- 97%57/59
- 100%59/59
- 100%59/59
Live: 59 of 59 repositories have the latest run on main green (0 red)
- 100%59/59
- 100%59/59
- 100%59/59
- 13%5/38
- 8%5/59
135/135 repositories resolved successfully · 100% healthy · metadata and versions are read from each repository manifest · dashboard generated by GitHub Actions · static GitHub Pages
Since your last visit
Health alerts
No alerts
Last published build: · workflow run
100%Registry health135 of 135 valid
How it is calculated: valid manifests divided by repositories. A repository counts as valid when its manifest was read and declares a known maturity and evidence. A repository whose manifest cannot be read, or has a manifest that was not listed, counts in the total but not as valid. It is the same set of problems as the Errors / unknown indicator.
Projects by ecosystem
ELECTRO-HOBBY-3D-UPDATER0.1.6TOOLS-GITHUB-ACTIONS-MONITOR2.1.7TOOLS-GITHUB-ACTIONS-MONITOR-ANDROID0.1.0TOOLS-GITHUB-ACTIONS-MONITOR-WATCH0.0.1
Progress and status by ecosystem
Same figures as the README.- 82%9/11
- 100%11/11
- 100%11/11
Live: 11 of 11 repositories have the latest run on main green (0 red)
- 100%11/11
- 100%11/11
- 100%11/11
- 0%0/9
- 9%1/11
- 90%18/20
- 100%20/20
- 100%20/20
Live: 20 of 20 repositories have the latest run on main green (0 red)
- 100%20/20
- 100%20/20
- 100%20/20
- 25%4/16
- 60%12/20
- 59%24/41
- 90%37/41
- 100%41/41
Live: 41 of 41 repositories have the latest run on main green (0 red)
- 100%41/41
- 100%41/41
- 80%33/41
- 0%0/16
- 0%0/41
- 50%2/4
- 75%3/4
- 100%4/4
Live: 4 of 4 repositories have the latest run on main green (0 red)
- 75%3/4
- 100%4/4
- 50%2/4
- 100%2/2
- 0%0/4
- Green
- Red
- Pending
- No CI / unknown
| Ecosystem | Green | Red | Pending | No CI / unknown |
|---|---|---|---|---|
| HYDRA-UMC | 59 | 0 | 0 | 0 |
| URTC | 11 | 0 | 0 | 0 |
| A.R.M.O.R. | 20 | 0 | 0 | 0 |
| EH3D | 41 | 0 | 0 | 0 |
| Tools | 4 | 0 | 0 | 0 |
No repository has a red CI.
| Month | Repositories |
|---|---|
| 2026-06 | 0 |
| 2026-07 | 0 |
| 2026-08 | 0 |
| 2026-09 | 0 |
| 2026-10 | 0 |
| 2026-11 | 133 |
- Scaffolding
- Functional
- Established
- Production
- Unknown
| Ecosystem | Scaffolding | Functional | Established | Production | Unknown | Total |
|---|---|---|---|---|---|---|
| HYDRA-UMC | 2 | 52 | 5 | 0 | 0 | 59 |
| URTC | 2 | 8 | 1 | 0 | 0 | 11 |
| A.R.M.O.R. | 2 | 6 | 12 | 0 | 0 | 20 |
| EH3D | 17 | 24 | 0 | 0 | 0 | 41 |
| Tools | 2 | 2 | 0 | 0 | 0 | 4 |
- None
- Host test
- Simulator
- Bench
- Field
- Unknown
| Ecosystem | None | Host test | Simulator | Bench | Field | Unknown | Total |
|---|---|---|---|---|---|---|---|
| HYDRA-UMC | 0 | 0 | 54 | 3 | 2 | 0 | 59 |
| URTC | 0 | 0 | 11 | 0 | 0 | 0 | 11 |
| A.R.M.O.R. | 0 | 0 | 16 | 0 | 4 | 0 | 20 |
| EH3D | 1 | 7 | 33 | 0 | 0 | 0 | 41 |
| Tools | 1 | 1 | 0 | 0 | 2 | 0 | 4 |
Latest releases
- HYDRA-UMC-DEV-SERVERv0.2.8
- HYDRA-UMC-BRIDGE-UAVv0.2.4
- EH3D-SUITE-LAUNCHERv0.0.7
- TOOLS-GITHUB-ACTIONS-MONITORv2.1.7
- EH3D-BUNGARD-GUIv0.3.0
- EH3D-PHOENIX-PCBv0.1.6
- EH3D-PHOENIX-3DPv0.0.8
Longest without a release
- HYDRA-UMC-EDITOR-STLv0.2.1
- URTC-DEVTOOLSv0.0.4
- JuanenRacv0.0.1
- HYDRA-UMC-PHYSICS-REPLICAv0.1.5
- EH3D-COMPONENT-INTELLIGENCEv0.0.4
- HYDRA-UMC-PATH-PLANNER-3Dv0.1.5
- HYDRA-UMC-COGNITIVE-NODEv0.2.2
3 repositories have never published a release
Progress over time
One snapshot of the progress and of the eight areas is saved per day (at most 90 days). Days without a snapshot are not filled in.
| Date |
|---|
HYDRA-UMC
The industrial multi-robot platform and cell controller.
HYDRA-UMC
overall progress59 repositories counted- 97%57/59
- 100%59/59
- 100%59/59
Live: 59 of 59 repositories have the latest run on main green (0 red)
- 100%59/59
- 100%59/59
- 100%59/59
- 13%5/38
- 8%5/59
- Scaffolding
- Functional
- Established
- Production
- Unknown
| Ecosystem | Scaffolding | Functional | Established | Production | Unknown | Total |
|---|---|---|---|---|---|---|
| HYDRA-UMC | 2 | 52 | 5 | 0 | 0 | 59 |
- None
- Host test
- Simulator
- Bench
- Field
- Unknown
| Ecosystem | None | Host test | Simulator | Bench | Field | Unknown | Total |
|---|---|---|---|---|---|---|---|
| HYDRA-UMC | 0 | 0 | 54 | 3 | 2 | 0 | 59 |
- Green
- Red
- Pending
- No CI / unknown
| Ecosystem | Green | Red | Pending | No CI / unknown |
|---|---|---|---|---|
| HYDRA-UMC | 59 | 0 | 0 | 0 |
No repository has a red CI.
Latest releases
- HYDRA-UMC-DEV-SERVERv0.2.8
- HYDRA-UMC-BRIDGE-UAVv0.2.4
- HYDRA-UMC-TWINv0.2.0
- HYDRA-UMCv0.3.0
- HYDRA-UMC-WATCHv0.3.5
- HYDRA-UMC-ANDROID-CONTROLv0.8.0.6
- HYDRA-UMC-OS-REBUILDERv0.3.9
- HYDRA-UMC-STUDIOv0.8.2.3
Longest without a release
- HYDRA-UMC-EDITOR-STLv0.2.1
- HYDRA-UMC-PHYSICS-REPLICAv0.1.5
- HYDRA-UMC-PATH-PLANNER-3Dv0.1.5
- HYDRA-UMC-COGNITIVE-NODEv0.2.2
- HYDRA-UMC-BRIDGE-OPENPNPv0.2.6
- HYDRA-UMC-SEMANTIC-PLANNERv0.2.1
- HYDRA-UMC-JOB-DISPATCHERv0.3.1
- HYDRA-UMC-VLA-ENGINEv0.2.4
URTC
An independent robot-tool subsystem with its own firmware and maintenance tools.
URTC
overall progress11 repositories counted- 82%9/11
- 100%11/11
- 100%11/11
Live: 11 of 11 repositories have the latest run on main green (0 red)
- 100%11/11
- 100%11/11
- 100%11/11
- 0%0/9
- 9%1/11
- Scaffolding
- Functional
- Established
- Production
- Unknown
| Ecosystem | Scaffolding | Functional | Established | Production | Unknown | Total |
|---|---|---|---|---|---|---|
| URTC | 2 | 8 | 1 | 0 | 0 | 11 |
- None
- Host test
- Simulator
- Bench
- Field
- Unknown
| Ecosystem | None | Host test | Simulator | Bench | Field | Unknown | Total |
|---|---|---|---|---|---|---|---|
| URTC | 0 | 0 | 11 | 0 | 0 | 0 | 11 |
- Green
- Red
- Pending
- No CI / unknown
| Ecosystem | Green | Red | Pending | No CI / unknown |
|---|---|---|---|---|
| URTC | 11 | 0 | 0 | 0 |
No repository has a red CI.
Latest releases
- URTC-UPDATERv0.3.2
- URTC-GATEWAYv0.1.0
- URTC-COMMONv0.3.1
- URTCv0.5.1
- URTC-SMART-RACKv0.2.9
- URTC-VISION-TOOLv0.2.2
- URTC-FLASHERv0.5.8
- URTC-TESTERv0.5.7
Longest without a release
- URTC-DEVTOOLSv0.0.4
- URTC-DOCSv0.1.8
- URTC-WEB-STUDIOv0.5.3
- URTC-TESTERv0.5.7
- URTC-FLASHERv0.5.8
- URTC-VISION-TOOLv0.2.2
- URTC-SMART-RACKv0.2.9
- URTCv0.5.1
A.R.M.O.R.
Perimeter security and home automation.
A.R.M.O.R.
overall progress20 repositories counted- 90%18/20
- 100%20/20
- 100%20/20
Live: 20 of 20 repositories have the latest run on main green (0 red)
- 100%20/20
- 100%20/20
- 100%20/20
- 25%4/16
- 60%12/20
- Scaffolding
- Functional
- Established
- Production
- Unknown
| Ecosystem | Scaffolding | Functional | Established | Production | Unknown | Total |
|---|---|---|---|---|---|---|
| A.R.M.O.R. | 2 | 6 | 12 | 0 | 0 | 20 |
- None
- Host test
- Simulator
- Bench
- Field
- Unknown
| Ecosystem | None | Host test | Simulator | Bench | Field | Unknown | Total |
|---|---|---|---|---|---|---|---|
| A.R.M.O.R. | 0 | 0 | 16 | 0 | 4 | 0 | 20 |
- Green
- Red
- Pending
- No CI / unknown
| Ecosystem | Green | Red | Pending | No CI / unknown |
|---|---|---|---|---|
| A.R.M.O.R. | 20 | 0 | 0 | 0 |
No repository has a red CI.
Latest releases
- ARMOR-DOCSv0.8.2.2
- ARMOR-COMMONv0.8.1.6
- ARMOR-ANDROID-CONTROLv0.8.0.9
- ARMOR-IOS-CONTROLv0.0.7
- ARMOR-NODE-HMIv0.5.0
- ARMOR-NODE-SOLARv0.6.8
- ARMOR-NODE-RADARv0.8.1.4
- ARMOR-NODE-ELECTRICALv0.6.0
Longest without a release
- ARMOR-VOICE-AIv0.3.7
- ARMOR-SERVER-AIv0.3.4
- ARMOR-NETWORKv0.1.9
- ARMOR-DEVOPSv0.7.8
- ARMOR-UPDATERv0.2.4
- ARMOR-VISIONv0.1.8
- ARMOR-SIMULATORv0.3.9
- ARMOR-HARDWAREv0.3.0
1 repositories have never published a release
EH3D
A PCB, CNC and laser fabrication suite.
EH3D
overall progress41 repositories counted- 59%24/41
- 90%37/41
- 100%41/41
Live: 41 of 41 repositories have the latest run on main green (0 red)
- 100%41/41
- 100%41/41
- 80%33/41
- 0%0/16
- 0%0/41
- Scaffolding
- Functional
- Established
- Production
- Unknown
| Ecosystem | Scaffolding | Functional | Established | Production | Unknown | Total |
|---|---|---|---|---|---|---|
| EH3D | 17 | 24 | 0 | 0 | 0 | 41 |
- None
- Host test
- Simulator
- Bench
- Field
- Unknown
| Ecosystem | None | Host test | Simulator | Bench | Field | Unknown | Total |
|---|---|---|---|---|---|---|---|
| EH3D | 1 | 7 | 33 | 0 | 0 | 0 | 41 |
- Green
- Red
- Pending
- No CI / unknown
| Ecosystem | Green | Red | Pending | No CI / unknown |
|---|---|---|---|---|
| EH3D | 41 | 0 | 0 | 0 |
No repository has a red CI.
Latest releases
- EH3D-SUITE-LAUNCHERv0.0.7
- EH3D-BUNGARD-GUIv0.3.0
- EH3D-PHOENIX-PCBv0.1.6
- EH3D-PHOENIX-3DPv0.0.8
- EH3D-PHOENIX-PNPv0.1.4
- EH3D-PHOENIX-LASERv0.1.6
- EH3D-PHOENIX-CNCv0.2.3
- EH3D-PHOENIX-CADv0.1.3
Longest without a release
- EH3D-COMPONENT-INTELLIGENCEv0.0.4
- EH3D-VALIDATION-LABv0.0.3
- EH3D-THERMAL-LABv0.0.3
- EH3D-PROJECT-PASSPORTv0.0.4
- EH3D-DESIGN-VERIFIERv0.0.4
- EH3D-NODE-BUNGARDv0.0.8
- EH3D-NODE-BUNGARD-AUXv0.1.0
- EH3D-UPDATERv0.1.1
Tools
The programs that sit above the four ecosystems: the Actions monitor and the updater of the suite.
Tools
overall progress4 repositories counted- 50%2/4
- 75%3/4
- 100%4/4
Live: 4 of 4 repositories have the latest run on main green (0 red)
- 75%3/4
- 100%4/4
- 50%2/4
- 100%2/2
- 0%0/4
- Scaffolding
- Functional
- Established
- Production
- Unknown
| Ecosystem | Scaffolding | Functional | Established | Production | Unknown | Total |
|---|---|---|---|---|---|---|
| Tools | 2 | 2 | 0 | 0 | 0 | 4 |
- None
- Host test
- Simulator
- Bench
- Field
- Unknown
| Ecosystem | None | Host test | Simulator | Bench | Field | Unknown | Total |
|---|---|---|---|---|---|---|---|
| Tools | 1 | 1 | 0 | 0 | 2 | 0 | 4 |
- Green
- Red
- Pending
- No CI / unknown
| Ecosystem | Green | Red | Pending | No CI / unknown |
|---|---|---|---|---|
| Tools | 4 | 0 | 0 | 0 |
No repository has a red CI.
Latest releases
Longest without a release
2 repositories have never published a release
Tools: the programs above the four ecosystems
One row per tool: version, latest release, state of its CI and tests. Read when the page was generated; what could not be read is never shown as fine.
| Tool | Version | Maturity | Latest release | CI | Tests | Links |
|---|---|---|---|---|---|---|
| ELECTRO-HOBBY-3D-UPDATER | 0.1.6 | functional | v0.1.6 2026-11-21 | Green | 73 | Repository · Releases · Changelog |
| TOOLS-GITHUB-ACTIONS-MONITOR | 2.1.7 | functional | v2.1.7 2026-11-21 | Green | 5936 | Repository · Releases · Changelog |
| TOOLS-GITHUB-ACTIONS-MONITOR-ANDROID | 0.1.0 | scaffolding | – | Green | 80 | Repository · Releases · Changelog |
| TOOLS-GITHUB-ACTIONS-MONITOR-WATCH | 0.0.1 | scaffolding | – | Green | not declared | Repository · Releases · Changelog |
TOOLS-GITHUB-ACTIONS-MONITOR 2.1.7
Watches the workflows of every repository you own and shows what is red, stuck or waiting on one local page, in eight languages, with metrics for Grafana. A workflow is red when the last completed run of its file on the default branch failed; this dashboard uses the same rule (see the Sentinel tab).
Two editions
Complete: the monitor and its five sentinels that never write to GitHub (CI with its checks A1 to A13, Collab, Security, Supply chain and Delivery), a score and the health of each repository and team, reports, wall screens, accounts with roles and an operation panel with backup and an updater.
Lite: the monitor alone, with no sentinel; the connection with an AI tool is the same in both editions.
Connect an AI tool
Five channels, the same in both editions: a brief of what to do now, a feed of alerts, the state as data, a stream of events and MCP tools, plus a gate script that adds the brief to the session of the AI. Everything is read only unless you allow an action.
Quota and refresh
The monitor spends little: conditional reads (an unchanged answer, a 304, costs nothing), batched GraphQL queries and saved reads. The central panel of this page is refreshed at least every half hour, and the Sentinel tab warns when it is more than two hours behind.
In the next version
Not published yet: a comparison of a period against the one before (share of failed runs, duration, time to resolve alerts, score), an operator command line with table, plain, JSON, NDJSON and CSV formats, a watch command, shell completions and a kiosk profile with larger type and high contrast for wall screens. This dashboard takes the same ideas of the sentinels in its own way: the Supply chain and Delivery cards of the Sentinel tab.
ELECTRO-HOBBY-3D-UPDATER
One window and one command line that hand over to the updater of each ecosystem (HYDRA-UMC-UPDATER, URTC-UPDATER, ARMOR-UPDATER and EH3D-UPDATER).
| Project | Type | Maturity | Stack | Deploy target | Version | Tests | Pipeline | Status | Last commit | Activity |
|---|---|---|---|---|---|---|---|---|---|---|
| HYDRA-UMC - the industrial multi-robot platform and cell controller, this dashboard's first ecosystem. | ||||||||||
| Cognitive AI Node 5 project(s) · HYDRA-UMC-COGNITIVE-NODE is this family's own integration parent | ||||||||||
| Service | Functional | python | CM5 | 0.2.2 |
OK
Version successfully resolved
|
Bump the github_actions_updates group with 2 updates (#14) | ||||
|
Notes
Real v0 family-readiness check (`family-status [--json]` subcommand): reads each of the 4 real children's own hydra-umc.project.json and reports presence/version/maturity through a real, versioned JSON schema. Sibling manifest reads are bounded (64 KiB) against a corrupted/oversized file. Also honestly reports this node's own shared model-weights directory (models/) as present/missing - real degradation-awareness for an integration hub that runs no model itself. Real LLM/VLA/voice orchestration stays future work, blocked on the children themselves.
Technology
PythonCM5Hailo-10
Build
python tools/build_test.py
|
||||||||||
| API | Functional | python | CM5 | 0.2.1 |
OK
Version successfully resolved
|
Bump the github_actions_updates group with 3 updates (#18) | ||||
|
Notes
Real v0 lexical retrieval: stdlib-only TF-IDF search (query subcommand) over ingested Markdown, defaulting to this repo's own README/CHANGELOG. Deterministic scoring (hash-seed-independent summation), traceable source#index citations, and a real document allowlist (only existing .md/.markdown paths are ingested; every other path is rejected with a distinct reason instead of being silently skipped or silently read as text). Honestly lexical, not yet embedding-based semantic search, and no LLM synthesis step - that and PDF ingestion are future work. Evidence is simulator: the automated test suite runs in CI on the host, and a simulation bench runs the real service against failures around it (reloads, tokens and, on a schedule drawn from a fixed seed, documents and corpus bundles that go wrong on disk); nothing here has run on real hardware.
Technology
PythonTF-IDF lexical retrieval
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.1 |
OK
Version successfully resolved
|
Move the container image and the tests to Python 3.15 (#21) | ||||
|
Notes
Real v0 rule-based task decomposition (`decompose` subcommand, small known goal vocabulary) and real rule-based semantic recovery over structured MCU error codes (`recover` subcommand). Real precondition validation (validation.py) refuses a plan that fails its own preconditions instead of handing it off as execution-ready; decompose_goal() is proven deterministic and fuzz-tested against 500 random invalid goals. Honestly template-based, not the local LLM the README describes - that and context-aware multi-step reasoning are future work. v0.0.8: validate_plan()/validate_step()/POST /decompose gained an optional real declared-capability catalog (capabilities) - a well-formed step for a primitive the target robot never declared support for is now a real, distinct issue (I38); real postcondition verification stays deliberately out of scope until a real executor exists to verify against.
Technology
Pythonlocal LLM (Llama-3 class)
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.4 |
OK
Version successfully resolved
|
Bump the github_actions_updates group with 2 updates (#19) | ||||
|
Notes
Real action-token encoding/decoding (OpenVLA/RT-2-style 256-bin discretization per action dimension) and trajectory generation (delta-pose integration, absolute gripper state) - the 'Action Tokens' and 'Trajectory Generator' steps of the VLA flow, exposed via 'tokens encode'/'tokens decode'/'trajectory integrate'. A real, versioned model manifest contract (model_manifest.py) plus shape/confidence validation for a future model's inference output, and an honest 'status' subcommand reporting real accelerator/model-weight availability (never a fake ready state). Added hailo_runtime.py: a real HailoRT (hailo_platform) integration boundary - open_vdevice()/load_hailo_vla_model() against the real confirmed API (VDevice, HEF, ConfigureParams, InputVStreamParams/OutputVStreamParams), lazily imported so this dev machine (no hailort installed) degrades to a clear RuntimeError; hailo_output_to_tokens() adapts a real InferVStreams result into this engine's own token/confidence contract and is fully unit-tested without hailort. integrate_trajectory() now rejects an accumulated pose that silently overflowed to infinite (two individually-finite values, e.g. a start position and a delta, summing past float's max range) instead of returning it as part of the trajectory. Real VLA model inference still needs a real .hef model and a physical Hailo-10 module, both future work. Evidence is host-test: the automated test suite runs in CI on the host; nothing here has run against a simulator bench or real hardware.
Technology
PythonVision-Language-Action model
Build
Vision-Language-Action engine. Expect large model downloads on first installation for the real model integration (still future work).
|
||||||||||
| Service | Functional | python | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Release 0.2.6 (#21) | ||||
|
Notes
Real v0 audio front-end (WAV loading + energy-gate voice activity detection, stdlib-only), rule-based intent/entity parser over already-transcribed text, and a bounded Watch voice-turn HTTP gateway. Server now relays authenticated client requests to a loopback CM5 service while retaining the gateway token; replies require primary-control confirmation for motion. Real intent normalization (Unicode NFKC, filler/punctuation stripping) and ambiguity-aware classification (classify_intent) reject a transcript that genuinely matches more than one known command instead of silently guessing. Whisper STT / neural TTS still require their real model/runtime dependencies.
Technology
PythonRule-based intent parserEnergy-gate VAD
Build
python tools/build_test.py
|
||||||||||
| Complementary Tools 5 project(s) | ||||||||||
| UI | Functional | node | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Bump the npm_updates group across 1 directory with 5 updates (#28) | ||||
|
Notes
Functional v0: Smart Summaries (min/max/average/latest/direction stats from real HYDRA-UMC-DATALAKE history, each of min/max/latest linked back to the real observation timestamp that produced it) and Anomaly Highlighting (real check against a real, already-fitted HYDRA-UMC-ANOMALY-DETECTOR baseline), each a real text panel, verified against live sibling services. Real AI-provider gate (aiProvider.ts): input/output schema validation, real rejection of unstructured provider output, and an honest statistical fallback for when no real AI provider is configured - wired into the Trend Summary panel. An actual LLM-backed provider, Trend Prediction (real forecasting), and embedding the panels inside STUDIO itself are future work. Evidence is simulator: the automated test suite runs in CI on the host, and a simulation bench runs the forecast, the chart, the verifier of the numbers, the clients of the services that go wrong and come back and the gate of the language model against seeded histories and a scripted local server; nothing here has run against the real services or real hardware.
Technology
TypeScript/ReactVite
Build
python tools/build_test.py
|
||||||||||
| Library | Functional | data | User PC | 0.1.1 |
OK
Version successfully resolved
|
build(deps): bump the github_actions_updates group with 2 updates (#13) | ||||
|
Notes
The shared 3D meshes (STL and GLB) of the HYDRA-UMC desktop and web apps, published once as a release package instead of being copied into each repository. The release carries one archive with every distinct mesh stored once and a manifest with the SHA-256 of the archive and of each file; HYDRA-UMC-STUDIO and HYDRA-UMC-SUITE download it on demand and check every file against the manifest before writing anything. The repository holds only the manifest and its checksums; the check in CI confirms that the manifest and the checksums agree. The same check can verify a downloaded archive and the folder it was unpacked into, and a simulated download that fails in every way (cut short, changed, padded, empty, absent, files missing or altered) is tested; the download by the apps from the real release has not been run.
Technology
GLBJSONSHA-256STL
Build
python tools/verify_pack.py
|
||||||||||
| Library | Functional | node | User PC | 0.0.5 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#26) | ||||
|
Notes
Small TypeScript package with no runtime dependencies and no framework: the sign-in session storage, the connection history, guided tours, local preferences and an HTTP client with timeouts, bounded retries and typed errors that the web clients of the family can share. Tested with fakes for fetch and for the storage, and by a simulation bench: the real HTTP client, session and connection history against a simulated server that signs people in, lets a token expire and fails on command (503, rate limit, cut answer, no answer, dropped connection, restart). Nothing in it has run inside a real browser or against the real server yet. HYDRA-UMC-STUDIO does not consume it yet. Evidence is host-test: the automated test suite runs in CI on the host; nothing here has run against a simulator bench or real hardware. Maturity is functional: the package has automated tests and a published release.
Technology
TypeScriptVitest
Build
python tools/build_test.py
|
||||||||||
| CLI | Functional | go | User PC | 0.2.2 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#20) | ||||
|
Notes
Real `version`/`status`/`robots`/`doctor`/`config validate`/`config apply --dry-run`/`shell` commands, plus `claims`/`claim`/`release`/`limits`/`jog-arm`/`jog` (reservations with a per-session controller id, safety limits read from the server, jog arming; a jog above the model's limit is refused before it is sent). `robots`/`status`/`doctor` are genuine clients of a live HYDRA-UMC-SERVER (GET /api/settings, GET /api/hydra-info). A --config file's own "server"/"timeoutSec" (already schema-validated by `config validate`/`apply`) now really drives every live command's HTTP target and request timeout via resolveTarget(), not just validation - a `shell` session persists both as session defaults too. A real, stable exit-code contract (exitcode.go) classifies every failure (usage/config/network/server/not-implemented/auth) instead of a bare exit 1. The tests use net/http/httptest and real temp-file config fixtures, including `doctor --json`'s structured checkId/severity/status report. Fleet-wide deploy/flash-all/audit commands and a live (non-dry-run) `config apply` remain future work, blocked on HYDRA-UMC-SERVER's own fleet write endpoints that don't exist yet.
Technology
Gostandard-library flag handling
Build
Go module root is src/. Build with go build ./cmd/hydra-cli from inside src/.
|
||||||||||
| UI | Functional | android | Wearable | 0.3.5 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#38) | ||||
|
Notes
Functional v0: differentiated haptic vibration patterns wired to Wear OS Vibrator, explicit microphone/system-STT capture and local TTS confirmation, plus typed voice_turn/assistant_reply/system_status messages. A paired Android Data Layer relay is implemented; it keeps the Server JWT on the phone and Voice UI token on CM5. Physical radio, wireless E-STOP and live CM5 telemetry remain hardware integration gates. Voice never directly actuates a robot. The relay now has a real, testable reconnection policy (RelayRetryPolicy) and a last-known-state cache with staleness tracking (LastKnownStateCache) so a stale relayed status/alert is never shown as current. The relay now only ever sends to a real CapabilityClient-verified companion node (hydra_umc_phone_companion, not any connectedNodes.firstOrNull()), preferring a nearby node over a cloud-relayed one, and cancels a pending retry outright once the screen closes instead of only bounding it by attempt count. Evidence is simulator: the Wear OS workflows install the debug APK on an emulator and run the instrumented tests there (started by hand); no run on a real watch is recorded.
Technology
KotlinWear OS
Build
./build.sh or build.bat (bumps version, runs JUnit tests, assembleDebug). Wear OS 3+ target.
|
||||||||||
| Core Backend & Clients 8 project(s) · HYDRA-UMC-SERVER is this family's own integration parent | ||||||||||
| API | Functional | node | CM5 | 0.8.3.3 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#81) | ||||
|
Notes
The real headless backend every control client (STUDIO, SUITE, DSI, the mobile apps) actually talks to. It relays bounded authenticated voice turns to a loopback HYDRA-UMC-VOICE-UI gateway while retaining the gateway token server-side (voice never becomes a direct robot command), and now relays the real CM5<->STM32H757 SPI-OTA link to a local spi_bridge service (admin-only for flashing, live progress over WebSocket) - not yet exercised against real STM32H757 hardware, since no populated board exists yet. A new /api/ai/:service/* real passthrough proxy (GET/POST/etc) reaches this cell's own local Hailo-accelerated AI services (cognitive-node/detection-hef/visual-servoing-api/vla-engine, all loopback-only systemd units) the same authenticated way, for a development/control machine that is not on this host itself. New GET /api/adapters and /api/adapters/:adapterId proxy this cell's own real HYDRA-UMC-CONNECTOR-HUB serve-catalog, making this the ecosystem's first real Server-side consumer of the adapter catalog rather than only a CLI/fixture calling its safety gate. src/serverPolicy.ts now holds 8 pure helper functions extracted out of src/server.ts (some previously trapped inside startServer()'s own closure with no external reason to be), each with real, direct node:test unit tests (tests/serverPolicy.test.ts, tests/kinematics.test.ts) run via the already-present tsx devDependency - this repository's first direct unit tests, alongside its existing 11 real end-to-end tools/verify_*_contract.mjs scripts. src/users.ts accounts now carry a real tokenVersion, bumped on every account mutation and re-checked on every authenticated HTTP request and WebSocket tick (not just at login), so deleting/demoting a user or changing its password revokes every already-issued token/session immediately instead of waiting for JWT expiry; users.json reads now distinguish a genuinely missing file from real corruption (throws instead of silently reseeding) and writes are atomic (temp file + rename); scrypt password hashing runs on Node's async threadpool instead of blocking the whole event loop per login, bounded to 4 concurrent operations plus a 20-deep queue with an explicit ScryptOverloadError/503 once both are full (an unbounded queue is just a slower-motion version of the same overload). v0.6.9: POST /api/robot/:id/command's speed case now returns a real warnings array for a rejected out-of-range setpoint instead of silently discarding it (I01); robot claim/release now append to a real, bounded reservationHistory recording who/when/why for every transition (claimed/renewed/force-claimed/released/force-released/expired), returned directly by both endpoints (I03).
Technology
Node.js 20+ExpressWebSocket (ws)REST/OpenAPI
Build
npm install && npm run build.
|
||||||||||
| UI | Functional | android | Mobile | 0.8.0.6 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#71) | ||||
|
Notes
Native Android control app with biometric login, talking to HYDRA-UMC-SERVER over Wi-Fi/Bluetooth. It receives paired Wear OS Data Layer voice/status messages, keeps the Server JWT encrypted on the phone and returns bounded replies; that path never issues robot commands. Declares a real Wear OS android_wear_capabilities resource (hydra_umc_phone_companion) so the paired Watch can discover this exact app via CapabilityClient instead of trusting connectedNodes ordering. A red STOP button in the top bar of every screen, on the control console and on the 3D view, the home-screen widget and the notification action all send one POST /api/stop-all (no hold, no confirmation) that stops every robot, with a retry, a per-robot fallback on 429 and a persistent STOP NOT CONFIRMED state. Evidence is simulator: the emulator workflows install the debug APK on Android emulators and run the instrumented tests there (started by hand); no run on a real phone is recorded.
Technology
KotlinJetpackbiometric authWebSocket client
Build
./gradlew assembleDebug. Needs Android SDK.
|
||||||||||
| UI | Functional | flutter | CM5 | 0.3.9 |
OK
Version successfully resolved
|
Pin the SDK to its 0.7.3 release (#47) | ||||
|
Notes
Native touch UI for the onboard 7" DSI touchscreen, same client role as the mobile apps but embedded on the CM5 itself. v0.1.5: a failed secure-storage delete during logout, or a partial saveToken() write (token saved, username save failed), could leave a real, orphaned token that a later loadToken() would resurrect - fixed with a plain, non-secret hydra_logged_out marker checked before secure storage is ever consulted, plus a best-effort revert of a partial write. Same fix ported to HYDRA-UMC-IOS-CONTROL. Evidence is simulator: the automated test suite runs in CI on the host against a stand-in server (the shared fake server of the SDK, a local socket server and a scripted network that goes wrong and comes back in an order drawn from a fixed seed), and the kiosk unit runs under a headless compositor in a container; nothing here has run on a Compute Module 5 with its panel.
Technology
Flutter 3.xDart
Build
flutter pub get && flutter build linux
|
||||||||||
| Tool | Functional | python | User PC | 0.2.1 |
OK
Version successfully resolved
|
Bump the github_actions_updates group with 2 updates (#13) | ||||
|
Notes
Browses and edits the two real model libraries HYDRA-UMC-STUDIO's own public/models/ and HYDRA-UMC-SUITE's own assets/meshes/ already ship (both reorganized into the same category folders in 2026-09: robots-5-dof/robots-6-dof/robots-7-dof, machine-pnp/machine-cnc/machine-laser, heatedbeds/racks/vacuum-tables). Real operations only, no simulated success: transform (translate/rotate/scale real vertex data via numpy-stl), replace, remove and add a part's own STL file, each backed by real file I/O against the actual checkout on disk - nothing is ever permanently deleted, a removed/replaced part is moved into that model's own .trash/ subfolder instead. `--cli` is stdlib+numpy-stl only (no Qt import at all) so a headless machine can still browse/edit the libraries; the default launch is the Qt Quick desktop shell, sharing HYDRA-UMC-UPDATER's own visual component set (GameButton/GameCombo/SectionPanel, dark cyan/blue/amber/red theme) rather than a separately styled UI. It pushes a model to HYDRA-UMC-SERVER's POST /api/models/submit catalog (independent pieces only are posed correctly; robot and machine parts arrive without joints) and shows a per-mesh 3D preview.
Technology
PythonPySide6Qt Quicknumpy-stl
Build
Desktop Python app, dependency-light core (numpy-stl only) plus an optional PySide6/Qt Quick GUI extra. run.bat/run.sh runs it from source; build.bat/build.sh creates a venv, installs the dev+gui extras and runs the real test suite; build-test.bat/build-test.sh validates sources without bumping the version.
|
||||||||||
| Tool | Established | python | User PC | 0.2.1 |
OK
Version successfully resolved
|
Say Python 3.12 in the README, the version the workflows run on (#19) | ||||
|
Notes
Desktop graphical URDF creator/editor with real, shipped features (GitHub/local source loading, mesh reference resolution, DOF feasibility validation, live 3D preview editing) - pushes finished models to HYDRA-UMC-STUDIO's own model catalog via HYDRA-UMC-SERVER's POST /api/models/submit. tests/ now covers 214 real cases across the app's own pure logic (URDF parse/write, DOF validation, kinematics, inertia estimation, mesh resolution, GitHub fetching, i18n) via pytest, none of which need PySide6 - wired into CI, closing a real zero-test-coverage gap.
Technology
PythonPySide6Qt6OpenGL
Build
Desktop PySide6/Qt6 app. run.bat/run.sh runs it from source; build-test.bat/build-test.sh validates sources without versioning; build_exe.bat/build_exe.sh produces a standalone PyInstaller executable.
|
||||||||||
| UI | Functional | flutter | Mobile | 0.3.7 |
OK
Version successfully resolved
|
Keep flutter_blue_plus on the 1.x line and ignore its major updates in Dependabot (#48) | ||||
|
Notes
iOS/iPadOS control app with real-time WebSocket sync to HYDRA-UMC-SERVER: sign-in, dashboard, manual control with a global STOP, live camera, 3D view, telemetry and an offline copy of the last state that is clearly marked as saved data. Evidence is simulator: the xcodebuild workflow runs the Swift tests on an iPhone simulator (started by hand); no run on a real iPhone is recorded.
Technology
Flutter 3.xDart
Build
flutter pub get && flutter test (checks, any platform); flutter build ios (real iOS build, requires Xcode on macOS)
|
||||||||||
| UI | Functional | node | CM5 | 0.8.2.3 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#83) | ||||
|
Notes
Web control dashboard with 3D multi-robot visualization, talking to HYDRA-UMC-SERVER's own API. Commands report their result, the E-STOP acts at once, and settings saves show when they fail.
Technology
React 19ViteThree.jsTailwind CSS
Build
npm install && npm run build
|
||||||||||
| UI | Functional | python | User PC | 0.8.0.8 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#68) | ||||
|
Notes
Desktop swarm command center, packaged as a standalone executable. Real server discovery (net/discovery.py): concurrent subnet /24 scan plus real mDNS (_hydra._tcp, matching what server.ts publishes and HYDRA-UMC-IOS-CONTROL already queries), deduplicated by (host, port). Every real-time robot control (joint jog, speed, acceleration) now fires the atomic, debounced /api/robot/:id/command path (net/client.py's HydraConnection.send_command(), debounce_ms) instead of the full-tree push_state() a continuously-dragged knob/slider used to call once per mouse event. tests/verify_discovery.py and tests/verify_send_command_debounce.py give the pure logic real, deterministic, no-network coverage (this repo has no pytest suite - verify_*.py scripts are its own real testing convention). A global STOP (net/client.py HydraConnection.stop_all(), POST /api/stop-all) sits in the command deck and the Qt Quick header with the application-wide F12 shortcut, covered by tests/verify_stop_all.py. tools/run_offline_verifiers.py now runs all of them headlessly (QT_QPA_PLATFORM=offscreen), wired into both tools/build_test.py and this repository's own CI workflow - a broken control no longer reaches main with a green CI. Evidence is simulator: the automated test suite runs in CI on the host and a simulation bench runs the real network client, sign-in and panel logic against fake HTTP and WebSocket servers that lie, refuse, cut an answer, go off and come back; nothing here has run against a real controller, a real robot or the real server.
Technology
Python 3.12PySide6
Build
No pyproject.toml. __init__.py's __version__ is the real source. Packaged as standalone .exe using build_exe.bat/.sh.
|
||||||||||
| Core Hardware 1 project(s) · HYDRA-UMC is this family's own integration parent | ||||||||||
| Firmware | Scaffolding | firmware-c | STM32 MCU | 0.3.0 |
OK
Version successfully resolved
|
Name the repository in the draft job of the firmware build (#123) | ||||
|
Notes
The motion-control firmware for the robot arm boards of this ecosystem: 6 independently-versioned MCU targets (G474 + H7 CM7/CM4, app + bootloader each), plus a CM5-side SPI-OTA bridge (src/cm5_host/spi_bridge/, Python, unit-tested against fakes). The H7 CM4 application has the PLL1 clock configuration, the FDCAN1 master and the relay tunnel to Tier 2 (KinematicBrainCan.c); the G474 application has the AXIS_STATUS responder and the FDCAN2 relay tunnel to the URTC head (RobotControllerRelay.c). Not written yet: the 6-axis motion generation on both G474 and CM7 (the CM7 application is still a FreeRTOS smoke test). Nothing here has run on a populated board yet; the status table of the README says what has been checked and how. Evidence is simulator: the Renode workflows boot the real firmware on an emulated STM32 board and the virtual CAN check drives the real CAN code over a virtual bus; no run on a real board is recorded.
Technology
STM32H757 (dual-core)STM32G474FreeRTOS (AMP)CMSIS-DSPFDCAN (classic CAN mode)SWD/JTAGPython (CM5 host: spi_bridge)
Build
6 independently-versioned components (G474 + H7 CM7 + H7 CM4, app + bootloader each). This registry tracks the G474 application as representative. Real install/build: python3 tools/build_firmware.py.
|
||||||||||
| Data & Analytics 4 project(s) · HYDRA-UMC-DATALAKE is this family's own integration parent | ||||||||||
| API | Functional | python | CM5 | 0.2.7 |
OK
Version successfully resolved
|
Warn when reading is open to the network and add --require-read-token (#17) | ||||
|
Notes
Real sqlite3-backed time-series store with a real HTTP API (POST /ingest, GET /query), tested over real HTTP. Real, reversible schema migrations (PRAGMA user_version), explicit UTC timestamp formatting (GET /stats/range), and validated, opt-in per-series retention (GET/POST /retention, POST /retention/apply).
Technology
Pythonsqlite3http.server (ThreadingHTTPServer)pytest
Build
Runs real SQLite-backed store and HTTP API checks; no external database is required by the current implementation.
|
||||||||||
| API | Functional | python | CM5 | 0.2.5 |
OK
Version successfully resolved
|
Release 0.2.5 (#18) | ||||
|
Notes
FFT + statistical baseline detector (no machine-learning model), empirically threshold-tuned against real synthetic healthy/faulty signal separation. Real model versioning (every Verdict carries the model_version/threshold it was scored against), real precision/recall metrics over the labeled synthetic fixture (metrics.py), and simulated drift detection (drift.py's DriftMonitor) for a sustained score-elevation trend distinct from any single window's own anomaly flag.
Technology
PythonFFTstatistical baseline detection
Build
python tools/build_test.py
|
||||||||||
| API | Functional | python | CM5 | 0.2.2 |
OK
Version successfully resolved
|
Bump actions/setup-python (#21) | ||||
|
Notes
Real OEE/availability calculation over real HYDRA-UMC-DATALAKE history via a real HTTP client, verified end-to-end with hand-checkable numbers. Real shift/day-boundary computation (shift.py), a real versioned formula plus a sha256 input fingerprint on every report (formula_version/input_fingerprint), and a real byte-for-byte reproducible CSV export (export.py, GET /reports/{oee,availability}/export).
Technology
Pythonreal OEE formulahttp.server
Build
python tools/build_test.py
|
||||||||||
| API | Functional | go | CM5 | 0.2.3 |
OK
Version successfully resolved
|
Bump actions/setup-go (#27) | ||||
|
Notes
Real CAN/WebSocket ingestion pipeline with a bounded buffer, retry-on-outage delivery, and a real DatalakeSink writing to HYDRA-UMC-DATALAKE - verified end-to-end. Real per-producer sequence deduplication with a bounded reorder window (dedup.Tracker) so a reconnect resend never inflates ingest counts, and a real transport-vs-invalid-data error classification on flush failures (sink.InvalidDataError) for real operational diagnosis. A permanently-rejected sample is now quarantined (dropped, counted in Stats().Quarantined) instead of being requeued forever and blocking every valid sample behind it - sink.InvalidDataError now carries that sample's exact batch Index so only it, not the whole batch, is dropped. Evidence is simulator: the automated test suite runs in CI on the host and a simulation bench runs the collector, its disk spool and the real sink against a data lake that flaps, rejects, resets and loses answers, on a fixed seed; nothing here has run against a real CAN bus, a real robot or a real HYDRA-UMC-DATALAKE.
Technology
GoCAN/WebSocket parsersbounded FIFO bufferHTTP API
Build
Go module root is src/. Run go build ./... from inside src/.
|
||||||||||
| Digital Twin & Simulation 4 project(s) · HYDRA-UMC-TWIN is this family's own integration parent | ||||||||||
| Service | Functional | rust | CM5 | 0.2.0 |
OK
Version successfully resolved
|
Name the repository in the draft job of the binaries build (#40) | ||||
|
Notes
Functional v0: real family-readiness check (`family-status` subcommand) reads each of the 3 real children's own hydra-umc.project.json and reports presence/version/maturity/role. Plus a real state-sync contract (src/contract.rs): a child must be at least `functional` maturity and report a major version no higher than this Twin's verified-compatible ceiling (0 today) before it is treated as sync-ready - an incompatible simulator version or an immature child is refused with a real reason, exposed via the new `family-sync` subcommand. Honest for an integration hub that runs no rendering/physics engine itself - the real Bevy-based engine and MuJoCo/PhysX backend stay future work, blocked on a real GPU/graphics toolchain and the children's own engines maturing. Besides the two subcommands the service signs risk certificates (`POST /interlock/certify`), predicts collisions for a short trajectory (`POST /predict/collision`) and dry-runs a trajectory (`POST /dryrun/trajectory`).
Technology
Rusttiny_http
Build
cargo build --release.
|
||||||||||
| Service | Functional | rust | CM5 | 0.1.8 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#25) | ||||
|
Notes
Functional v0 mode-based command routing (`route`/`mirror` subcommands): a real safety interlock blocks commands to real hardware whenever the twin reports an imminent collision, simulation-mode commands are never gated, and mirroring shadows a command into the twin unconditionally. CommandSink::send() now returns Result<(), TransportError> (Timeout/Disconnected); a new SimulatedTransport implementation models a slow or disconnected link with configurable latency/timeout, so the bridge's fail-safe behavior (a transport failure is reported as RouteOutcome::TransportFailure, never mistaken for SentReal/SentSimulation) is testable and exercisable via CLI flags without any real hardware. Real gRPC/WebSocket transport stays future work - genuinely blocked on no live HYDRA-UMC-TWIN/HYDRA-UMC controller to talk to yet, not just unwritten code.
Technology
Rusthardware-in-the-loop I/O
Build
cargo build --release
|
||||||||||
| Library | Functional | rust | CM5 | 0.1.5 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#23) | ||||
|
Notes
Functional v0: real forward kinematics and joint-limit validation (`fk`/`validate-limits` subcommands) over a real, documented-partial URDF subset (revolute/continuous/prismatic/fixed joints). src/urdf.rs now walks each joint's real `parent`/`child` link names from the document's own real root link into a single serial chain, in genuine root-to-leaf order regardless of `<joint>` element order in the file - a real branching tree (one link with more than one child joint) is refused, not flattened. Plus a real joint-limits corpus (src/corpus.rs) and a fail-safe `fk-checked` subcommand: forward_kinematics_checked() refuses to compute a world-frame pose for a joint position outside its declared URDF limit, closing the gap where plain fk would silently compute a physically-unreachable pose. Inverse dynamics by the recursive Newton-Euler algorithm over the `<inertial>` blocks, self-collision with capsules built from `<collision>`, a trajectory check and an HTTP service with `/fk`, `/ik`, `/check-trajectory` are in. Contact simulation with MuJoCo or PhysX stays future work: MuJoCo is only an optional bridge behind a Cargo feature that is off by default. Reader hardening: only joints directly under <robot> are read (ros_control transmissions parse), a link with two parents, duplicate joint names, unknown links and a zero axis are errors, a continuous or fixed joint has no position limit, and a misspelt or missing joint in --joints is an error instead of a pass. fk --json prints the position and the rotation as a unit quaternion.
Technology
RustroxmltreeURDF forward kinematicsInverse dynamics (RNEA)Capsule self-collisionMuJoCo MJCF
Build
cargo build --release
|
||||||||||
| Tool | Established | python | User PC | 0.1.6 |
OK
Version successfully resolved
|
Bump the github_actions_updates group with 2 updates (#10) | ||||
|
Notes
Real v0 procedural 2D scene generation, BMP and PNG rasterization with the standard library alone, and real YOLO/COCO annotation export (`generate` subcommand), verified end to end incl. visual inspection of a real rendered image. A real dataset manifest (manifest.json, with a reproducible-seed flag and a per-image sha256 checksum) and real post-generation validation (scene bounds, BMP integrity against a genuinely truncated file, label-distribution sanity) now run automatically at the end of every `generate` call. Honestly 2D placeholder shapes, not the photorealistic 3D rendering through HYDRA-UMC-TWIN's own engine the README describes - that engine doesn't exist yet (TWIN is itself still scaffolding). TFRecord export (`--tfrecord`) is in; real 3D rendering through TWIN is future work.
Technology
Pythonprocedural scene generationYOLO/COCO annotation export
Build
python tools/build_test.py
|
||||||||||
| Ecosystem Operations 7 project(s) | ||||||||||
| Service | Functional | python | Dev Server | 0.2.8 |
OK
Version successfully resolved
|
Release 0.2.8 | ||||
|
Notes
Functional - DS01-DS10 of the 10-delivery plan, now complete (contracts, limits and a verifiable skeleton). DS01-DS03 only read and describe. DS01: state-ownership table + 17-relationship grouping + a real, tested configuration schema (config.py) whose default policy grants NO task deployment permission. DS02: validated remote-station profile + read-only host preflight + dry-run provisioning plan, never executed. DS03: conservative migration (migration.py) - hashes/classifies every file in a source checkout, records unpushed commits, plans each class into one of four provably-separate destinations, refuses a private file to a shareable root; the source is never touched. DS04: the first thing that executes - workspace.py refuses a path that escapes a per-task workspace (.., absolute, symlink-out) and refuses two tasks sharing one; recipe.py pins a revision and refuses any command not in the policy's own allowed_commands; runner.py runs that one allow-listed command in the isolated workspace with a SCRUBBED environment (no inherited *_TOKEN/*_KEY/*_SECRET/ANTHROPIC_*/GITHUB_*/SSH_*), under a bounded timeout, killing the whole process group on timeout or cancel - proven by a real test that spawns a child-of-a-child and confirms both die. It still deploys nothing. Injectable WorkspaceFs/ProcessLauncher seams keep the control tests off a real host. DS05: a durable SQLite queue (durable_queue.py) with leases and an append-only execution journal that survive a restart - a duplicate enqueue never becomes a second job, a crashed worker's lease expires and reconcile() returns the task to queued, a result from a worker that no longer holds the lease is rejected (never a false success), a base that moved since enqueue blocks promotion even on exit 0, and the journal keeps only bounded tails. DS06: an interchangeable AI provider behind a safety contract (ai_provider.py) - only the DETERMINISTIC FAKE ships (kind must be "fake"; a real provider and its authorization are a user decision). run_provider_step() turns a provider timeout / quota exhaustion / malformed output into a bounded named outcome (never an escalating exception), stops before a configured calls/tokens/cost budget would be exceeded, and treats the provider's answer as DATA never instructions - an "ignore previous instructions / deploy now / grant me root" suggestion is copied verbatim, flagged, and still grants_no_permissions / triggers_no_deploy for every outcome; the result is inert (wired to nothing). DS07: an authenticated incident transport (incident_transport.py) for the round trip with HYDRA-UMC-OPS-AGENT - every message is HMAC-signed by a registered node with a nonce, timestamp and contract version; verify_message() rejects an unregistered identity, a bad signature, a node impersonating another, a replayed nonce, a stale timestamp, an overloaded sender and an incompatible version, each with a named code. IncidentSession runs the full submit -> diagnosis -> post-deploy-verification round trip (not just the submit); a dropped connection leaves it pending-ack and reconcile() re-sends the same incident with a fresh nonce, the peer de-duping on the stable incident_id so nothing is lost or double-counted. It is a real protocol object, never a manual file drop. DS08: one fully controlled repair cycle (repair_cycle.py) chaining repro-confirmed -> incident-raised -> candidate-received -> regression-passed -> build-test-passed -> approved -> installed -> verified (or -> recovered / -> blocked). Every step is gated: an unreproduced failure never opens the cycle; a candidate whose HMAC signature does not verify, or that is pinned to a different incident / base fingerprint / target, is blocked; regression and build-test must be genuinely green; an approval must be signed by a registered approver AND issued for this exact base and target; install goes through an injectable isolated-installer seam (never the real target); a failed post-install verification rolls back and ends the cycle recovered, and the I60/T07 apparent-success control voids the result (with rollback) if the base moved mid-cycle or the evidence is for a different repro case. DS09: stable operation + verified restore (operations.py) - create_backup() records a sha256 per file with a backup_id, instance_id and schema_version; verify_backup() re-hashes and flags drift; restore_backup() refuses a backup for a different instance id or schema version, or one whose files no longer match, otherwise restores the exact bytes. check_operational_health() reports a named finding for orphaned (expired-but-leased) entries, a journal over its row cap, and free disk under the floor. All filesystem contact is behind the injectable BackupFs seam. DS10: the delivery package + an HONEST maturity evaluation (delivery.py). build_delivery_manifest() enumerates what the repo actually ships (every module/doc/config plus pyproject/manifest/CHANGELOG/README) with a sha256 per file, and reads back the real CLI surface, version and test count. evaluate_maturity() reports each of the ten deliveries as shipped/partial/not-started against real evidence (DS06 is partial on purpose - only the deterministic fake ships), lists seven known limitations in plain words, and hard-codes overall_maturity to "functional" - the code has no branch that returns anything higher, so the evaluation cannot over-claim. CLI: config / inventory / station / migrate / task / queue / provider / incident / repair / ops / deliver subcommands. The ten-delivery plan is complete; the maturity is "functional" (built and covered by tests, not verified on real hardware or services) - nothing here provisions a host, runs a deploy, or has touched target hardware, and `deliver evaluate` says exactly that. Evidence is simulator: the bench workflows start the swarm cell as real containers, run the bridges in front of an emulator and a Moonraker stand-in, and drive a simulated tool head on a virtual CAN bus, each with named scenarios that include refused and malformed input; no run on real hardware is recorded.
Technology
Python 3.11+JSON Schemasystemd (host role, Linux only)SQLite (durable task/state journal, from DS05 onward)
Build
Ten-delivery plan (DS01-DS10, see this repository's own README Roadmap section): a reproducible development host (target platform: Raspberry Pi 5 or Compute Module 5, both 8GB, NVMe-booted) that hosts this ecosystem's source code, an isolated task/workspace runner for bounded programming/build/test work, and a durable queue coordinating with HYDRA-UMC-OPS-AGENT's own incident lifecycle - never a new AI to train, never a replacement operating system, never itself the authority that decides a machine is safe. Real build: python -m build (see pyproject.toml).
|
||||||||||
| Service | Scaffolding | python | Dev Server | 0.0.4 |
OK
Version successfully resolved
|
Bump the github_actions_updates group with 3 updates (#14) | ||||
|
Notes
Developer node of the HYDRA-UMC ecosystem: reads evidence bundles of the local technician (read only), keeps a cache of checkouts by revision (git worktrees over a bare mirror, with a limit in GB), runs executable repro cases and the shared CI of a workspace only after a person approved the exact plan, and writes a status document for dashboards. Every part is closed by default, every tuned value is in the configuration file and nothing leaves the machine. It has not run on a real development host yet.
Technology
Python 3.11+JSON SchemaSQLitegit worktree
Build
Python package built with python -m build (see pyproject.toml); the standard library only, plus the optional SDK extra for the verdict of a before and after run.
|
||||||||||
| Documentation | Functional | python | User PC | 0.1.5 |
OK
Version successfully resolved
|
Name Python 3.12 in the README source of the URDF editor (#71) | ||||
|
Notes
The documentation and cross-repository tooling hub of the ecosystem: the architecture, the interfaces, the security baseline, the threat model and the capability matrix, written from what the other repositories document and marked where a statement could not be traced. Tools (standard library only): a project catalogue generated from the manifests of the repositories found next to it, a banner, icon and family map generator, a generator of its own eight READMEs, a publication check for anything private, and a script that runs every repository's own CI validation and prints one PASS, FAIL or SKIP line each. Scaffolding on purpose: the tools are unit-tested and run in CI, but the nightly comparison of all the repositories has not yet been exercised over time. Evidence is simulator: the automated test suite runs in CI on the host, and the simulated tool head runs on a simulated bus that loses, doubles and breaks frames and goes down and comes back; nothing here has run against a real head or a physical CAN bus.
Technology
MarkdownMermaidPythonSVG
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.3 |
OK
Version successfully resolved
|
Pin the SDK to its 0.7.3 release (#20) | ||||
|
Notes
Read-only diagnostic CLI for the HYDRA-UMC/URTC ecosystem: ten OBSERVE-level tools with real handlers (storage, ports, processes, systemd units, logs, temperature, connectivity, manifests, pending updates, local knowledge search) behind a symbolic allow-list, secret redaction at every output boundary, JSON Schema contracts (ToolRequest, ToolResult, MaintenanceProposal, EvidenceBundle, PatchVerificationReport), a local TF-IDF knowledge index, evidence-bundle and proposal builders grounded only in valid tool results, and an adversarial test that a poisoned document can never produce a tool execution or leak a secret. Its test suite (`pytest tests`) passes in full. There is no inference engine, no transport for the escalation outbox and no HYDRA-UMC-SERVER integration yet - see the README ROADMAP. Evidence is simulator: the Simulation bench runs the real outbox and HTTP sender against a simulated receiver in CI; nothing here has run against real hardware or a real server.
Technology
Python 3.11+JSON SchemaHailoRT GenAI (optional, later phase)systemd (cm5 role, Linux only)
Build
Six-phase plan (see the README's own Roadmap section): a local, policy-gated AI technician for the ecosystem itself - observes, explains, diagnoses and proposes maintenance, never acts by generating text. The LLM proposes; a deterministic orchestrator validates against real contracts and a real risk-level policy before anything executes, and the highest-risk levels (privileged change, physical action) are not implemented at all in this delivery. Real build: python -m build (see pyproject.toml).
|
||||||||||
| Service | Functional | python | CM5 | 0.2.4 |
OK
Version successfully resolved
|
Bump actions/setup-python (#17) | ||||
|
Notes
Its own real test suite (`pytest tests`) passes in full (a few tests need the optional SDK and are skipped without it), including a real end-to-end canary-deploy cycle (git clone, apply, build, promote) run against a real throwaway git repository, not a mock. Deliveries 1-5 of the 6-delivery plan ("Observabilidad read-only" + "Diagnostico" + "Cambio aprobado" + "Despliegue canario" + "Verificacion") from an explicit recommended-new-project proposal. Coordinates existing pieces (HYDRA-UMC-OS, COGNITIVE-NODE, VOICE-UI, SERVER/SDK, NODE-HEALING, UPDATER, SUITE/STUDIO) rather than reimplementing any of them - this repo owns the maintenance-incident lifecycle (evidence -> diagnosis -> human-approved change -> canary deploy following the same atomic-by-verification pattern as HYDRA-UMC-UPDATER's install.py, as its own implementation -> verification), not inference, updates, or MCU safety themselves. Complementary to the proposal's other recommended project, HYDRA-UMC-CONNECTOR-HUB (external machine/adapter capabilities) - this one operates the health and changes of the ecosystem's own components instead. diagnosis.py is genuinely provider-agnostic (a minimal AIProvider Protocol, two real concrete implementations - Anthropic and OpenAI - either optional); canary_deploy.py never touches a live checkout until an independent staging clone with the approved diff applied has already proven it builds, verified end-to-end against a real throwaway git repository in this project's own test suite (not a mock of subprocess/git). Notifications go to the MQTT broker of the ecosystem because HYDRA-UMC-VOICE-UI has no outbound surface. The optional features are tested against simulated nodes, a simulated broker and a simulated model server, and have not run on a real fleet. Real end-to-end calls against the live Anthropic/OpenAI APIs have not been exercised in this development session (no API key available here); every other real path (missing key, provider error, malformed response, diff-apply failure, build failure, promotion) is real and tested, including a live local git clone+apply+build+promote cycle.
Technology
Python 3.11+JSON Schemasystemd (edge role, Linux only)two optional AI provider APIs (provider-agnostic diagnosis)git (staging clone + git apply for the real canary deploy path)
Build
Two roles in one package: the edge role (deployed on a CM5) collects a real, sanitized inventory (installed project versions, service health checks, redacted logs) and derives read-only observations; the control-plane role (deployed on a development host, e.g. a Raspberry Pi 5) renders a saved snapshot, asks a provider-agnostic AI provider (Anthropic and/or OpenAI, optional `[ai-anthropic]`/`[ai-openai]` extras) to propose a root-cause explanation, manages a human-approved change proposal, applies a change approved with an Ed25519 signature to a staged clone and promotes it only if that project's own real build-test passes (optionally stopping and health-checking a service around the swap, with automatic rollback), records everything in a hash-chained audit journal, and re-verifies the original check. Optional, off by default and configured in one file: an authenticated edge-to-control channel, incidents from the node watchdog, notifications to the MQTT broker, a canary across several nodes, a proposal drafted by a local model (always pending) and Prometheus metrics. Real build: python -m build (see pyproject.toml).
|
||||||||||
| Tool | Functional | python | User PC | 0.3.9 |
OK
Version successfully resolved
|
Make the partitions of the base image an option of the platform instead of a fixed boot and root pair (#31) | ||||
|
Notes
Its own real test suite (`pytest tests`) passes in full. The real remote-build path (SSH to a Linux host, chroot, profile-build reproducibility) has already been exercised and fixed for real against the CM5's own ecosystem checkout. Windows/Linux desktop tool, CLI and Qt Quick GUI (reusing HYDRA-UMC-UPDATER's own visual language). Depends on hydra-umc-updater as a library for ecosystem discovery/version-checking rather than duplicating that logic. The image-build pipeline (loop-mount, chroot) is Linux-only and reports that honestly on Windows instead of pretending to run. Evidence is simulator: the automated test suite runs in CI on the host and a simulation bench runs the image pipeline, the first-boot files, the release and golden-image commands, the profile signing and the boot test over fakes (a recorder instead of the build, a launcher instead of QEMU, a twin of the Linux host that makes each step fail); nothing here has built, flashed or booted a real image on a real board.
Technology
Python 3.10+PySide6QMLRaspberry Pi OSGitHub raw content
Build
Builds a ready-to-flash Raspberry Pi OS image for the CM5, pre-loaded with the most current version of every ecosystem project it can reach on GitHub, and lets the operator set first-boot Wi-Fi/user/hostname/SSH configuration before writing the image - the same customization Raspberry Pi Imager itself offers, one step earlier.
|
||||||||||
| Tool | Established | python | User PC | 0.5.9 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#27) | ||||
|
Notes
Administrative tooling for the ecosystem catalog. Now included in the public dashboard's own project count (JuanenRac/ecosystem-catalog.json's dashboard_exclude list no longer excludes it). Real bounded retry with backoff for transient GitHub network errors; a malformed remote catalog fails loudly (caught by both gui.py and main.py, falling back to local-only state) while one malformed manifest is isolated without aborting discovery of the rest. install.py's clone_or_pull never replaces a healthy local checkout - pull is always --ff-only, never a reset. Updating an existing checkout is now atomic-by-verification: the candidate is cloned+merged+built in an independent staging clone first, and only promoted (via two back-to-back directory renames, previous install kept at <name>.backup-xxxxxxxx) once that build actually succeeds - a build failure, crash, or full disk before promotion leaves the previous installation completely untouched. v0.3.0: the staging clone's own git remote is reset to the real GitHub upstream immediately after creation (a plain `git clone --local` used to leave it pointing at the local install path instead, silently stopping all future updates once promoted), and the installed checkout's own real local data (anything git considers untracked/ignored, e.g. data/settings.json - excluding regenerable build artifacts) is carried into the staging clone before it is built, so a project's own real operational state survives an update instead of being stranded in the kept backup checkout. A real, genuinely uncommitted edit to an already-TRACKED file is a separate case - checked explicitly upfront and refused outright, never silently discarded via the same staging-clone path. Evidence is host-test: the automated test suite runs in CI on the host; nothing here has run against a simulator bench or real hardware.
Technology
Python 3.10+PySide6 / Qt Quick (QML)Tkinter (legacy fallback)GitGitHub raw content
Build
Detects local projects, compares their manifests with GitHub and performs explicit install or update actions.
|
||||||||||
| External Automation Bridges 5 project(s) | ||||||||||
| Service | Functional | python | CM5 | 0.2.9 |
OK
Version successfully resolved
|
Pin the SDK to its 0.7.3 release (#33) | ||||
|
Notes
High-level CNC cell coordinator. It normalizes read-only controller evidence and can now query real GRBL status and send GRBL's own real-time control bytes (status query, feed hold, cycle resume, soft reset) over a real serial connection; the controller (GRBL today) retains real-time trajectory, limits, spindle and safety authority - this never streams a G-code program. Interlocks (door/E-STOP) are always re-read live immediately before cycle_start_resume/a new job is gated - no cached snapshot is ever reused for a real safety decision, and (v0.0.9) the live read itself now happens AFTER the blocking GRBL status query completes, not before it, so a real E-STOP/door event during that query is never missed. A real short/zero-byte serial write is reported as not executed, never as a false success; v0.1.0: this now also requires an exact, real int byte count equal to the command length - a None/bool/mismatched-but-real return value from a non-conforming serial-like implementation is no longer silently trusted as a confirmed write either.
Technology
PythonLinuxCNCG-codeGRBL status evidenceGRBL serial transportHYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Pin the SDK to its 0.7.3 release (#28) | ||||
|
Notes
High-level laser cell coordinator. It normalizes read-only safety evidence and can now read the 3 independent key/enclosure/interlock safeguards from real GPIO lines (one shared line request, wiring declared per line), controller-neutral by design; it cannot arm or fire a laser, and preserves the controller's own authority. The 3 safeguards are always re-read live immediately before a new job is gated - no cached snapshot is ever reused for a real safety decision.
Technology
PythonLaser controllerG-codeSafety evidenceGPIO interlock readingHYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Bump the github_actions_updates group across 1 directory with 3 updates (#16) | ||||
|
Notes
Safe high-level board-flow coordinator for OpenPnP. It never claims placement kinematics or bypasses a pick-and-place controller. v0.1.2: the test suite's own unknown-phase reproduction (a raw string BridgeJob construction) was updated to match HYDRA-UMC-SDK's own real construction-time rejection - split into a real construction-rejection test plus a minimal double proving BoardFlow.plan()'s own defensive fallback still fails safe. v0.1.3: BoardIdentity gained optional slot_id/rack_id/table_id fields feeding into fingerprint(), so two physical placements of the same board are never fingerprinted as indistinguishable (I46).
Technology
PythonOpenPnPHYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Bump actions/setup-python (#33) | ||||
|
Notes
Safe coordination boundary for native 3D-printing controllers. It checks Moonraker readiness, inspects slicer artifacts and matches profile evidence read-only, and now sends only fail-closed, SDK-gated real job commands (start/pause/resume/cancel, plus a hard emergency stop) through Moonraker's own REST API - every command still passes the same evaluate_job() safety gate every bridge in this ecosystem uses. It never streams raw G-code and never replaces firmware, heater interlocks or motion authority - Moonraker/Klipper keep that. start_job() only ever accepts a genuine PROCESS-phase job (ABORT is refused outright, never reaching print/start), and print_stats parsing correctly unwraps Moonraker's own real result envelope.
Technology
PythonMoonrakerKlipperG-code3MFResin slice artifactsPrint profile evidenceHYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.0 |
OK
Version successfully resolved
|
Pin the SDK to its 0.7.3 release (#37) | ||||
|
Notes
Dependency-free safety coordinator with a ROS 2 interface plan, plus a real, lazily-imported rclpy transport for the 2 interfaces that have a real standard ROS 2 message type today (safe_stop_service via std_srvs/Trigger, state_topic via std_msgs/String) - inspect_service/job_action still need a custom .srv/.action definition this repository doesn't have yet. The optional rclpy runtime is intentionally not installed until it is tested against a real ROS 2 deployment. The safe-stop call uses call_async()+spin_until_future_complete() with a real bounded timeout on the response itself, not just on service availability. v0.0.7: the test suite's own unknown-phase reproduction (a raw string BridgeJob construction) was updated to match HYDRA-UMC-SDK's own real construction-time rejection - split into a real construction-rejection test plus a minimal double proving the coordinator's own defensive fallback still fails closed.
Technology
PythonROS 2rclpyHYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Industrial Gateway 4 project(s) · HYDRA-UMC-GATEWAY-INDUSTRIAL is this family's own integration parent | ||||||||||
| API | Functional | node | CM5 | 0.2.1 |
OK
Version successfully resolved
|
Bump supertest from 7.3.1 to 7.3.2 in the npm_updates group (#23) | ||||
|
Notes
Functional v0: GET /status makes a real, live reachability check against all 3 children on every request - verified end-to-end, including detecting a real killed sibling. Plus a real command allowlist + backpressure + timeout decision layer (src/command.ts, CommandDispatcher) in front of POST /command: default-deny per-protocol allowlist (only read-like/publish operations allowed in v0), bounded concurrency (no more than maxConcurrent commands in flight, rejected immediately beyond that), and a real timeout - all enforced before any relay attempt, mapped to real HTTP status codes (403/429/504/502). The default executor reuses the existing real reachability probes; an actual protocol-level write is not implemented yet, honestly. src/tls.ts now provides real, opt-in TLS 1.2+/mutual-TLS on this gateway's own HTTP API (TLS_CERT_PATH/TLS_KEY_PATH, optionally TLS_CLIENT_CA_PATH), closing a real gap where the README already advertised mTLS as delivered while the server was actually plain HTTP - verified with a genuine end-to-end mTLS handshake test against real openssl-generated certificates. Evidence is simulator: the test suite runs the Modbus TCP adapter and its write gate end to end against a simulated device, including exception responses and timeouts; no run against a real PLC or meter is recorded.
Technology
Node.jsExpressreal TCP/HTTP reachability probes
Build
npm install && npm run build
|
||||||||||
| Service | Functional | node | CM5 | 0.2.2 |
OK
Version successfully resolved
|
Bump actions/setup-node (#25) | ||||
|
Notes
Real MQTT broker over real TCP. Optional MQTT CONNECT username/password authentication (MQTT_AUTH_JSON) gives the opt-in per-client-ID-prefix topic ACL (MQTT_ACL_JSON) a verified session identity; ACL scope checks prevent wildcard SUBSCRIBE escalation. An optional MAX_PAYLOAD_BYTES cap is enforced on every PUBLISH. An optional retainedTtlMs/MQTT_RETAINED_TTL_MS actively expires a retained message older than the limit instead of retaining it forever. All controls are verified with real Aedes and MQTT clients, not only unit math. Evidence is simulator: the automated test suite runs in CI on the host, and a simulation bench runs the broker with a simulated fleet that goes silent, takes over identities, is away and returns, and survives a broker restart; nothing here has run on real hardware.
Technology
Node.jsTypeScriptAedes MQTT broker (MQTT 3.1.1)MQTT over WebSockets (ws)TLS and mutual TLS (optional)
Build
npm install && npm run build.
|
||||||||||
| API | Functional | node | CM5 | 0.2.4 |
OK
Version successfully resolved
|
Bump actions/setup-node (#27) | ||||
|
Notes
Real MTConnect /probe and /current XML endpoints, spec-shape verified with real HTTP tests (supertest). Real, versioned DataItem mapping (unit conversion, quality, UTC timestamp, error code - src/units.ts/dataitem.ts) separated from reading a machine (src/reader.ts's polling-frequency-limited CachedReader), with real degraded-mode output (MTConnect's own UNAVAILABLE value) when a source is down or returns invalid data. src/hydraServerReader.ts (HYDRA_UMC_SERVER_URL) is a real, non-fixture MachineReader polling a live HYDRA-UMC-SERVER's own GET /api/settings for one real robot's own AVAILABILITY/EXECUTION - closing a real gap where FixtureMachineReader was the only reader that ever existed, and Execution/Availability were hardcoded XML literals regardless of which reader was configured. Verified live against a real running HYDRA-UMC-SERVER instance, not just unit tests. Evidence is simulator: the automated test suite runs in CI on the host and a simulation bench runs the SHDR reader and server against a machine that misbehaves, and the readers of HYDRA-UMC-SERVER and of the MQTT broker against a server and a broker that fail, on a fixed seed; nothing here has run against real hardware, a real machine, the real server or a real broker.
Technology
Node.jsExpressMTConnect 1.7 XML
Build
npm install && npm run build
|
||||||||||
| Service | Functional | node | CM5 | 0.2.9 |
OK
Version successfully resolved
|
Bump actions/setup-node (#33) | ||||
|
Notes
Real OPC-UA address space, verified with a real OPCUAClient session over the real binary protocol (browse, read, write). Real, explicit, versioned namespace URI and stable string NodeIds; a real quality/units/UTC-timestamped DataItem (SpindleTemp, real EngineeringUnits); real per-session read/write authorization on MaintenanceMode (anonymous read-only, authenticated write via OPCUA_ADMIN_USERNAME/PASSWORD). Every real session now requires Sign/SignAndEncrypt with a real security policy (Basic256Sha256/Aes128_Sha256_RsaOaep/Aes256_Sha256_RsaPss) - a real SecurityPolicy.None connection is genuinely refused, closing a real gap where node-opcua's own unconfigured defaults still exposed one; OPCUA_ALLOW_INSECURE=1 is an explicit, off-by-default local-dev escape hatch, verified with a real protocol-level test proving both the refusal and the escape hatch actually work. Evidence is simulator: the automated test suite runs in CI on the host and a simulation bench reads the server with a real OPC UA client while a data lake, a broker with a vehicle and the HYDRA-UMC-SERVER of the roster fail, on a fixed seed; nothing here has run against real hardware, a real SCADA, a real broker or a real data lake.
Technology
Node.jsnode-opcuaOPC-UA address space
Build
npm install && npm run build
|
||||||||||
| Mobile & Autonomous Bridges 3 project(s) | ||||||||||
| Library | Functional | python | CM5 | 0.2.2 |
OK
Version successfully resolved
|
Bump the github_actions_updates group across 1 directory with 3 updates (#20) | ||||
|
Notes
Real, dependency-free coordination boundary for AGV/AMR fleets: a real, hand-checkable 2D rigid-body transform (FrameTransform, now carrying a required real map_id per site/AMR) from the factory's shared coordinate frame into a specific AMR's local frame, a minimal VDA-5050-inspired order-action vocabulary (MOVE_TO_*/PICK_LOAD/DROP_LOAD/CANCEL_ORDER) gated through HYDRA-UMC-SDK's shared safety contract, and a real VDA 5050 MQTT publisher (Vda5050Publisher) that only ever publishes an already-gated dispatch to the correct real order/instantActions topic, checks the real MQTT client rc before ever reporting a dispatch as delivered, and reports success honestly as "queued for delivery", not AGV acceptance. Never plans a path or avoids an obstacle itself - that authority stays with the AMR's own onboard navigation or its fleet manager. Not yet exercised against a real fleet manager or physical AMR. v0.0.6: the test suite's own unknown-phase reproduction (a raw string BridgeJob construction) was updated to match HYDRA-UMC-SDK's own real construction-time rejection - split into a real construction-rejection test plus a minimal double proving the coordinator's own defensive fallback still fails closed.
Technology
PythonVDA 5050 (subset)MQTTHYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Library | Functional | python | CM5 | 0.1.9 |
OK
Version successfully resolved
|
Pin the SDK to its 0.7.3 release (#27) | ||||
|
Notes
Real, dependency-free coordination boundary for legged/humanoid droids: validates a small named action-trigger vocabulary (WALK_TO/PICK_OBJECT/PLACE_OBJECT/RETURN_HOME/HOLD_POSITION/STAND/SIT) and its real required parameters before forwarding through HYDRA-UMC-SDK's shared safety gate, plus a real bosdyn-client command sender (SpotDroidControl) for Boston Dynamics Spot, lazily imported. Never computes gait/balance/joint trajectories itself - that authority stays with the droid's own onboard controller. Not yet exercised against a real Spot robot. v0.0.5: the test suite's own unknown-phase reproduction (a raw string BridgeJob construction) was updated to match HYDRA-UMC-SDK's own real construction-time rejection - split into a real construction-rejection test plus a minimal double proving the coordinator's own defensive fallback still fails closed.
Technology
PythonBoston Dynamics Spot SDK (bosdyn-client)HYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Library | Functional | python | CM5 | 0.2.4 |
OK
Version successfully resolved
|
Release 0.2.4 | ||||
|
Notes
Real, dependency-free coordination boundary for camera-equipped UAVs: a named high-level flight-request vocabulary (ARM/TAKEOFF/GOTO_WAYPOINT/HOVER_AND_CAPTURE/RETURN_TO_LAUNCH/LAND) gated through HYDRA-UMC-SDK's shared safety contract, a real, deterministic HeartbeatMonitor link-loss failsafe state machine, and a real MAVLink command sender (MavlinkFlightControl) mapping each request to its real, numbered MAV_CMD - only an already-gated dispatch is ever sent. A GOTO_WAYPOINT goes out as a COMMAND_INT with an explicit altitude frame (relative by default), an optional FlightEnvelope limits altitude and distance from home, and RETURN_TO_LAUNCH/LAND/DISARM are always sent. ARM (renamed from the misleading PRE_FLIGHT_CHECK) requires the literal boolean confirm_arm is True (v0.0.6: never a merely-truthy value like the string 'false'); takeoff altitude and waypoint lat/lon/alt are validated as real, non-boolean, finite numbers within real geographic range before ever reaching a real COMMAND_LONG. Never computes flight control or stabilization itself; real-time failsafe execution stays the flight controller's own authority. Not yet exercised against a real flight controller or SITL.
Technology
PythonMAVLinkHYDRA-UMC-SDK
Build
python tools/build_test.py
|
||||||||||
| Orchestration & Swarm 5 project(s) · HYDRA-UMC-ORCHESTRATOR is this family's own integration parent | ||||||||||
| Service | Functional | rust | CM5 | 0.2.7 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#27) | ||||
|
Notes
Functional v0: a real mission state machine (src/mission.rs: Pending, Dispatched, InProgress, Completed, Cancelled, Failed, Unknown after an unclean restart; idempotent cancel; node-failure recovery; an attempt number that fences late reports and moves forward on every dispatch and every requeue) kept on disk, served over plain JSON/HTTP (tiny_http, a pool of worker threads) with an optional bearer token. src/job_dispatcher.rs is a real HTTP client for HYDRA-UMC-JOB-DISPATCHER (submit with routing hints, dispatch pass, close by cancelling queued work or completing assigned work, requeue after a lost node) with connect and total timeouts. A mission's terminal outcome is recorded as unconfirmed in the registry and in a durable outbox (src/outbox.rs) before the dispatcher is asked, and retried with backoff. HYDRA-UMC-NODE-HEALING calls POST /nodes/:node/recover. The shared gRPC contract in proto/ is not used by this binary. Contract test against the real dispatcher binary (HYDRA_UMC_JOB_DISPATCHER_BIN).
Technology
Rusttiny_httpureqJSON/HTTP
Build
cargo build --release.
|
||||||||||
| API | Functional | go | CM5 | 0.3.1 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#25) | ||||
|
Notes
Real priority-based job queue with a real HTTP API, tested. Deterministic priority ordering is now covered by an explicit repeated-run regression test. Engine.SubmitJob() adds real, opt-in DedupKey-based deduplication (POST /jobs/submit alongside the unchanged POST /jobs): a repeat submission of an in-flight or already-done job is returned unchanged, and a retry after a real failure reuses the same job ID instead of running the work twice. NewEngineWithStore() (src/sqlitestore) adds real, opt-in (-db PATH) SQLite persistence - the mission queue now survives a real process restart, verified end-to-end with an actual killed-and-relaunched binary, not just unit tests; GET /health reports a degraded store via LastPersistError() instead of failing silently.
Technology
Gopriority queueHTTP APISQLite (modernc.org/sqlite, pure Go)
Build
go build ./...
|
||||||||||
| Service | Functional | go | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#21) | ||||
|
Notes
Functional v0: health watchdog (gRPC HealthService or HTTP health URL) with guarded recovery - OrchestratorReactor (watchdog.Reactor's real implementation) calls HYDRA-UMC-ORCHESTRATOR's own POST /nodes/:node/recover only for registry entries with a dispatchNode, for a node that stays UNREACHABLE, INVALID or UNHEALTHY past a confirmation wait, a start-up grace, a per-node cooldown and a global budget (a MISCONFIGURED health URL never recovers). checkNode retries a transport-level failure (dial/RPC error) via a bounded, deterministic RetryPolicy (default 3 attempts, exponential backoff 250ms->2s cap) before classifying a node UNREACHABLE, masking transient network blips without an open-ended retry loop. A node that answers but misreports its own identity (missing or mismatched NodeIdentity.Name) is classified StatusInvalid and never trusted or retried - retrying cannot fix a node lying about its own name. A registry entry with a healthUrl is probed with one HTTP GET (2xx healthy, 5xx unhealthy, no answer unreachable and retried, redirect or 4xx invalid and never retried) because no service of the ecosystem implements the gRPC HealthService; the recovery request carries a bearer token when -orchestrator-token or ORCHESTRATOR_TOKEN is set. Evidence is simulator: the Simulation bench runs the real watchdog and reactor against simulated services and a simulated orchestrator in CI; nothing here has run against a real fleet.
Technology
GogRPC health checksHTTP health probes
Build
go build ./...
|
||||||||||
| Library | Functional | rust | CM5 | 0.1.5 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#34) | ||||
|
Notes
Real RRT-based 3D path planner, tested against real obstacle/workspace scenarios (RRT with an optional rewiring pass in the style of RRT*, off unless `optimize_iterations` is above zero). Real wall-clock planning time limit (PlannerConfig.max_duration_ms) independent of max_iterations, a reusable obstacle/workspace test corpus, and a new `validate` subcommand that re-checks an externally-supplied path for safety (workspace bounds, obstacle clearance, segment-level collisions) before it is trusted for real execution. The nearest-neighbour search is a KD-tree rebuilt balanced as it grows and walked without recursion; the workspace is the volume the whole body must stay in (robot_radius is kept clear of its walls); validate also requires the path to begin at the scenario start and end at its goal; stdout carries only the JSON result.
Technology
RustRRT* path planning
Build
cargo build --release
|
||||||||||
| Service | Functional | rust | CM5 | 0.2.1 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#30) | ||||
|
Notes
Real CRDT LWW-Element-Map implementation with property-tested commutativity/associativity/idempotence, backed by a real Lamport logical clock for event ordering. merge_report() gives real, inspectable per-key conflict visibility (which cell's write beat which other's), and a 4-cell multi-round simulated partition/reconnect test proves eventual convergence at swarm scale, not just a single two-cell merge. src/store.rs (serve --state-file) adds real, opt-in per-node persistence - POST /reconcile now merges into and durably updates this node's own state instead of discarding it every call, verified end-to-end against a real killed-and-relaunched binary, closing a real gap where the server was fully stateless. merge_report() now also rejects (IdentityCollision) the one case the CRDT's own single-writer invariant should make impossible - an identical (time, writer) stamp claiming two different values - instead of silently, order-dependently keeping one side; LamportClock::tick/observe use checked_add and a real ClockOverflowError instead of silently wrapping at u64::MAX.
Technology
RustCRDT (LWW-Element-Map)Lamport logical clockJSON/HTTP
Build
cargo build --release.
|
||||||||||
| Platform Foundation & Contracts 3 project(s) | ||||||||||
| Tool | Functional | python | User PC | 0.2.1 |
OK
Version successfully resolved
|
Bump actions/setup-python (#23) | ||||
|
Notes
Its own real test suite (`pytest tests`) passes in full (the tests that need the optional `[sdk]` extra or real hardware and are not silently counted as passing). v0.0.4 - a second independent review pass found and fixed 4 more real issues: the capability gate now enforces a write/abort capability's own declared requiredPermission/requiredCellState/requiresHumanConfirmation and the manifest's own requiredSafetyGates (previously ignored - a call was gate-allowed on the generic SDK motion check alone, zero policy enforced); adapterId is now restricted to a safe charset and save_certification_record() independently verifies its resolved path stays inside the target directory (previously a path-traversal adapterId could write outside it); the certification store now creates records with real atomic O_CREAT|O_EXCL semantics, refusing to silently overwrite a different record under the same id (previously not actually append-only despite documenting it as such); the JSON-Schema-subset validator no longer accepts a bool as a number or raises on an invalid regex pattern/non-string required entry (previously it could crash or silently accept wrong-typed evidence). Deliveries 1-4 of the 4-delivery plan ("Esquema JSON, CLI validate y diez fixtures" / "Registro ... y endpoint read-only de catalogo" / "Integracion con SDK: validacion de capability + safety gate + evidencia" [P0] / "Certificar adaptadores uno a uno sobre sus maquinas" [P1][HARDWARE]). Complementary to HYDRA-UMC-OPS-AGENT (maintenance-incident lifecycle for the ecosystem's OWN components) - this one is about discovering, validating and safely invoking what an EXTERNAL machine/adapter can do, never about CAN/SPI/UART access or motor control directly (those stay owned by HYDRA-UMC-SERVER/SDK/MCU). Explicitly does not replace HYDRA-UMC-GATEWAY-INDUSTRIAL, HYDRA-UMC-OPCUA-SERVER, HYDRA-UMC-MQTT-BROKER or HYDRA-UMC-MTCONNECT-ADAPTER - it is a declarative registry/validator/gate ABOVE them, not a second protocol implementation. The ten fixtures reference those real projects (and the real bridge repos) as their own ownerProject, never an invented one. A capability whose mode is write/abort is rejected by the validator unless it also declares risk, requiredPermission, requiredCellState and timeoutMs - an explicit rule ("las acciones write nunca son seleccionables si falta una de esas condiciones"), enforced as a real, tested check. Delivery 3's own gate never reimplements HYDRA-UMC-SDK's safety logic - it calls the SDK's real `evaluate_job()` directly (a `read` capability skips the SDK entirely; `write`/`abort` map onto a documented, honestly-simplified single JobPhase - see sdk_gate.py's own module docstring for exactly why). Delivery 4's own certification log is real, append-only record-keeping of a human-attested evidence payload checked against the adapter's own evidenceSchema - it explicitly does NOT and cannot independently verify that evidence came from real hardware without the real hardware itself, which this development machine does not have; that gap is documented, not hidden.
Technology
Python 3.11+JSON Schema (hand-written structural validator, no jsonschema dependency)http.server (stdlib-only real read-only catalog endpoint)HYDRA-UMC-SDK (optional, real safety-gate integration via evaluate_job())
Build
A real, hand-written structural validator (schema.py, now also exposing a small JSON-Schema-subset checker for evidence/config payloads) for this ecosystem's "adapter-manifest" contract; a real CLI (`hydra-umc-connector-hub validate/catalog/serve-catalog/gate/certify/certifications`); ten real fixtures, one per already-existing bridge/protocol project this hub registers; a real, GET-only http.server catalog endpoint (registry.py + catalog_server.py); a real integration with HYDRA-UMC-SDK's own evaluate_job() safety gate (sdk_gate.py, optional `[sdk]` extra); and a real, append-only certification-record log (certification.py). Still never touches CAN/SPI/UART or a real machine directly - the catalog only serves manifests already on disk, and certification records a human's own attested evidence rather than independently probing hardware. Real build: python -m build (see pyproject.toml).
|
||||||||||
| Service | Functional | python | CM5 | 0.6.4 |
OK
Version successfully resolved
|
Pin the SDK to its 0.7.3 release (#32) | ||||
|
Notes
A reproducible HYDRA-UMC product layer on Raspberry Pi OS, not a new Linux distribution. Provides a read-only agent, validated configuration, profiles, diagnostics, Server staging and an optional loopback Voice UI gateway. The preflight is proven idempotent and side-effect-free by a real test (two runs, byte-identical output, zero files touched), and a real, host-independent backup/rollback mechanism (provisioning/rollback.py) protects the one system file install_local_agent.sh unconditionally overwrites. Real WiFi first-contact provisioning (provisioning/wifi_provision.py, hydra-umc-wifi-provision.service): a real NetworkManager AP-mode fallback + local HTTP form for a headless CM5 with no known network yet, state machine unit-tested against a fake NetworkManager including a real HTTP round-trip (tools/verify_wifi_provision.py). CM5 validation and production image assembly remain release gates.
Technology
Raspberry Pi OS ARM64Debian/aptsystemdudevNetworkManager
Build
Functional read-only CM5 platform agent, validated configuration, systemd unit and host-side tests.
|
||||||||||
| Library | Established | python | User PC | 0.7.4 |
OK
Version successfully resolved
|
Bump jsonschema in /clients/rust in the cargo_updates group (#83) | ||||
|
Notes
The common contract and conformance boundary for services, UIs, CM5 adapters and URTC. Provides the v1 JSON Schemas, a Python reference validator, Go, TypeScript and Rust clients that enforce the published schemas, conformance fixtures, a schema/validator compatibility matrix (tools/verify_contract_matrix.py), a differential test of the Python validator against the schemas, and an OpenAPI v1 slice of HYDRA-UMC-SERVER's discovery, sign-in, settings and command routes (a subset of the server's own docs/openapi.json). Bridge helpers: BridgeJob/GateDecision, job intake with an idempotency store, MQTT connection settings, an Operation lifecycle with evidence-based success, Capability checks and a durable promotion journal. Protobuf is not published yet. Evidence is simulator: the integration workflow runs the container bench of the swarm cell against the contract on pull requests that touch it; no run on real hardware is recorded.
Technology
PythonTypeScriptGoRustJSON SchemaOpenAPI
Build
Functional JSON Schema v1 contracts, a real first OpenAPI v1 slice for HYDRA-UMC-SERVER's own HTTP surface, plus real, tested Python, Go, TypeScript and Rust reference clients and conformance fixtures.
|
||||||||||
| Vision AI Node 5 project(s) · HYDRA-UMC-VISION-NODE is this family's own integration parent | ||||||||||
| Service | Functional | python | CM5 | 0.2.4 |
OK
Version successfully resolved
|
Refresh the project status lights and test counts of the README (#18) | ||||
|
Notes
Real v0 family-readiness check (`family-status` subcommand): reads each of the 4 real children's own hydra-umc.project.json and reports presence/version/maturity/role. Real pipeline manifest (pipeline.py) describing the perception pipeline's shape and per-stage hardware needs, real hardware-independent frame corruption validation (frame.py), and real degraded-mode detection (hardware.py) that honestly probes for a real camera/Hailo-8 device and reports exactly which stages can run - `pipeline-status` and `validate-frame` subcommands. Honest for an integration parent that runs no Hailo-8 runtime or camera pipeline itself - the real capture loop and Hailo-8 inference runtime stay future work, blocked on real CM5+Hailo-8 hardware this environment does not have.
Technology
PythonCM5Hailo-8
Build
python tools/build_test.py
|
||||||||||
| Service | Functional | python | CM5 | 0.2.1 |
OK
Version successfully resolved
|
Bump python from 3.12-slim to 3.15-slim in the docker_updates group (#19) | ||||
|
Notes
Functional v0: real compiled-model registry - JSON schema validation (including a required, validated hailo_arch field), duplicate-version detection, latest-version lookup by name/task, and sha256 checksum verification against local files, exposed via 'registry validate'/'registry latest'. Real, combined safe-load gate (compatibility.py's safe_load()) checks Hailo-architecture compatibility before checksum integrity and never reports a model ready unless both real checks pass - exposed via the new 'registry load' subcommand. hef_path is now validated against path traversal (rejects an absolute path at parse time, and verify_checksum() confirms the resolved join stays inside models_dir) before any file is read or hashed. The ONNX/Dataflow-Compiler/HEF toolchain that would populate the registry still needs real Hailo hardware and is future work.
Technology
PythonHailo SDKHEF model registry (no model compiled yet)
Build
Requires Hailo runtime/SDK on the real CM5 for the compilation toolchain itself (ONNX export, Dataflow Compiler, HEF packaging); the model registry (registry.py) needs neither.
|
||||||||||
| Service | Functional | python | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Refresh the project status lights and test counts of the README (#26) | ||||
|
Notes
Real v0 zone-breach checking and E-STOP requesting (`check` subcommand): real AABB zone geometry, real Warning/Danger breach detection, and a real EStopRequester interface that only ever requests a stop, never asserts one - honest for a safety service whose enforcement boundary is deliberately outside this repo. Real calibration-freshness enforcement (calibration.py/safety_state.py): a zone set with no calibration metadata, or one older than its own declared max_age_days, forces SafetyState.INHIBITED before any breach logic runs, never falls through to a silent READY. Real 3D occupancy mapping from Hailo-8 and the real CAN transport stay future work, both genuinely blocked on hardware this environment does not have.
Technology
Python3D visionHailo-8
Build
python tools/build_test.py
|
||||||||||
| Service | Established | python | CM5 | 0.2.7 |
OK
Version successfully resolved
|
Bump the github_actions_updates group with 2 updates (#9) | ||||
|
Notes
Established v0: real per-camera config validation, GStreamer pipeline description generation, and MediaMTX paths.yml generation, exposed via 'config validate'/'config gst'/'config mediamtx'. Real bounded frame buffering (drop-oldest backpressure, provably bounded memory under a slow consumer) and a real deterministic reconnect/backoff policy for a dropped camera link, exposed via 'stream simulate'. Added hailo_runtime.py: a real HailoRT (hailo_platform) integration boundary - open_vdevice()/load_hailo_detection_model() against the real confirmed API, lazily imported so this dev machine (no hailort installed) degrades to a clear RuntimeError; real pre-flight frame/input-shape validation against config.py's CameraConfig, fully unit-tested without hailort. mjpeg_server.py's real capture path now buffers a Frame (frame_id/session_id/capture_ts, not just bare bytes) - a genuinely new session_id on every real reconnect, HTTP wire format unchanged. Actually running the generated pipeline (and parsing a real model's NMS output) against physical USB cameras and a real Hailo-8 module is future work.
Technology
PythonGStreamerMediaMTX
Build
python tools/build_test.py
|
||||||||||
| API | Functional | python | CM5 | 0.2.6 |
OK
Version successfully resolved
|
Release 0.2.6 (#16) | ||||
|
Notes
Functional v0: real Position-Based Visual Servoing (PBVS) correction law - pose-delta computation with shortest-turn angle wrapping, proportional velocity command with direction-preserving clamping, and convergence check, exposed via the 'correct' subcommand. Plus a safety-gated authorization layer (authorization.py) that refuses to turn a visual detection into motion unless the upstream safety_state is READY and the data is confident/fresh enough (INHIBITED wins over REJECTED, mirroring HYDRA-UMC-SAFETY-ZONES), exposed via the 'request' subcommand. Added hailo_runtime.py: a real HailoRT (hailo_platform) integration boundary - open_vdevice()/load_hailo_pose_model() against the real confirmed API, lazily imported so this dev machine (no hailort installed) degrades to a clear RuntimeError; hailo_output_to_pose() adapts a real InferVStreams result into this project's own Pose6D and is fully unit-tested without hailort. A non-object top-level JSON body and an extreme-but-individually-finite pose delta/gain combination that could overflow to inf/NaN now both surface as a real 400 instead of crashing the handler thread or returning a corrupted velocity command. Real 6-DOF pose *estimation* (needs the Hailo-8 NPU and a real compiled .hef) and the low-latency gRPC feed to the HYDRA-UMC core remain future work.
Technology
Python6-DOF pose estimation
Build
python tools/build_test.py
|
||||||||||
| URTC - an independent product with its own firmware and maintenance tools, coordinated with HYDRA-UMC over FDCAN. | ||||||||||
| Ecosystem Operations 3 project(s) | ||||||||||
| Tool | Scaffolding | python | User PC | 0.0.8 |
OK
Version successfully resolved
|
ci: analyze the workflows with zizmor and document the workflow_run trigger (#16) | ||||
|
Notes
The shared repository tools of the URTC family as a versioned set: the canonical copy of the documentation-policy and README-parity checks, the picture generators, the live panel of the build scripts, the manifest version bumper, the status-light check and the dashboard notice (files/), with the version of the repository as the version of the set. A repository takes the set with the synchroniser of URTC-DOCS, which writes a lock file with the version and the SHA-256 of every file; the check compares the lock with the files, so a hand edit and a copy that fell behind are both found. Every repository stays buildable with its own checkout. A virtual family of repositories (tools/fleet_simulation.py) takes the set, is broken in the ways the checks know and heals, so the set is exercised from end to end without the real repositories. Scaffolding on purpose: the repositories of the family have not yet moved to the lock.
Technology
PythonGitHub ActionsJSON
Build
python tools/build_test.py
|
||||||||||
| Documentation | Functional | python | User PC | 0.1.8 |
OK
Version successfully resolved
|
Carry the reason for the workflow_run trigger in the template of the dashboard notice (#54) | ||||
|
Notes
The documentation and cross-repository tooling hub of the URTC ecosystem: the architecture, the interfaces, the security baseline, the threat model, the capability matrix and a study note on the implementations of the CAN protocol, written from what the other repositories document and marked where a statement could not be traced. Tools (standard library only): a project catalogue generated from the manifests of the repositories found next to it, banner, icon and ecosystem map generators, a generator of its own eight READMEs, a publication check for anything private, and urtc_doctor, which runs every repository's own CI validation and checks the shared copies and the firmware images. A virtual ecosystem (src/urtc_docs/simulation.py) runs simulated nights of the comparison with faults and repairs; the nightly comparison of all the repositories has not yet been observed over real days.
Technology
MarkdownMermaidPythonSVG
Build
python tools/build_test.py
|
||||||||||
| Tool | Established | python | User PC | 0.3.2 |
OK
Version successfully resolved
|
Take the shared tool set 0.0.8 lock (#28) | ||||
|
Notes
Adapted from HYDRA-UMC-UPDATER's own proven design (the same atomic-by-verification staging-clone install/update, manifest discovery and evidence log), scoped to just the URTC ecosystem's own urtc.project.json/ecosystem: "URTC" repositories instead of HYDRA-UMC's. Every URTC repository is public, so remote discovery works with no GITHUB_TOKEN at all. Run for real, end to end, against GitHub in an isolated workspace: a clean install and a later update of a real repository both cloned, built, and promoted correctly, with the previous checkout kept as an atomic backup.
Technology
Python 3.10+PySide6 / Qt Quick (QML)GitGitHub raw content
Build
Detects local URTC projects, compares their manifests with GitHub and performs explicit install or update actions.
|
||||||||||
| URTC Tool Platform 8 project(s) | ||||||||||
| Firmware | Functional | firmware-c | STM32 MCU | 0.5.1 |
OK
Version successfully resolved
|
Take the shared tool set 0.0.8 lock (#88) | ||||
|
Notes
Firmware for the Universal Robot Tool Controller PCB (no populated board exists yet, so it has never run on a real board; its receive path and the bootloader of the master run on a PC against virtual boards) - 25+ tool profiles, CAN bus.
Technology
STM32F303bxCAN25 tool profiles
Build
4 independently-versioned components (main + slave, app + bootloader each). This registry tracks the main application as representative. Real build: bash build_firmware.sh.
|
||||||||||
| Library | Functional | python | User PC | 0.3.1 |
OK
Version successfully resolved
|
Take the shared tool set 0.0.8 lock (#80) | ||||
|
Notes
The CAN protocol of the tool head as data: spec/can_protocol.json holds every identifier with its direction, length rule and payload layout, the magic payloads, the registers of the expansion slave, the version procedures, the SLCAN subset the host tools rely on and the multi-node section. Generators write a C header, a TypeScript module and a Python module (constants plus pure frame builders and parsers), conformance vectors are written and checked, and a Python package offers the SLCAN codec, the transports and the interface helpers the desktop tools share. FLASHER, TESTER and WEB-STUDIO carry checked copies (tools/vendor_copy.py); the tests that compare the specification with the other repositories skip when those are not next to this one.
Technology
JSONPython
Build
python tools/build_test.py
|
||||||||||
| Tool | Functional | python-qtquick | User PC | 0.5.8 |
OK
Version successfully resolved
|
Take the shared tool set 0.0.8 lock (#52) | ||||
|
Notes
GUI flashing tool (CAN-OTA + full-chip SWD/JTAG), exercised only against a simulated bootloader (MockCAN), never against a real board. The Qt Quick deck is the only window: CAN-OTA update of the main board or the expansion slave (with the downgrade authorisation, F-RAM erase and CAN backup of the running firmware), bit-rate selector and listen-only search, GitHub release download, bus-activity check, Board Snapshot, the device-configuration writes, read-only SWD/JTAG discovery and full-chip SWD/JTAG programming. Deadlines use time.monotonic(), not the civil clock, so a clock change mid-flash cannot shorten or lengthen a timeout.
Technology
PythonPySide6Qt QuickCAN-OTASWD/JTAG
Build
Standalone Python desktop tool. pip install -r requirements.txt; the Qt Quick deck starts with no arguments, --cli is the headless update.
|
||||||||||
| Tool | Scaffolding | python | User PC | 0.1.0 |
OK
Version successfully resolved
|
Take the shared tool set 0.0.8 lock (#41) | ||||
|
Notes
A gateway in front of the URTC CAN bus, for an installation that cannot keep the bus closed: only the identifiers a host may send (made from the protocol specification) cross towards the bus, dangerous ones (bootloader, persistent settings) are refused unless unlocked, the payload must be one the specification allows, the rate is limited, and a peer must prove it holds a session key (HMAC with a nonce and a counter, so nothing recorded can be replayed). Every protection can be switched off from the file or the command line and the program says so. It is tested on a PC, including end to end through an SLCAN bus against a virtual board. It has never been run with a real bus, adapter or board.
Technology
PythonCANHMAC
Build
python tools/build_test.py
|
||||||||||
| Firmware | Functional | firmware-c | STM32 MCU | 0.2.9 |
OK
Version successfully resolved
|
Take the shared tool set 0.0.8 lock (#22) | ||||
|
Notes
Hardware-independent pure-C logic: tool-ID decoding, usage and maintenance tracking, the Smart Idle pre-heat target with a safe-state target, a versioned rack command protocol (framing with CRC8, actuation-limit validation, a link watchdog), a multi-slot inventory with per-tool limits, and the layer that holds every slot in the safe state while the link is lost. It is tested on a PC and run against a virtual rack with injected faults (compared with the Python model of the rack). It has never been run on a board. A host-side scenario harness plays valid, corrupted, out-of-range, timed-out and duplicate frames through the whole pipeline, and the ARM cross-compile is verified with a size report. Real CAN/F-RAM/GPIO I/O needs a real PCB, which does not exist yet.
Technology
CSTM32G4arm-none-eabi-gcc
Build
bash build_firmware.sh. Real host-native unit tests run before the arm-none-eabi-gcc cross-compile.
|
||||||||||
| Tool | Functional | python-qtquick | User PC | 0.5.7 |
OK
Version successfully resolved
|
Update ruff requirement (#59) | ||||
|
Notes
Live CAN-bus diagnostic tool including CSV export of live graphs, exercised only against fake transports and never against a real board. The Qt Quick deck is the only interface and starts in listen-only mode. Every output the screens send goes through the encoders in advanced_protocol.py, which reject out-of-range values instead of clamping them.
Technology
PythonPySide6Qt Quicklive CAN bus diagnostics
Build
Standalone Python desktop tool. pip install -r requirements.txt; it starts the Qt Quick diagnostics deck.
|
||||||||||
| Firmware | Functional | firmware-c | STM32 MCU | 0.2.2 |
OK
Version successfully resolved
|
Take the shared tool set 0.0.8 lock (#22) | ||||
|
Notes
Firmware has a hardware-independent sensor protocol (versioned framing with CRC8, range validation, minimum-interval throttling, replay and stale-sensor detection, error and latency counters kept apart from any accept or reject decision), with a host-side sensor simulator covering valid, truncated, CRC-corrupt, out-of-range and disconnected frames - 217 host-native test assertions. It is also run through a virtual sensor link with injected faults, and has never been run on a board or with a real sensor. The Python vision_companion package is test-covered: synthetic thermal and RGB frames, false-colour rendering, and RGB-to-thermal ROI alignment with temperature statistics, and a thermal calibration (gain, offsets, dead pixels) loaded from a JSON file, verified by 251 real pytest cases. Wiring a real MLX9064x/RGB sensor over I2C/SPI, real dual-modal capture and per-pixel calibration still need a real PCB and real cameras.
Technology
CSTM32Python vision companion
Build
bash build_firmware.sh. Also contains an independent Python vision companion package (cd src/vision_companion && pip install -e '.[dev]' && pytest).
|
||||||||||
| UI | Functional | node | User PC | 0.5.3 |
OK
Version successfully resolved
|
Bump the npm_updates group with 6 updates (#50) | ||||
|
Notes
Browser-based alternative to URTC-TESTER via the Web Serial API (the Web Serial path has never been run against a USB-CAN adapter), extensive version history (jumper sync, OLED reset fixes, ...).
Technology
Web Serial APITypeScript
Build
npm install && npm run build.
|
||||||||||
| A.R.M.O.R. - a separate, public perimeter-security and home-automation ecosystem, same author. | ||||||||||
| Clients 5 project(s) | ||||||||||
| Android operator client | Established | Kotlin / Jetpack Compose | Android 8+ | 0.8.0.9 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#56) | ||||
|
Notes
The mobile client authenticates to ARMOR-SERVER with the user-facing Studio session. Camera credentials and RTSP URLs remain server-side; a production signed release still requires a private signing key and HTTPS deployment.
Technology
KotlinJetpack ComposeMJPEGHTTP Studio sessionONVIF PTZ via server
Build
./gradlew.bat test assembleDebug
|
||||||||||
| iPhone operator client | Scaffolding | Swift / SwiftUI | iOS 16+ | 0.0.7 |
OK
Version successfully resolved
|
Regenerate the README and the file tree so the test count matches the tests (#47) | ||||
|
Notes
The iPhone client authenticates to ARMOR-SERVER with the user-facing Studio session, like the Android client. Camera credentials and RTSP URLs remain server-side. The rules of the app are in the ArmorCore Swift package, tested without a phone; the screens are built with Xcode (XcodeGen description in project.yml) and have not been run on a device yet. The project also holds the Apple Watch app of ARMOR-WATCH as an embedded watchOS target (off until the phone says so; built for the simulator in GitHub Actions, never run on a real watch). The app also runs end to end without hardware against a simulated server and a simulated watch (Tests/ArmorCoreTests/Simulation), cuts and failures included; nothing has run on an iPhone or an Apple Watch.
Technology
SwiftSwiftUIFace ID / Touch IDKeychainMJPEGHTTP Studio session
Build
python tools/ios_lint.py && swift test
|
||||||||||
| Touch panel: shows the state of the system on a wall screen, arms, disarms and acknowledges from there, and is where the voice assistant will live | Functional | C++17 (host-tested core); ESP-IDF firmware for the ESP32-S3 with LVGL, with its web page | Waveshare ESP32-S3-Touch-LCD-7C-BOX (the firmware has built once with ESP-IDF 5.5.5 and has never run on the board) | 0.5.0 |
OK
Version successfully resolved
|
Declare the contracts at 0.8.1.6 (#60) | ||||
|
Notes
The settings, what the panel knows of the server (the summary, the frames of its live channel, the state of the link and when what it shows is old), the voice conversation and its binary request, the pins of the board and the words of the screen in eight languages, and the glyphs of its fonts, are tested on a computer (30,783 checks in fourteen test programs, all passing), and the web page is tested in a real browser against a stand-in of the panel. The firmware (the screen, the touch, the sound and the link to ARMOR-SERVER, on top of the node firmware the family shares) has been built once with ESP-IDF 5.5.5 (not by the continuous integration) and has never run on a board: it is written from Waveshare's examples for this family of boards. There is no wake word (it is push to talk) and the speech service the assistant talks to (ARMOR-VOICE-AI, through the server) has not been tried with a real panel.
Technology
C++17ESP-IDFLVGLI2SMQTT
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host
|
||||||||||
| Web administration and calibration console | Established | React 19 / TypeScript / Vite | Browser console served by its own static host or by the nginx of the container image | 0.8.2.6 |
OK
Version successfully resolved
|
Check that the scripts served to the browser parse in the CI validation (#71) | ||||
|
Notes
Operational browser dashboard for the A.R.M.O.R. server API. Studio keeps no password, token or session in the browser; it keeps, per server, the drawings, the camera names and addresses and the preferences.
Technology
ReactTypeScriptRESTCanvas
Build
npm run typecheck && npm test && npm run build
|
||||||||||
| Minimal watch client: alarm notices and alarm confirmation | Scaffolding | Kotlin (JVM core) / Wear OS | Wear OS 3+ watch paired with the phone | 0.0.3 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#32) | ||||
|
Notes
The watch shows the alarms and notices the phone chooses to send and lets the person confirm an alarm with one tap; it keeps nothing on the watch and holds no session of ARMOR-SERVER. Today the repository holds the message contract (docs/watch_alert_contract.json, checked in Python), the Kotlin core that reads and checks every message, keeps the alerts in memory and runs the conversation of the watch end against a simulated phone (tested on a computer with Gradle), the Wear OS app (wear/: a Compose for Wear OS screen and the Data Layer service, built only with -ParmorWear, compiled and unit-tested in GitHub Actions) and the protocol notes. The Apple Watch app (watchos/: a Swift core with the same contract and a SwiftUI app with WatchConnectivity, built only on a Mac, tested and built for the simulator in GitHub Actions) follows the same rules and is also a target of the Xcode project of ARMOR-IOS-CONTROL, embedded in the iPhone app and built there for the simulator in GitHub Actions. The sending side is in ARMOR-ANDROID-CONTROL for Wear OS and in ARMOR-IOS-CONTROL for the Apple Watch (both off by default); an iPhone does not pair with a Wear OS watch and an Android phone does not pair with an Apple Watch. The whole conversation between a phone and a watch also runs in memory (a simulated link with faults, the sending side of the phone apps and a simulated server: Simulation.kt and tools/watch_simulation.py). Nothing has run on a real watch, an emulator or a simulator of the watch system.
Technology
KotlinWear OSData LayerGradle
Build
python tools/watch_contract.py && python tools/watch_simulation.py && python -m unittest discover -s tests && ./gradlew test
|
||||||||||
| Core Services 4 project(s) · ARMOR-SERVER is this family's own integration parent | ||||||||||
| Central event coordinator and web API | Established | Node.js 22+ / TypeScript / Docker Compose | Raspberry Pi CM5 (tested); NVIDIA Jetson Orin NX (planned) | 0.8.2.4 |
OK
Version successfully resolved
|
Declare the contracts at 0.8.1.6 (#95) | ||||
|
Notes
Provides a secured development API, WebSocket state relay and deterministic event reducer. Native HTTPS/WSS (TLS_CERT_PATH/TLS_KEY_PATH, same convention as HYDRA-UMC-SERVER's own) is real and tested against a real self-signed certificate, but has never been run with a real publicly-trusted certificate on real hardware. Broker, camera and field-device adapters remain separately validated deployment work.
Technology
TypeScriptExpressWebSocketMQTTDocker
Build
npm run typecheck && npm test && npm run build
|
||||||||||
| Camera movement observation service (CM5 today, Jetson planned) | Established | Python 3.11+ (standard library) | CM5 test bench (systemd); NVIDIA Jetson Orin NX planned | 0.3.4 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#30) | ||||
|
Notes
Decides and explains; never actuates. Day/night profile selection with hysteresis, an explainable visual/radar fusion policy and a checksum-verified engine registry. It watches the cameras for movement on the CM5 test bench; a person detector and TensorRT need Jetson validation.
Technology
HTTPGrey-frame movement detectionRadar fusion policyTensorRT engine registry (planned)
Build
python -m unittest discover -s tests
|
||||||||||
| Camera zones, masks and detection rules | Functional | Python 3.11+ | NVIDIA Jetson Orin NX (planned); developer workstation today | 0.1.8 |
OK
Version successfully resolved
|
Release 0.1.8 (#26) | ||||
|
Notes
The tested core of the camera analysis service: zones and masks drawn on normalised coordinates, a motion check that decides whether the detector runs at all, a tracker that notices what stands still, and rules for dwell time, labels and arming profiles that produce alert or detection events. No real detector model exists yet (the Jetson Orin NX version is not written) and nothing has run on a camera or on the Jetson; the simulation uses scripted detections.
Technology
PythonMotion gatingObject trackingAlert and detection eventsLibrary without a service
Build
python -m unittest discover -s tests
|
||||||||||
| Offline speech and intent gateway | Established | Python 3.11+ | CM5 test bench (systemd) or any computer | 0.3.7 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#35) | ||||
|
Notes
Closed allow-list of spoken intents in eight languages, service-issued signed single-use confirmation tokens for arm and disarm, and a decision-only audit (no audio, no raw text). The speech engines are programs the installation chooses (ARMOR_VOICE_STT_COMMAND, ARMOR_VOICE_TTS_COMMAND); none is installed with it.
Technology
STTTTSHTTPJSON lines
Build
python -m unittest discover -s tests
|
||||||||||
| Documentation 1 project(s) · ARMOR-DOCS is this family's own integration parent | ||||||||||
| System architecture and operator documentation | Established | Markdown / Mermaid | All users and maintainers | 0.8.2.2 |
OK
Version successfully resolved
|
Release 0.8.2.2 (#67) | ||||
|
Notes
The documentation of the family and the tools that keep it coherent: the READMEs of every repository in eight languages generated from one source, the banners and the map of the family, the version catalogue, the CI tools copied into every repository, the check for what would leak before publishing, and a nightly run that compares all the repositories side by side.
Technology
MarkdownMermaidMQTTVLAN
Build
python tools/make_readmes.py --check && python tools/make_catalog.py --check && python tools/make_brand.py --check
|
||||||||||
| Ecosystem Operations 1 project(s) | ||||||||||
| Detects, installs and updates the A.R.M.O.R. ecosystem's repositories on the machine it runs on | Established | Python 3.10+ (standard library only; PySide6 optional for the desktop shell) | Any machine that installs or updates A.R.M.O.R. repositories (a developer PC or the central server) | 0.2.4 |
OK
Version successfully resolved
|
Refresh the test count in the repository tree of the READMEs (#26) | ||||
|
Notes
Manifest discovery, version comparison, and a staged install and update (a build is verified in an isolated staging clone before being promoted, never in place; the promotion is journalled, the replaced checkout is kept as a backup and can be put back). Every A.R.M.O.R. repository is public, so remote discovery and manifest fetches work with no GITHUB_TOKEN at all (an optional token just raises the 60/hour unauthenticated ceiling), and the manifest reader matches A.R.M.O.R.'s own armor.project.json shape (free-text role/deployment_target, a flat native_version string). Run for real, end to end, against GitHub in an isolated workspace: a clean install and a later update of a real repository (with its own real sibling dependency also installed) both cloned, built, tested and promoted correctly, with the previous checkout kept as a backup.
Technology
PythonPySide6 / Qt Quick (QML)GitGitHub REST API and raw content
Build
python -m pytest tests -q
|
||||||||||
| Field Nodes 5 project(s) | ||||||||||
| Local network monitor: the devices on the house's network, the internet's state and what changes | Established | Python 3.11+ (standard library only) | Any machine on the network being watched (a PC, the CM5, a Raspberry Pi); it reports to ARMOR-SERVER | 0.1.9 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#38) | ||||
|
Notes
The scanner, the inventory and its events, the internet state machine and the message are tested on a computer against a scripted network: 93 tests, all passing, and every message is accepted by ARMOR-COMMON. The program runs as a service on the CM5 test bench, watching the real network the bench is on. It only observes: it sends echoes, TCP connections to a few ports and the multicast questions of mDNS and SSDP to the private network it is on, and a few questions to public addresses to know whether the internet is there. It never configures a router or a device and does not capture packets.
Technology
PythonARP / ICMP / TCP / mDNS / SSDPIEEE OUI registerMQTT-compatible JSON
Build
python -m unittest discover -s tests
|
||||||||||
| Alarm node and panel: the zones, the arming, the delays, the siren and the PIN of the house's alarm, with the server or without it | Functional | C++17 host-tested core; ESP-IDF 5.5.5 firmware for the ESP32-S3 (PoE Ethernet and Wi-Fi boards) | ESP32-S3 nodes (Waveshare ESP32-S3-ETH on PoE, or an ESP32-S3-WROOM-1 on Wi-Fi) on the electrical node's base board; nothing has run on a board yet | 0.5.8 |
OK
Version successfully resolved
|
Declare the contracts at 0.8.1.6 (#53) | ||||
|
Notes
The core (zones, the end-of-line loop with a debouncer, the panel: away and stay arming, exit and entry delays, a latching alarm, a siren that times out, forced bypass, PIN with lockout, a log of events) and, on it, the node firmware: the settings, eight wired zones, the siren on a relay, the web panel in eight languages, the update from GitHub, the Zigbee radio bridge and relays of the base board, and the messages armor/alarm/<node>/state|command|result (the server may arm and disarm only through three independent switches). 486,024 checks on a computer plus the panel in a real browser against a stand-in node; the firmware builds for both boards. Nothing has run on a board or on a real installation, no zone or siren has been wired, the node's own Zigbee conversation with a coordinator (taking the radio over from the server) is written and tested against a stand-in, but nothing has talked to a real coordinator or sensor, and this is not a certified alarm system.
Technology
C++17Zigbee (IAS Zone / IAS WD)end-of-line loopsMQTT
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host
|
||||||||||
| Electrical node: reads the voltages, currents, power and energy of the house's AC and DC network, and holds the rules for switching it | Functional | C++17 (host-tested core); ESP-IDF firmware for the ESP32-S3 (Wi-Fi or Ethernet), with its web panel | ESP32-S3 nodes inside the electrical panels (the firmware builds; it has never run on a board) | 0.6.0 |
OK
Version successfully resolved
|
Declare the contracts at 0.8.1.6 (#52) | ||||
|
Notes
The core (the meters' frames, the bus that asks them, the message, the switching rules) and the node's settings are tested on a computer with stand-in meters and contactors: 388,635 checks across twenty-four host test programs (the interlock and switch-set suites are exhaustive/property-based), all passing. Every state, command and result it makes is accepted by ARMOR-COMMON's shared contract checker. The commands to a switch and the node's answers (`core/switch_set.hpp`) are tested the same way and against ARMOR-COMMON's shared vectors. The firmware builds for both boards. It reads the meters and the analog inputs; its only outputs are the base board's relays, which an administrator switches from the panel, and the Zigbee radio it lends to one address; neither the rules for switching nor the commands are linked into it. It has never run on a board, nothing has been connected to a meter, a contactor, a relay or the radio, and nothing here switches anything by itself.
Technology
C++17PZEM-004T / PZEM-017Modbus RTUMQTT
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host
|
||||||||||
| ESP32-S3 PoE field-node firmware | Established | ESP-IDF C++ | ESP32-S3-ETH-PoE | 0.8.1.4 |
OK
Version successfully resolved
|
Declare the contracts at 0.8.1.6 (#56) | ||||
|
Notes
The core, the node state machine, the sensor decoders and the Wi-Fi board layer are tested on a computer: 2,212 checks across nine host test programs, all passing (no frame captured from a real LD2450 is kept as a fixture yet). Its emitted payloads and device states are all accepted by ARMOR-COMMON's shared contract checker. The firmware runs on a real Waveshare ESP32-S3-ETH board with a radar; the LD2461 and presence-sensor decoders, the pin assignment of the PoE board and the Wi-Fi board have not been verified on hardware.
Technology
ESP-IDFESP32-S3MQTTUARTI2C
Build
idf.py build
|
||||||||||
| Solar gateway node: the firmware that reads inverters and batteries through up to ten serial ports, its web panel and the protocol library | Functional | C++17 (host-tested core); ESP-IDF 5.5.5 firmware for the ESP32-S3-WROOM-1 N16R8 (Wi-Fi) and the Waveshare ESP32-S3-ETH (Ethernet); JavaScript web panel | ESP32-S3 gateway nodes: N16R8 on Wi-Fi, or Waveshare ESP32-S3-ETH on Ethernet | 0.6.8 |
OK
Version successfully resolved
|
Declare the contracts at 0.8.1.6 (#48) | ||||
|
Notes
The firmware builds in the ESP-IDF container and its core, node state machine and Ethernet board layer are tested on a computer with stand-in equipment: 5,412 checks across thirteen host test programs, all passing. Both the polled and the pushed messages it makes are accepted by ARMOR-COMMON's shared contract checker. It has never run on a board and nothing has been connected to a real inverter or battery. The ANT-BMS decoder is tested with frames captured by other people on real units, not by this project.
Technology
C++17Voltronic / MPP SolarPylontechRS232RS485MQTT
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host
|
||||||||||
| Hardware 1 project(s) | ||||||||||
| Open hardware and parametric enclosure designs | Functional | OpenSCAD | Outdoor field nodes | 0.3.0 |
OK
Version successfully resolved
|
Refresh the test count in the README (#20) | ||||
|
Notes
Provides a parametric enclosure concept; RF, ingress and thermal validation require a physical prototype.
Technology
OpenSCADPoEmmWave
Build
OpenSCAD render of every piece: the radar base and cover, and the box and lid of the three node presets
|
||||||||||
| Operations 1 project(s) | ||||||||||
| Repeatable deployment and operations | Established | Docker Compose / Bash / YAML | CM5 test bench (systemd) and any Docker host | 0.7.8 |
OK
Version successfully resolved
|
Set GH_REPO in the draft release job (#36) | ||||
|
Notes
Compose validates the initial service topology; credentials and production network identifiers are externalized.
Technology
Docker ComposesystemdBashCaddyMosquitto
Build
docker compose config
|
||||||||||
| Platform Contracts 1 project(s) · ARMOR-COMMON is this family's own integration parent | ||||||||||
| Shared contracts and validation library | Established | Python 3.11+ / JSON Schema / OpenAPI | All A.R.M.O.R. services and tools | 0.8.1.6 |
OK
Version successfully resolved
|
Move the contract pins of the nineteen consumers to 0.8.1.6 (#75) | ||||
|
Notes
A real, substantial version history under this ecosystem's own odometer scheme (many real releases), and genuinely depended on by every other real A.R.M.O.R. project, not just its own test suite: its conformance vectors are run by ARMOR-SERVER and by ARMOR-NODE-SOLAR's own checks (per ARMOR-DOCS' own capability matrix), and ARMOR-NODE-RADAR/-ELECTRICAL/-NETWORK's own emitted payloads and device states are all independently accepted by this project's own shared contract checker. JSON Schemas are the single source of truth for every message. The Python validators interpret them directly, and TypeScript and Kotlin types are generated from them.
Technology
PythonJSON SchemaMQTTOpenAPI
Build
python -m unittest discover -s tests && python tools/generate_types.py --check
|
||||||||||
| Testing 1 project(s) | ||||||||||
| Offline telemetry and environment simulator | Established | Python 3.11+ | Developer workstation and CI | 0.3.9 |
OK
Version successfully resolved
|
Release 0.3.9 (#23) | ||||
|
Notes
Generates deterministic valid telemetry and health for several scenarios (patrol, crossing, two intruders, empty), the readings of a solar inverter with its battery and of an electrical node, and injects repeatable faults (silent node, out-of-order, duplicates, flapping, and deliberately invalid messages) without needing a broker or hardware. The geometry is illustrative, not a model of the real radar.
Technology
PythonMQTT-compatible JSON
Build
python -m unittest discover -s tests
|
||||||||||
| EH3D Fabrication Suite - the newest ecosystem, for PCB milling, CNC, laser and pick-and-place work, same author. | ||||||||||
| CNC Workflow Tools 21 project(s) | ||||||||||
| UI | Scaffolding | Kotlin / Jetpack Compose | Mobile | 0.1.7 |
OK
Version successfully resolved
|
Read the scheme of an address whatever its case, refuse ports that are not plain digits and addresses with a path, skip | ||||
|
Notes
Scaffolding v0.1.7: the Android client of the EH3D programs, in the Forge theme. It signs in to EH3D-SERVER and follows the Bungard machines of the workshop in real time (state, job, progress, steps done and time left), stops them with an emergency stop that asks twice, confirms the steps a job waits for, shows the job in a 2D and a 3D view with the place of the head, and shows the camera and the sensors of the drill head through the server. It updates itself from the releases of this project on GitHub, checking the package and the signature before Android installs it. The parsers and the geometry are tested on the JVM; the app has not been run on a phone against a real server.
Technology
KotlinJetpack ComposeMJPEGHTTP API of EH3D-SERVERGitHub Releases update
Build
./gradlew.bat test assembleDebug
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.2 |
OK
Version successfully resolved
|
Add behaviour tests of the words, the numbers, the settings and the surface | ||||
|
Notes
Functional v0.1.2: a library and a command line that rewrite a G-code program (absolute millimetres, straight moves) with the height map of the table of EH3D-COMMON: cutting moves are cut in pieces that follow the surface, bilinear or bicubic, and arcs, G91 and G20 are refused with their line. A window (egui, built with the gui feature: open a program and a map, see the table as a coloured mesh and the levelled paths, export) is written and its logic is tested, but it has not been looked at on a screen; the map comes from the probe of the suite or from a simulated table; nothing has been tried on a machine.
Technology
RustG-codeheight mapegui
Build
cargo build --release; cargo test
|
||||||||||
| UI | Functional | rust | User PC | 0.3.0 |
OK
Version successfully resolved
|
Release 0.3.0 | ||||
|
Notes
Functional v0.3.0: the modern interface for the Bungard CCD machines, with the classic drilling and routing programs as the other choices of its launcher. The window opens drilling files, HP-GL files and ZIP archives of jobs (a folder browser, drag and drop, the command line), draws the plan view of the table with every hole by drill size and the routing paths on top of it, the 3D view of the machine fitted to the table size of the configuration (turned, zoomed and moved with the mouse), panelizes the board in three modes, changes and saves the table size, and shows the live progress of the job and the questions of the machine to the operator. It is themed (Forge by default, Neon Arcade, Glass Pro, classic Windows, or the colours of Windows) and in eight languages. The machine is driven by the engines of EH3D-DRILLPRO and EH3D-ROUTEPRO with the same machine file, so the computer's RS-232, a USB-RS232 adapter and the EH3D node, directly or through the EH3D server, work as they do there; it has run on the simulator and on a computer only, never on a real machine. Missing: a pause in the middle of a move, the height of the tool paths in the 3D view, loading and saving the whole configuration and a package with the classic programs.
Technology
Rustegui / eframeSerial port
Build
cargo build --release --features gui; cargo test --features gui
|
||||||||||
| Tool | Functional | rust | User PC | 0.0.8 |
OK
Version successfully resolved
|
Add behaviour tests of the debugger | ||||
|
Notes
Functional v0.0.8: a library and a command line that read a G-code program with the interpreter of EH3D-CORE-LIBS, work out its time with the acceleration of the machine (an upper bound: every move from rest to rest) and look for what would hurt: a rapid move into the material, a move outside the table or below the lowest Z, a cut with the spindle off or with no feed, a feed above a limit and a head that passes over a clamp. A window (egui, built with the gui feature: open a program, see its path from above, go through the moves step by step, read the findings) is written and its logic is tested, but it has not been looked at on a screen; no 3D simulation and no removal of material yet; nothing has been tried on a machine.
Technology
RustG-codeegui
Build
cargo build --release; cargo test
|
||||||||||
| Tool | Functional | rust | User PC | 0.2.1 |
OK
Version successfully resolved
|
Add behaviour tests of the protocol, the file limits, the languages and the progress, and fix the Greek keyboard layout | ||||
|
Notes
Functional v0.2.1: a re-implementation of the drilling workflow of the classic DOS software for Bungard CCD machines. It reads Excellon .NCD drill files, the tool table and the machine configuration, orders the holes and runs the job against a simulator, in a text-mode look-alike interface in 37 languages. The serial protocol was worked out by watching how the original program talks to its control unit and has been tried on one Bungard CCD MTC: the reference run of the three axes and the moves of every axis. A whole job has not run on a machine yet, and the spindle, the vacuum output and the steps per millimetre are not confirmed.
Technology
Rustegui / eframeExcellonSerial port
Build
cargo build --release --features gui; cargo test --workspace
|
||||||||||
| Tool | Functional | rust | User PC | 0.0.9 |
OK
Version successfully resolved
|
Add behaviour tests of the view, the file list, the regions, the settings, the layer stack and the ruler | ||||
|
Notes
Functional v0.0.9: a library and a command line that turn Gerber layers and Excellon drill files (read with the readers of EH3D-CORE-LIBS) into shapes in millimetres, draw them as one SVG picture with a colour and a transparency per layer, measure the narrowest track and the smallest gap between round pads and tracks, and check that the layers line up and that the holes are inside the copper. A window (egui, built with the gui feature: add or drop layers and drill files, show or hide each, zoom and pan, measure, check the alignment, export the picture) is written and its logic is tested, but it has not been looked at on a screen; clear polarity, step and repeat and macro apertures are noted, not drawn; nothing has been tried on real production files.
Technology
RustGerber RS-274XExcellonSVGegui
Build
cargo build --release; cargo test
|
||||||||||
| UI | Scaffolding | Swift / SwiftUI | Mobile | 0.1.6 |
OK
Version successfully resolved
|
Count the lines of a program with Windows line ends, read addresses with line breaks and plain-digit ports as the Androi | ||||
|
Notes
Scaffolding v0.1.6: the iPhone client of the EH3D programs, the twin of EH3D-ANDROID-CONTROL, in the Forge theme. It signs in to EH3D-SERVER and follows the Bungard machines of the workshop in real time, stops them with an emergency stop that asks twice, confirms the steps a job waits for, shows the job in a 2D and a 3D view with the place of the head, and shows the camera and the sensors of the drill head through the server. The rules are in the Swift package Eh3dCore (the readers of the server's JSON, the geometry, the address checks, the session, the languages) and are tested without a phone; the screens are built with Xcode and have not been compiled or run on a device.
Technology
SwiftSwiftUIKeychainMJPEGHTTP API of EH3D-SERVER
Build
swift test
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.6 |
OK
Version successfully resolved
|
Refuse dot names and text-reversing characters in artifact names and add edge tests | ||||
|
Notes
Scaffolding: a library and a command line that read the copper of a Gerber layer (with the reader of EH3D-CORE-LIBS and the shapes of EH3D-GERBERVIEWER), draw it on a fine grid, work out the exact distance from every point to the copper and trace the paths of one or several tools at their radius (plus a stepover for each further pass) as closed or open polylines, clear the leftover copper, choose the direction, the starting points and the order of the cuts, report where two pieces of copper are closer than the tool is wide and write the paths as G-code or HP-GL. It also reads and writes drill files and board outlines (with the Bungard formats of EH3D-COMMON), lays a board out for both sides and for a panel with alignment holes checks the design rules, can lay a path along gaps the tool does not fit in, scale and correct the work from reference points, write the openings of the solder mask and draw a preview of the passes; every tuned value is a setting. A window (egui, built with the gui feature: open or drop a layer, drills, an outline, a mask layer and reference points, turn the board over and repeat it on a panel, zoom and pan, edit the main parameters, export) is written and its logic is tested, but it has not been looked at on a screen; no Voronoi strategy and nothing has been milled.
Technology
RustGerber RS-274XExcellonG-codeHP-GLdistance transformegui
Build
cargo build --release; cargo test
|
||||||||||
| node | Functional | C++17 (host-tested core); ESP-IDF firmware for the ESP32-S3 (Ethernet with PoE, or Wi-Fi) | STM32 MCU | 0.0.8 |
OK
Version successfully resolved
|
Test the language choice of the panel and the codes of the node against hostile, capitalised and former-variant codes | ||||
|
Notes
Functional v0.0.8: a thin network pipe to the RS-232 control unit of the Bungard CCD MTC and CCD/2 MTC drilling and routing machines (serial-port control units of the DOS software 4.6). A program on the network sends the bytes of a move over a framed TCP link; the node sends them out of its UART at the line's own pace, so the timing no longer depends on a PC or a USB adapter. The rules that keep the machine from moving by accident (login with a token, arming, a stop when the link goes quiet or drops) are in the host-tested core. The firmware has not been built or flashed yet and nothing has been connected to a machine; the protocol itself was tried on a CCD MTC through a USB adapter.
Technology
C++17ESP-IDFRS-232TCP
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host -C Debug
|
||||||||||
| node | Scaffolding | C++17 (host-tested core); ESP-IDF 5.5 firmware for the ESP32-S3 with a W5500 wired network port | STM32 MCU | 0.1.0 |
OK
Version successfully resolved
|
Test the language choice of the panel and the codes of the node against hostile, capitalised and former-variant codes | ||||
|
Notes
Scaffolding v0.1.0: a node for the drill head of a Bungard CCD machine that measures the height of the table with a laser, reads the speed and the temperature of the spindle, opens a protect output when the spindle is too fast or too hot and shows the camera, all over a wired network behind a token. The rules are in the host-tested core: the settings file with safe defaults, the measure from several laser readings (the odd ones thrown away, a spread that is too large refused), the speed of the spindle from pulses, the temperature with a hysteresis, the protect output and the words of the network. The ESP-IDF firmware is written but has not been built, flashed or run on a board; the drivers of the laser and of the temperature sensors are not written and nothing has been connected to a machine.
Technology
C++17ESP-IDFESP32-S3W5500
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host -C Debug
|
||||||||||
| UI | Scaffolding | C++17 (host-tested core); ESP-IDF 5.5 firmware with LVGL 9 for the ESP32-P4 | STM32 MCU | 0.1.2 |
OK
Version successfully resolved
|
Take the shared tests of hostile language codes from the firmware base | ||||
|
Notes
Scaffolding v0.1.2: a 10.1-inch touch screen (Guition JC8012P4A1C, ESP32-P4) meant to control a Bungard CCD machine on its own, with the screens of the desktop programs in eight languages and spoken alerts and orders. The rules are in the host-tested core: the settings file with safe defaults, the gate that every move and spindle order passes (confirmation on the screen, hardware-stop latch, armed link), the spoken orders and alerts and the screen texts in eight languages. The ESP-IDF firmware with the LVGL screens is written but has not been built, flashed or run on a module; the link to the server is written and tested against a simulated server, but nothing has run against a real server or been connected to a machine. It draws the settings of EH3D-SERVER from the contract of the server (a copy travels in the screen and is shown read only when the live one cannot be read; nothing of the server is changed unless hmi.cfg says so), tested on a computer against a simulated server.
Technology
C++17ESP-IDFLVGLESP32-P4
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host -C Debug
|
||||||||||
| UI | Scaffolding | C++17 (host-tested core); ESP-IDF 5.5 firmware with LVGL 9 for the ESP32-P4 | STM32 MCU | 0.1.2 |
OK
Version successfully resolved
|
Take the shared tests of hostile language codes from the firmware base | ||||
|
Notes
Scaffolding v0.1.2: a 10.1-inch touch screen (Guition JC8012P4A1C, ESP32-P4) meant to control a CNC machine, a laser or a pick and place machine on its own, on the hardware and the firmware base of EH3D-NODE-BUNGARD-HMI, with the screens in eight languages, a visual check of the board (CNC only) and spoken alerts and orders. The rules are in the host-tested core: the settings file with safe defaults and the kind of machine (only a CNC machine gets a spindle button; the screen never fires a laser or starts a pick and place head), the gate that every move and spindle order passes (confirmation on the screen, hardware-stop latch, armed link), the spoken orders and alerts and the screen texts in eight languages. The ESP-IDF firmware with the LVGL screens has not been flashed or run on a module and nothing has been connected to a machine. It draws the settings of EH3D-SERVER from the contract of the server (a copy travels in the screen and is shown read only when the live one cannot be read; nothing of the server is changed unless hmi.cfg says so), tested on a computer against a simulated server.
Technology
C++17ESP-IDFLVGLESP32-P4
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host -C Debug
|
||||||||||
| UI | Functional | rust | User PC | 0.2.3 |
OK
Version successfully resolved
|
Edge-case tests of the program readers and a fix for a semicolon inside a comment (#19) | ||||
|
Notes
EH3D-PHOENIX-CNC is the machine-control program of the PHOENIX family (violet phoenix), formerly EH3D-MOTIONCNC. Scaffolding v0.0.9: the EH3D machine-control program. A window (phoenix-cnc, feature gui) over the controller interface of eh3d-controller in EH3D-COMMON (a sibling folder is needed), working on a simulated machine only: position, jog, zero, home, a line of commands, a G-code program run line by line with pause, stop and the emergency stop, and its path drawn on the table; five themes and eight languages. Nothing has run on a real machine; by default only a simulated machine is driven, a serial port is opened only with --grbl, --grblhal or --marlin and a network connection only when a server address is set or --klipper names one. GRBL, grblHAL, Marlin and Klipper have been tried against simulators of their dialects only, and the Gerber and Excellon files of a board are read with EH3D-CORE-LIBS and cut as isolation, profile and drill programs.
Technology
Rust
Build
cargo build --release --features gui; cargo test --features gui
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.6 |
OK
Version successfully resolved
|
Refuse negative power and speed and numbers that are not numbers before a vector job leaves, and add edge tests | ||||
|
Notes
EH3D-PHOENIX-LASER is the laser engraving program of the PHOENIX family (bright red phoenix), formerly EH3D-LASER-CAM. Scaffolding v0.0.3: a library and a command line that read a PNG or PGM image, turn it into dots (threshold, Floyd-Steinberg or Atkinson dithering) and write the G-code of a laser that burns line by line with M4 dynamic power, S0 while it travels, an overscan worked out from the acceleration and alternating directions, plus a power-against-feed calibration grid. It also reads SVG and DXF drawings and writes the G-code that cuts their contours (inside ones first, the width of the beam taken into account, tabs, several passes) and engraves their shapes with lines, with the settings of each layer in a file and the program checked line by line before it is given. A window (egui, built with the gui feature: open a picture or a drawing, see the passes of the laser, edit the table of layers, export) is written and its logic is tested, but it has not been looked at on a screen; the window is the docking frame of the suite (twelve panels, four workspaces for tasks, a menu bar, the stop and the interlocks always in sight, the arrangement kept in the settings file). It also reads LBRN2 project files, measured on twelve real projects of one laser maker (all read; the one picture among them is decoded and engraved by dots, and the direction of their y axis was never confirmed with a real part), and, with the feature common-machining (on by default), the starting points of the common database of materials as they are (never verified). Nothing has been tried on a laser.
Technology
RustG-codePNGditheringegui
Build
cargo build --release; cargo test
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.4 |
OK
Version successfully resolved
|
Keep server numbers from wrapping round to another slot, refuse positions that are not numbers, write server texts safel | ||||
|
Notes
EH3D-PHOENIX-PNP is the pick and place program of the PHOENIX family (green phoenix), formerly EH3D-PNP-MANAGER. A library, a command line and a window (feature gui) for pick and place assembly: they read the position file of a board (KiCad .pos or a CSV from another CAD program, in mm, mil or inches), give every part the feeder that holds its value and package, plan the order of the placements for a head with one to six nozzles (ranks, nozzle tips, angle limits, the time of the moves, panels with fiducials of two levels and boards marked as bad), write a load sheet and carry the job out on a simulated machine with a simulated camera and vision, with a work report as text, HTML, CSV and JSON. The window is made of the panels of the docking frame of the suite (they dock, float and fold, in three arrangements, with a menu bar and a panel for eleven commands), a map of the work table, problems with fixes and a Job panel with decisions when an error stops the job; its logic is tested without a screen. No real camera and no machine control; nothing has placed a part.
Technology
RustKiCadCSVJSON
Build
cargo build --release; cargo test
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.8 |
OK
Version successfully resolved
|
Add behaviour tests of the routing job | ||||
|
Notes
Functional v0.1.8: a re-implementation of the routing (milling) workflow of the classic DOS software for Bungard CCD machines. It reads HP-GL .PLT plot files, the tool table and the machine configuration, plans multi-pass milling paths (rapid move, plunge, cut, lift) and runs them against a simulator, in a text-mode look-alike interface in 37 languages. It reuses the serial protocol, the screen and the language support of EH3D-DRILLPRO (a sibling folder is needed). The serial protocol was worked out by watching how the original program talks to its control unit and has been tried on one Bungard CCD MTC: the reference run of the three axes and the moves of every axis. A whole job has not run on a machine yet, and the spindle, the vacuum output and the steps per millimetre are not confirmed.
Technology
Rustegui / eframeHP-GLSerial port
Build
cargo build --release --features gui; cargo test --workspace
|
||||||||||
| API | Functional | rust | CM5 | 0.2.2 |
OK
Version successfully resolved
|
Edge-case tests of the pure parts of the server and the bugs they found (#30) | ||||
|
Notes
Functional v0.2.2: the center of the EH3D programs. EH3D-DRILLPRO, EH3D-ROUTEPRO and, later, the Android app, the Studio and the panel next to the machine send it jobs (holes and paths, or an Excellon or HP-GL file with its tool table); it plans them with the libraries of the programs, using the constants and the tuning of the machine they are for, shows a preview, and runs them on the machine through its node (or a simulator), with the dialogs of the operator, a stop, and every change told over a websocket. Users with three roles, tokens for programs, one job at a time on a machine, and a job never starts by itself. Tested on a computer against the simulator; nothing has run on a real machine through it, and it does not terminate TLS yet.
Technology
RustaxumRESTWebSocket
Build
cargo build --release; cargo test
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.2 |
OK
Version successfully resolved
|
Leave empty paths out of the order of the cuts, refuse dot names in artifact names, and add edge tests | ||||
|
Notes
Functional v0.1.2: a library and a command line that turn the pads of a paste layer (read with the Gerber reader of EH3D-CORE-LIBS and turned into shapes by EH3D-GERBERVIEWER) into the openings of a solder paste stencil: each is shrunk by a percentage for the material, rounded, reduced by the kerf of the laser, checked against the aspect and area ratios that decide whether the paste leaves the opening, and written as G-code for a laser or HP-GL for a plotter, mirrored for the bottom side if asked. Macro pads, tracks and regions are not turned into openings and a polygon pad is cut as a circle; nothing has been cut. A window (egui, built with the gui feature: open or drop a layer, choose the material, edit the parameters, see the pads and the cutting paths, export) is written and its logic is tested, but it has not been looked at on a screen.
Technology
RustGerber RS-274XG-codeHP-GLegui
Build
cargo build --release; cargo test
|
||||||||||
| UI | Functional | React 19, TypeScript and Vite; served by EH3D-SERVER | CM5 | 0.1.9 |
OK
Version successfully resolved
|
Keep a damaged address from stopping the page, refuse links with a name before the host and hexadecimal numbers, and add | ||||
|
Notes
Scaffolding v0.1.9: the configuration panel of EH3D-SERVER, in the look of EH3D-BUNGARD-GUI (the Forge theme) and organised in three groups: Monitor (overview, machines by station, jobs, live events), Configure (machines with their node and their configuration files, users and tokens, the settings of the server) and Account (password, language, theme). Eight languages. It talks only to the API of the server and the server can serve its pages at the root of its own address. The client of the API, the rules of the forms and the eight languages are tested; the pages have been looked at in a browser against a local server with simulated machines, and not against machines on a bench.
Technology
ReactTypeScriptViteRESTWebSocket
Build
npm ci && npm run typecheck && npm run lint && npm test && npm run build
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.2 |
OK
Version successfully resolved
|
Add behaviour tests of the cut, the crib, the check, the tables and the settings | ||||
|
Notes
Functional v0.1.2: a library, a command line and, built with the gui feature, a window (egui) for the tool crib: drills, end mills, V-bits, engravers, lasers and nozzles with their geometry, wear counters (hours, metres, holes) with limits and warnings, JSON and CSV storage, and starting feeds and speeds by material limited by the spindle, with the numbers of the calculation in tools.cfg. The window (a table of the tools with a form, filters, CSV import and export, and a calculator that shows its formulas and conditions) is written and its logic is tested, but it has not been built or looked at on a screen. No database (the crib is a JSON file) and the numbers are starting points that have not been tried on a machine.
Technology
Rustserde_jsonCSVegui
Build
cargo build --release; cargo test
|
||||||||||
| Tool | Functional | rust | User PC | 0.1.3 |
OK
Version successfully resolved
|
Add edge-case tests of what reads text and numbers and fix the bugs they found | ||||
|
Notes
Functional v0.1.3: a library and a command line that find round fiducial marks in a PNG or PGM image (Otsu threshold, connected spots, size, fill and elongation checks, marks cut by the edge refused), turn pixels into millimetres with the scale of a mark of known size, fit and correct the radial distortion of a lens from a grid of marks, work out the turn, the shift and (if allowed) the size that take a board from its design to where the camera found it, and, as plain CSV or JSON files, find cross-shaped marks, take uneven light out, refine a centre to a fraction of a pixel, work the scale out from the pitch of a target, measure parts and report on a folder of pictures. With the gui feature there is also a window (eh3d-vision-gui, egui) that opens a picture, shows the marks, sets the scale, fits the lens from a grid, aligns with a design and exports the options of ISOCAM; it has not been looked at on a screen. No live camera, no neural network and no Hailo acceleration yet; nothing has looked at a real board.
Technology
RustPNGcomputer vision
Build
cargo build --release; cargo build --release --features gui; cargo test
|
||||||||||
| Documentation 1 project(s) · EH3D-DOCS is this family's own integration parent | ||||||||||
| Documentation | Functional | docs | User PC | 0.1.2 |
OK
Version successfully resolved
|
Register the container codes EH3D-PKG-015 to EH3D-PKG-024 of the project container with branches | ||||
|
Notes
Scaffolding v0.1.2: the hub of the documentation of the EH3D Fabrication Suite: the map of the suite (what each repository is for and whether it exists yet), the architecture rules, and a capability matrix that says what has been run and what has not. It documents the repositories of the suite that are published and lists as plans the ones that are not.
Technology
MarkdownPython 3.10+
Build
python -I tools/ci_validate.py; python -I tools/check_links.py
|
||||||||||
| Ecosystem Operations 3 project(s) | ||||||||||
| Service | Scaffolding | rust | User PC | 0.0.3 |
OK
Version successfully resolved
|
Add 65 tests for the shared tools of the base | ||||
|
Notes
Scaffolding v0.0.3: PLANNED, no code yet. Adapter between the EH3D and HYDRA-UMC ecosystems. It talks to the EH3D server and is the counterpart of the bridge adapters of HYDRA-UMC. What exists is the base of every repository of the suite and the written scope, requirements, dependencies, threat model and decisions in docs/. The planned stack is Rust (a library with a command-line interface); nothing has been run on a machine.
Technology
Rust (planned)MarkdownPython 3.10+
Build
python -I tools/ci_validate.py
|
||||||||||
| Tool | Functional | rust | User PC | 0.0.7 |
OK
Version successfully resolved
|
Release 0.0.7 | ||||
|
Notes
Functional v0.0.7: a library and a command line that find the programs of the suite by their eh3d.project.json manifests, start a program and start it again a limited number of times when it fails, keep the list of recent projects and read and write the project file (.eh3dproj: name, files with kind and SHA-256, work zero). A window (egui, built with the gui feature: the recent projects, open or create a project, add and remove its files, the programs of the suite with a button to start each one) is written and its logic is tested, but it has not been built or looked at on a screen. No discovery of machines on the network, and nothing here has run on a machine.
Technology
Rustserde_jsonSHA-256egui
Build
cargo build --release; cargo test
|
||||||||||
| Tool | Functional | python | User PC | 0.1.1 |
OK
Version successfully resolved
|
Read versions as plain digits, check repository names and relative paths of a manifest, refuse repeated alternatives in | ||||
|
Notes
Scaffolding v0.1.1: the updater of the EH3D ecosystem, with the design of the updaters of the other ecosystems. It finds the EH3D repositories through their eh3d.project.json manifests, compares each local checkout with GitHub and installs or updates one at a time after a confirmation: the replacement is built and checked in a staging clone first and the previous installation is kept as a backup. Tested on a computer with a fake GitHub and local repositories; it has not been used for real work yet.
Technology
Python 3.10+PySide6 / Qt Quick (QML)GitGitHub raw content
Build
python -m pytest tests -q; python -I tools/ci_validate.py
|
||||||||||
| PHOENIX 12 project(s) | ||||||||||
| Library | Scaffolding | rust | User PC | 0.0.4 |
OK
Version successfully resolved
|
Add behaviour tests of the library and the command line | ||||
|
Notes
Scaffolding v0.0.4: early code. Technical database of electronic components in which every value carries its provenance: a Rust library and command line that read a JSON database, find duplicate identities, observations without a source, sources that disagree, records marked verified while something is pending and records without a recorded origin or licence, and refuse to export data without a licence. It also imports a CSV list of parts into a JSON database file kept next to the project, with bounded reading and stable refusal codes. It has only been tested with data made by its own tests and small example files; other import formats and the link with the symbols and footprints of the design tool are not written.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Documentation | Scaffolding | rust | User PC | 0.0.4 |
OK
Version successfully resolved
|
Add tests for the shared tools and the comparison of alternatives | ||||
|
Notes
Scaffolding v0.0.4: early code. A Rust library and command line that read a set of alternatives of a design (parameters and figures that each name their tool and version) with its criteria, check it, show the place of every alternative (non-dominated, or dominated and by whom) without naming a single best one, list what differs between two alternatives and generate the grid of a parametric sweep in a fixed order. It has only run on data made by its own tests; it computes no figure, applies no change to a design and starts no job, and the comparison of branches of the design history is not written.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Library | Scaffolding | rust | User PC | 0.0.4 |
OK
Version successfully resolved
|
Add tests for the shared tools and the history | ||||
|
Notes
Scaffolding v0.0.4: early code. Semantic history of a design: a Rust library and command line that read a history of operations keyed by the stable identity of an object and chained by SHA-256, find a change in the past, a state that points nowhere and a reference to a deleted object, print the log, compare two histories from where they separate and find the first bad operation by halving. It has only run on data made by its own tests; it reads no design, the producer of the operations belongs to the design tool, and the comparison by the content of the design is not written.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Documentation | Scaffolding | rust | User PC | 0.0.4 |
OK
Version successfully resolved
|
Add tests for the shared tools and the report | ||||
|
Notes
Scaffolding v0.0.4: early code. Integral verification of a design: a Rust library and command line that gather the results of the checks from a JSON file, check that every finding has a well-formed PHOENIX code and names its object, and give a verdict that keeps confirmed failures apart from inconclusive results (a required area that did not run makes the verdict inconclusive). The report is canonical, with a hash, and can be printed redacted with counts only; the review document writes the verdict, the rules applied, the data used and the findings as text. It has only run on data made by its own tests; it runs no check itself, does not read a design, and the comparison with a manufacturing profile is not written.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Library | Scaffolding | rust | User PC | 0.0.4 |
OK
Version successfully resolved
|
Add tests for the shared tools, the curves and the records | ||||
|
Notes
Scaffolding v0.0.4: early code. A Rust library and command line that read a measured curve from one defined CSV format with units in the header, check a test record (instrument, conditions, quantity, SHA-256 of the data) and compare the measured curve with a simulated one, giving the deviation of every point with its expanded uncertainty and a verdict only against a tolerance that the person sets. It has only run on curves made by its own tests; the files of real instruments, instrument control and the link with the technical passport are not written, and no tolerance exists by default.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Library | Scaffolding | rust | User PC | 0.0.4 |
OK
Version successfully resolved
|
Add tests for the shared tools, the profile and the settings | ||||
|
Notes
Scaffolding v0.0.4: early code. Profiles of the capabilities of workshop and factory machinery and of the restrictions of suppliers: a Rust library and command line that read a JSON profile whose limits always carry their unit (min_/max_ plus a quantity that ends in a unit), find numbers without a unit, bad limits, conflicting tool definitions and restrictions without a source, cite a profile by the SHA-256 of its content, and compare the measurements of a design with the limits, naming the limit, its source and the remedy. It has only run on data made by its own tests; the reading of the machine files that the server and the tool manager use, and the link with the verifier, are not written.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Tool | Functional | rust | User PC | 0.0.8 |
OK
Version successfully resolved
|
Add edge-case tests of the formats, the profiles and the slicer and fix the bugs they found | ||||
|
Notes
EH3D-PHOENIX-3DP is the 3D printing program of the PHOENIX family. Scaffolding v0.0.1: a Cargo workspace of eight libraries (none uses a crate from crates.io; the server bridge depends by path on the common client of the suite in EH3D-COMMON, a sibling folder is needed) and a command line (printers, info, check-profile, convert, filaments, slice, project, version) that read STL, OBJ and basic 3MF models with size limits and ordered refusals, read the objects, plates and a few settings of the 3MF projects that other programs save (read only), read the headers of binary G-code and of open resin layer packages (with an estimate of the exposure; the closed resin formats are refused), write binary STL, OBJ, an own 3MF package and G-code files, hold a catalog of 120 printer profiles (FDM and resin, built from public data, every one verified = false, each saying whether its temperature limits are the maker's or the conservative ones of the program) and the 477 filaments and seven resins of the common database of printing materials checked against a printer and a nozzle, slice a closed model into flat layers with perimeters, rectilinear, grid or gyroid infill, solid skins, bridges, plain supports, seams and variable width, estimate the time with acceleration, and write G-code for the Marlin, Klipper, RepRapFirmware and Bambu dialects with temperature limits that refuse a value instead of cutting it down, check any program against those limits, drive Moonraker and OctoPrint printers through one normalized state and one list of commands over a plain HTTP transport, and keep the settings and a client of the printer3d plugin routes of the EH3D server. Everything has been tried only against simulators written for the tests and a socket of the same computer: no real printer, server or network was used. Bambu LAN is a design marked experimental that does not connect. The slicer lays a skirt, a brim or a raft and has no tree supports, no multi-material and no print from it has run on a printer. The model of a 3D viewer (camera, picking, range of layers, lines of a plan) is the crate p3dp-viewer; its wgpu drawing is behind a feature and has never been built. A window in the docking frame of the suite (panels that move, float and fold, named workspaces for the four old tabs, a menu bar) is written for the gui feature, with filtered lists, a check before printing, the plan by layers, the queue of the server with a confirmed start, shortcuts and a picture mode; its logic is tested, but it has not been built or looked at on a screen.
Technology
RustG-code3MFMoonrakeregui
Build
cargo build --release; cargo test --workspace
|
||||||||||
| Documentation | Scaffolding | rust | User PC | 0.1.3 |
OK
Version successfully resolved
|
Add grow, shrink and shell to the history, sections, shadows and inertia of solids, five mesh formats, the entry line fo | ||||
|
Notes
Scaffolding v0.1.3: early code. Mechanical design for the PHOENIX family: parametric enclosures, chassis and heat sinks that live in the same project as the board. What exists is the base of every repository of the suite, the written scope, requirements, dependencies, threat model and decisions in docs/, and a Rust library with a parametric enclosure (lengths with their own range, round holes, change operations that report conflicts, an OpenSCAD-style script and a fit check, and a 2D sketch with constraints and a solver that is kept in the project file and whose closed outlines can be raised from a face or cut into it). It has only been tried on models made by its own tests; it has a command line and a window (never opened on a screen), a STEP file checked only for its structure that does not draw the cooling fins or the parts, patterns of holes and posts, sets of cooling fins with a simple estimate of their heat, a catalog of generic parts (profiles, screws and brackets) placed in the model with a bill of materials, a cut list and the mass of each material, simple assemblies of those parts (mates along one axis and a check of which parts share volume), an STL file of the solid and DXF and SVG drawings of the three views that no slicer, CAD program or drawing program has opened, and nothing has been run on a machine.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test; python -I tools/ci_validate.py
|
||||||||||
| Documentation | Scaffolding | rust | User PC | 0.1.6 |
OK
Version successfully resolved
|
Release 0.1.6 | ||||
|
Notes
Scaffolding v0.1.6: early code. EDA environment planned as schematic capture, PCB layout, verification tools, circuit simulation and specialized modules for switch-mode power supplies and RF; it is the main project of the PHOENIX family. What exists is a Rust workspace: a redacted inventory of a private corpus, the design model foundations, the .phoenixproj project file, the board model with silkscreen and reversible operations, connectivity and rule checks, the schematic model with a drawing, a reader of EAGLE XML with a measured read rate, a writer and reader of boards as EAGLE XML, the contract for pairs of EAGLE binary files, the Gerber and Excellon export, a viewer and editor prototype whose window has never been seen on a screen, and a command line. All of it has been tried on generated or hand-made data only; the editors, the readers of foreign designs and the reader of the EAGLE binary files are not written and nothing has been run on a machine.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Documentation | Scaffolding | rust | User PC | 0.0.4 |
OK
Version successfully resolved
|
Add tests for the shared tools and the passport | ||||
|
Notes
Scaffolding v0.0.4: early code. Technical passport of a project: a Rust library and command line that read a passport (a JSON file whose sections are always written) with the tools and versions, the input and output files with their SHA-256, the results of the checks, the limitations and the pending decisions, find bad hashes, unsafe paths, duplicates and tools without a version, print the hash of the canonical content, compare the passport with the files of a folder (only read) and compare two passports. It has only run on data made by its own tests; the generator from a project, the rebuilding of the manufacturing files and the signature are not written.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Documentation | Scaffolding | rust | User PC | 0.0.3 |
OK
Version successfully resolved
|
Add tests for the shared tools and the thermal code | ||||
|
Notes
Scaffolding v0.0.3: early code. A Rust library and command line that compute the temperature rise of a copper trace with a current (Joule effect with temperature-dependent resistance, cooling by convection) and across a series and parallel network of thermal resistances, and refuse a result that lacks hypotheses, boundary conditions, materials, uncertainty, limitations or range of validity; a case out of range or without convergence gives no number. It has only run on cases made by its own tests and closed-form values; nothing was compared with a solver or a measurement, and the heat sources, the geometry coupling and the comparison with measurements are not written.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Documentation | Scaffolding | rust | User PC | 0.0.3 |
OK
Version successfully resolved
|
Add behaviour tests of the laboratory | ||||
|
Notes
Scaffolding v0.0.3: early code. Test environment of the PHOENIX family: a Rust library and command line that write and verify the checksum manifest of a golden set against redundant copies, read results in which every failure carries a class, and compare a run with a baseline through a gate that a measured rate never drops. Reports hide the names of cases. It has only run on data made by its own tests; it reads no design file and the comparison with golden files, the case reducer and the invariant tests are planned.
Technology
RustMarkdownPython 3.10+
Build
cargo build --release; cargo test
|
||||||||||
| Pick and Place Nodes 2 project(s) | ||||||||||
| node | Functional | C++17 (host-tested core); ESP-IDF firmware for the ESP32-S3 (not written yet) | STM32 MCU | 0.0.2 |
OK
Version successfully resolved
|
Show the maturity of the project as functional in the README and its translations | ||||
|
Notes
Functional v0.0.2: the node of the tape feeders of a pick-and-place machine. It advances the tape of each feeder by a pitch of 2, 4, 8, 12 or 16 mm over RS-485, CAN or I2C (a setting) with its own checked frame, counts the parts, watches the pocket, the tension and the cover film, holds a fault when the tape jams and stops by itself when it loses the node or the server. The rules are in the host-tested core and are tried against a simulator of the chain; the ESP-IDF firmware is not written, and nothing has been connected to a motor, a tape sensor or a bus.
Technology
C++17ESP-IDFESP32-S3RS-485CANI2C
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host -C Debug
|
||||||||||
| node | Functional | C++17 (host-tested core); ESP-IDF firmware for the ESP32-S3 (not written yet) | STM32 MCU | 0.0.2 |
OK
Version successfully resolved
|
Show the maturity of the project as functional in the README and its translations | ||||
|
Notes
Functional v0.0.2: the node of the head of a pick-and-place machine. It runs the vacuum pump and the valve of each nozzle with a pressure sensor or a vacuum switch, the puff of air that lets a part go, the Z axis and the rotation of each nozzle and two lights, under interlocks, tells the machine when every nozzle is raised so that it may move in X and Y, and goes to a safe state by itself when it loses the server. The rules are in the host-tested core and are tried against a simulator of the head; the ESP-IDF firmware is not written, and nothing has been connected to a pump, a valve, a sensor or a motor.
Technology
C++17ESP-IDFESP32-S3Vacuum
Build
cmake -S tests -B build/host && cmake --build build/host && ctest --test-dir build/host -C Debug
|
||||||||||
| Platform Foundation 2 project(s) | ||||||||||
| Library | Functional | rust | User PC | 0.3.0 |
OK
Version successfully resolved
|
Test that hostile and capitalised language codes are refused and leave the language as it was | ||||
|
Notes
Functional v0.3.0: the .eh3dproj project container (a ZIP archive with project.json, integrity.json and layers/, toolpaths/, vision/), its metadata with format versions, SHA-256 checks and limits against unsafe entry names and oversized entries, the machine telemetry types (axes, spindle, states, GRBL alarm codes) and the colours of the suite. No program of the suite uses it yet and nothing here has been run on a machine. It also keeps the firmware base of the ESP32 nodes (firmware_base/ and tools/sync_firmware_base.py): the part of the node firmware that every EH3D node shares, written once and copied into each node.
Technology
Rustserdezipsha2
Build
cargo build --release; cargo test
|
||||||||||
| Library | Functional | rust | User PC | 0.1.2 |
OK
Version successfully resolved
|
Refuse coordinate formats with absurd digits, tell words from numbers that are too large, keep lengths from overflowing, | ||||
|
Notes
Functional v0.1.2: tolerant readers of Gerber RS-274X layers, Excellon drill files and G-code (a modal interpreter for the GRBL, Marlin and PlanetCNC dialects) that never panic and report what they could not understand as diagnostics with line numbers. They draw nothing, compute no toolpath and have not been tried on a collection of real production files; HP-GL and 2D geometry are planned. Nothing here has been run on a machine.
Technology
RustGerber RS-274XExcellonG-code
Build
cargo build --release; cargo test
|
||||||||||
| Tools - the programs that sit above the four ecosystems and update or watch any of them. | ||||||||||
| Ecosystem Operations 1 project(s) · ELECTRO-HOBBY-3D-UPDATER is this family's own integration parent | ||||||||||
| Tool | Functional | python | User PC | 0.1.6 |
OK
Version successfully resolved
|
Prepare version 0.1.6 (#10) | ||||
|
Notes
Does not belong to HYDRA-UMC, URTC, A.R.M.O.R. or EH3D - it sits above all four, dispatching to whichever ecosystem's own updater the user picks via --ecosystem. No sub-updater discovers this manifest (none of the four matches ecosystem: "ELECTRO-HOBBY-3D"), by design; this repository is meant to be found the same way the JuanenRac profile dashboard finds every other top-level tool. Never run against a real machine end to end yet.
Technology
Python 3.10+hydra-umc-updaterurtc-updatereh3d-updaterarmor-updater
Build
python -m pytest tests -q
|
||||||||||
| Tools 3 project(s) · TOOLS-GITHUB-ACTIONS-MONITOR is this family's own integration parent | ||||||||||
| Tool | Functional | python | User PC | 2.1.7 |
OK
Version successfully resolved
|
Merge pull request #335 from JuanenRac/fix-218-pagina | ||||
|
Notes
A standalone tool: it belongs to no ecosystem and needs none of them. It watches the GitHub Actions workflows of the repositories of its owner and shows them on a local page. Part of the Tools group of the profile.
Technology
Python 3.11+GitHub CLI (gh)PrometheusGrafana
Build
python -m unittest discover -s monitor/tests
|
||||||||||
| UI | Scaffolding | Kotlin / Jetpack Compose | Mobile | 0.1.0 |
OK
Version successfully resolved
|
Bump the actions group with 2 updates (#2) | ||||
|
Notes
Preliminary release v0.1.0: the native Android client of GitHub Actions Monitor. It reads the read API v1 of a monitor with a bearer token kept sealed in the Android Keystore, shows the summary, the alerts, the repositories and the action proposals, lets a person approve or reject a proposal (the server enforces that a second person approves) and opens ghamon://proposal links. Push notifications are critical-only and OFF by default, and the Firebase side is a null implementation until a Google project exists. The rules are tested on the JVM and the release APK is built by a manual GitHub Actions workflow and signed by the maintainer; the app has not been run on a phone yet.
Technology
KotlinJetpack ComposeRead API v1 of GitHub Actions MonitorAndroid Keystore
Build
./gradlew testDebugUnitTest lintDebug
|
||||||||||
| UI | Scaffolding | Wear OS | Wearable | 0.0.1 |
OK
Version successfully resolved
|
Say that the planned watch version targets Wear OS (#1) | ||||
|
Notes
Scaffolding v0.0.1: the planned minimal Wear OS watch version of the Android app of GitHub Actions Monitor, which follows TOOLS-GITHUB-ACTIONS-MONITOR-ANDROID and would show a small part of what the monitor knows. It only reads from the monitor and writes nothing to GitHub. There is no code yet: this repository holds only its base (documents, manifest, licence and shared checks). Nothing has been built, tested or run on a Wear OS watch; how the watch gets its data is not decided.
Technology
KotlinWear OSRead API v1 of GitHub Actions Monitor
Build
Not built yet: there is no watch project. The repository checks run with python -I tools/ci_validate.py.
|
||||||||||
Automated tests by ecosystem
Declared, not run hereNumber of automated tests that each repository declares in its own README or glance file. It is a declared figure: the dashboard reads it and does not run the tests.
How is this calculated?
No test count could be read in this build, so nothing is verified.
Total tests per ecosystem
| Ecosystem |
|---|
Evolution of the total
One point per day with a snapshot. A day without a snapshot breaks the line. The total can rise because a repository starts to declare its tests, not only because tests were written.
| Date |
|---|
Tests per project
Horizontal bars ordered by the declared count. Select a project to open its panel.
No project matches the filters.
| Project |
|---|
Code statistics
Counted on shallow clonesLines of code, tests and documentation of every repository, counted once a day on a fresh shallow copy. Generated, vendored, dependency, binary and image files are left out.
How is this counted?
No code statistics could be read in this build, so nothing is verified.
Code, tests and documentation
Share of each kind of line in the total of every group. Tests are test files; documentation is text files including the translated READMEs.
- Code
- Tests
- Documentation
| Repository |
|---|
Languages
Lines of code per programming language (tests included), for the whole ecosystem and for each ecosystem. The smallest languages are grouped as Other.
| Repository |
|---|
Evolution
Lines per day from the daily snapshots (recent days one by one, older ones thinned out to one per week or month).
Whole ecosystem
By ecosystem
| Repository |
|---|
Largest repositories
The repositories with most lines, stacked by code, tests and documentation.
| Repository |
|---|
Detail
Click a column heading to sort. Filter by ecosystem and type, search by name or group the rows.
| Repository |
|---|
No row matches the filters.
Changelog statistics
Read from the CHANGELOG filesVersions and entries of the changelog of every repository, with the kind of each entry decided by words. The pace of publishing comes from the tags of GitHub, never from the files.
How is this counted?
Kinds of change
Share of each kind among the published entries and those waiting, for the repositories selected by the filters.
| Repository |
|---|
Kinds of change by ecosystem and type
| Repository |
|---|
Changes over time
Entries per day from the daily snapshots (recent days one by one, older ones thinned out to one per week or month).
Whole ecosystem
Published entries by ecosystem
| Repository |
|---|
Ranking of repositories
Most changes (published and waiting)
Highest share of fixes
| Repository |
|---|
| Repository |
|---|
Detail of a repository
Choose a repository to see its versions, how many entries each has and of which kind. Choose a version to read its entries, search them and filter them by kind.
| Version |
|---|
| Repository |
|---|
Changelogs by repository
Click a column heading to sort. The filters, the grouping and the search apply to the table, the donut, the ranking and the exported file.
Exports the rows shown in the table, as they are sorted and filtered.
| Repository |
|---|
No row matches the filters.
Workflows and scripts
Read from the repositoriesThe GitHub workflows of every repository (when they run, the actions they use and how the version of each action is fixed) and the scripts of its tools/ and scripts/ folders by language. Read once a day from a fresh shallow copy; nothing is run.
How is this counted?
No workflow or script figures could be read in this build, so nothing is verified.
When the workflows run
Workflows per trigger for the whole, each ecosystem and each type; a workflow with several triggers counts once for each.
| Repository |
|---|
How the actions are pinned
Uses of actions by how their version is fixed, for the whole, each ecosystem and each type. A full commit cannot move; a tag or a branch can.
| Repository |
|---|
Most used actions
The actions used most often (the repository of the owner included), stacked by how their version is fixed.
| Repository |
|---|
Scripts by language
Files of tools/ and scripts/ by language, for the whole and for each ecosystem.
| Repository |
|---|
Repositories with most workflows
The repositories with most workflows, stacked by trigger.
| Repository |
|---|
Workflow lint: three levels
The same rules and the same three levels as the local Actions monitor: information, warning and error. Actions not pinned to a commit, workflows with no permissions block, what looks like a secret written in the file, text of outsiders pasted into a script and a pull-request-target that checks out the pull request. Only counts are kept: never a value, a name or a line.
Not verified yet: the figures of the daily count do not carry the lint findings until the next count runs.
| Repository |
|---|
Findings by rule
| Repository |
|---|
Actions not pinned to a commit
Actions of other owners whose version is a tag or a branch, with the number of uses and of repositories.
Every action of another owner is pinned to a commit.
Workflows and scripts by repository
Sort with the column headings, filter by ecosystem or type, search by name, and group the rows by ecosystem or type.
| Repository |
|---|
No row matches the filters.
Actions / Builds
Near real timeState and figures of the GitHub Actions runs of the last 24 hours and 7 days. Updated every 30 minutes and when a build of a repository of the ecosystem ends. No author, branch or commit text is kept.
The builds could not be read in this generation and there is no earlier read: not verified.
Result · last 24 hours
Success rate · 7 days · lowest first
No finished runs in 7 days
Duration of the last run · longest first
No finished runs in 7 days
Timeline · last 24 hours
No runs in the last 24 hours
Runs per day
The daily history starts with the first reading; it fills in one day at a time.
Repositories
| Repository |
|---|
No repository matches.
Speed, queue and flaky runs
The p50 and p95 of how long the successful runs of the last 7 days take and of how long a run waits before it starts, and the workflows that failed and then passed on the very same commit. Same definitions as the analytics of the local Actions monitor, taken per run; the label "flaky" never changes what counts as red.
Not measured yet: the last reading does not carry these figures.
Slowest workflows (p95)
Nothing to show
Flaky workflows
Nothing to show
How it is calculated
Computing minutes are approximate: wall-clock time of the finished runs of the last 24 hours, not the minutes GitHub bills (which depend on the runner and are rounded per job).
Sentinel: watch over the ecosystem
Rules, no AIA public-safe summary read when the page was generated: builds on the main branch, releases ready, signatures, public hygiene, computing and trends. It is decided by fixed rules, with no AI. Security alerts, pull requests, issues and people are deliberately not shown.
The sentinel has no reading yet: not verified.
The sentinel is switched off in the configuration.
Report (made by rules, no AI)
Signature of the latest releases
Builds on the main branch
| Repository |
|---|
Releases ready
| Repository |
|---|
Public hygiene
By ecosystem
| Ecosystem |
|---|
Computing (approx.)
Daily trend
Version raised, not published
The manifest of the repository declares a newer version than its latest release.
| Repository |
|---|
Recent changes
Signature and attestation of the latest release
Presence only: whether each release file has a signature file next to it and whether GitHub holds a build attestation for its digest. No key and no proof is verified here.
| Ecosystem |
|---|
By repository
| Repository |
|---|
Unknown means it could not be read (no release data, no digest, a refused token or the request budget used up); it is never shown as absent.
Shared contract pins
The version of the shared contracts each repository was last checked against, read from the files the repositories publish, compared with the current version of the contracts. A count of what is written there, not a test of compatibility.
Supply chain
What the workflow files of each ecosystem allow: actions not fixed to a commit, actions that follow a branch, permissions that are missing or too wide and the pull request trigger that runs with the rights of the repository. Counts only, read from the daily copy of the repositories.
| Ecosystem |
|---|
Delivery
Whether what was released is complete: a version tag newer than the latest release, a release with an empty description and a CHANGELOG that does not name the version released. Read when the page was generated.
| Ecosystem |
|---|
How it is calculated
What this tab never shows: Security alerts (code scanning, dependencies and secrets, not even a count), titles of pull requests and issues, people, branches and texts written by outsiders. They are left out on purpose so that nothing sensitive is published.
Ecosystem analysis
Maturity funnel by ecosystem
Each bar is the number of repositories that reached at least that level; the figures beside it give how many are exactly at that level.
| Ecosystem |
|---|
The eight areas, side by side
Same scale for the five groups, from the centre (0%) to the edge (100%). The table below holds the same figures.
| Area |
|---|
Weekly activity by ecosystem
Commits and releases of the last twelve weeks. The number is always written in the cell; a darker cell means more, compared with the same kind of figure in the other ecosystems.
Only the releases that still exist are counted: older ones are removed on purpose. A week of an ecosystem counts only the repositories that could be read.
| Ecosystem and measure |
|---|
The weekly activity could not be read when this page was generated, so nothing is drawn.
Who uses whom
Read from the depends_on list of every manifest. Arrows point from the repository that uses to the one it uses.
Dependency cycles
Dependencies on repositories that do not exist
No manifest declares dependencies, so there is nothing to show.
Next milestones
The titles of the cards are kept in the language of the boards.
Compare two dates
Only the snapshots that exist are offered; nothing is estimated between them.
Visible technical debt
Three rules, applied to the data read when this page was generated. A repository that cannot be judged is listed apart as not verified, never counted as debt.
Evidence checklist by repository
What each repository can show: tests, green CI, a release, a signed release, README in 8 languages, simulation and real hardware. Every item is verified, missing, not verified or not applicable.
The signature is read from the latest release of each repository and checked when the page is generated, with the public key that its family pins: verified, invalid, present but not verifiable here, no signature or not verified (a read failed).
Counts by ecosystem
| Repository |
|---|
| Repository |
|---|
No repository matches these filters.
There is no repository data in this page, so no checklist is shown.
Declared against verified
The manifest of each repository declares how far it has been proven. Here that claim is compared with the records its own CI publishes: a passing run on the main branch, not older than the limit of its kind, proves its level. Nothing published means not verified.
No evidence was read in this build, so nothing is verified.
| Repository |
|---|
No repository matches this state.
v3: real maturity/role classification, family/parent trees and richer per-project notes - see the Maturity legend below for exactly how each level was decided.
- Scaffolding
- Functional
- Established
- Production
- Unknown
| Ecosystem | Scaffolding | Functional | Established | Production | Unknown | Total |
|---|---|---|---|---|---|---|
| HYDRA-UMC | 2 | 52 | 5 | 0 | 0 | 59 |
| URTC | 2 | 8 | 1 | 0 | 0 | 11 |
| A.R.M.O.R. | 2 | 6 | 12 | 0 | 0 | 20 |
| EH3D | 17 | 24 | 0 | 0 | 0 | 41 |
| Tools | 2 | 2 | 0 | 0 | 0 | 4 |
- None
- Host test
- Simulator
- Bench
- Field
- Unknown
| Ecosystem | None | Host test | Simulator | Bench | Field | Unknown | Total |
|---|---|---|---|---|---|---|---|
| HYDRA-UMC | 0 | 0 | 54 | 3 | 2 | 0 | 59 |
| URTC | 0 | 0 | 11 | 0 | 0 | 0 | 11 |
| A.R.M.O.R. | 0 | 0 | 16 | 0 | 4 | 0 | 20 |
| EH3D | 1 | 7 | 33 | 0 | 0 | 0 | 41 |
| Tools | 1 | 1 | 0 | 0 | 2 | 0 | 4 |
Not a feature timeline with dates - the real path every project in this ecosystem follows, and exactly where each one sits on it right now (see the Maturity section above for the live counts).
-
Scaffolding 25
Real, tested code exists, but a large part of the project is still a skeleton or an open plan, and nothing has been tried in its real context. A project can finish its own plan and stay here until it is proven on its real target.
Becomes functional once the real capability exists and is tested on a developer machine and in CI against fakes, emulators or fixtures, with the README saying what the tests stand in for.
-
Functional 92
A real capability, tested on a developer machine and in CI against fakes, emulators or fixtures. No run on the real target (board, machine, CM5, cameras) is recorded; the README says what the tests stand in for.
Becomes established once the repository records a run on its real target (real hardware, a real CM5, a real board) or, for a pure software tool, its whole workflow runs end to end in CI and on a developer machine.
-
Established 18
The repository itself records a run on its real target (real hardware, a real CM5, a real board), or it is a pure software tool whose whole workflow runs end to end in CI and on a developer machine. Passing its own tests is never enough on its own.
Becomes production only after a continuous, unattended deployment with no known open gap.
-
Production 0
A continuous, unattended deployment with no known open gap. Reserved: no repository claims it yet.
The top of the ladder. Nothing advances past here - it just keeps shipping real fixes.
Real project counts by role and deployment target, with row, column and grand totals, drawn straight from every repository's own manifest. A count, not a claim that any two specific projects interoperate.
| Role \ Target | CM5 | Field Node | STM32 MCU | AI Server | Container Host | User PC | Mobile | Wearable | Browser Client | Dev Server | Shared / Ecosystem-wide | Total |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| API | 11 | – | – | 1 | – | – | – | – | – | – | – | 12 |
| UI | 4 | – | 2 | – | – | 4 | 6 | 2 | 1 | – | 2 | 21 |
| CLI | – | – | – | – | – | 1 | – | – | – | – | – | 1 |
| Firmware | – | 5 | 4 | – | – | – | – | – | – | – | – | 9 |
| Library | 5 | – | – | – | – | 10 | – | – | – | – | 1 | 16 |
| Service | 23 | – | – | 2 | – | 2 | – | – | – | 2 | 1 | 30 |
| Tool | – | – | 4 | – | 1 | 29 | – | – | – | – | – | 34 |
| Documentation | – | – | – | – | – | 10 | – | – | – | – | 1 | 11 |
| Hardware Design | – | 1 | – | – | – | – | – | – | – | – | – | 1 |
| Total | 43 | 6 | 10 | 3 | 1 | 56 | 6 | 2 | 1 | 2 | 5 | 135 |
| Role \ Target | CM5 | STM32 MCU | User PC | Mobile | Wearable | Dev Server | Total |
|---|---|---|---|---|---|---|---|
| API | 10 | – | – | – | – | – | 10 |
| UI | 3 | – | 1 | 2 | 1 | – | 7 |
| CLI | – | – | 1 | – | – | – | 1 |
| Firmware | – | 1 | – | – | – | – | 1 |
| Library | 5 | – | 3 | – | – | – | 8 |
| Service | 23 | – | – | – | – | 2 | 25 |
| Tool | – | – | 6 | – | – | – | 6 |
| Documentation | – | – | 1 | – | – | – | 1 |
| Total | 41 | 1 | 12 | 2 | 1 | 2 | 59 |
| Role \ Target | STM32 MCU | User PC | Total |
|---|---|---|---|
| UI | – | 1 | 1 |
| Firmware | 3 | – | 3 |
| Library | – | 1 | 1 |
| Tool | – | 5 | 5 |
| Documentation | – | 1 | 1 |
| Total | 3 | 8 | 11 |
| Role \ Target | Field Node | AI Server | Container Host | User PC | Mobile | Browser Client | Shared / Ecosystem-wide | Total |
|---|---|---|---|---|---|---|---|---|
| API | – | 1 | – | – | – | – | – | 1 |
| UI | – | – | – | – | 1 | 1 | 2 | 4 |
| Firmware | 5 | – | – | – | – | – | – | 5 |
| Library | – | – | – | – | – | – | 1 | 1 |
| Service | – | 2 | – | 1 | – | – | 1 | 4 |
| Tool | – | – | 1 | 2 | – | – | – | 3 |
| Documentation | – | – | – | – | – | – | 1 | 1 |
| Hardware Design | 1 | – | – | – | – | – | – | 1 |
| Total | 6 | 3 | 1 | 3 | 1 | 1 | 5 | 20 |
| Role \ Target | CM5 | STM32 MCU | User PC | Mobile | Total |
|---|---|---|---|---|---|
| API | 1 | – | – | – | 1 |
| UI | 1 | 2 | 2 | 2 | 7 |
| Library | – | – | 6 | – | 6 |
| Service | – | – | 1 | – | 1 |
| Tool | – | 4 | 14 | – | 18 |
| Documentation | – | – | 8 | – | 8 |
| Total | 2 | 6 | 31 | 2 | 41 |
| Role \ Target | User PC | Mobile | Wearable | Total |
|---|---|---|---|---|
| UI | – | 1 | 1 | 2 |
| Tool | 2 | – | – | 2 |
| Total | 2 | 1 | 1 | 4 |
URTC tool rack
State of the tool rack as the snapshot of URTC-COMMON describes it: slots, heads, maintenance and alarms. This page only shows a published snapshot; it never reads a rack.
Alarm · Simulated rack
Source: simulation
This is simulation data published as an example, not a real rack.
Slots in use: 3 of 8 · heating: 1 · maintenance due: 1
Alarms
- Slot 0: maintenance due for unit 0:7
Slots
| Slot | Tool | Unit | Target °C | Thermal check | Wear |
|---|---|---|---|---|---|
| 0 | 0 | 0:7 | 250 | target reached | Maintenance due |
| 1 | 19 | 19:2 | 0 | idle and cold | No data |
| 2 | 4 | 4:5 | 0 | idle and cold | No data |
| 3 | Empty | – | 0 | Not verified | No data |
| 4 | Empty | – | 0 | Not verified | No data |
| 5 | Empty | – | 0 | Not verified | No data |
| 6 | Empty | – | 0 | Not verified | No data |
| 7 | Empty | – | 0 | Not verified | No data |
Electro Hobby 3D is four independent engineering ecosystems by the same author: HYDRA-UMC (industrial multi-robot platform), URTC (its universal robot-tool subsystem, an independent product with its own firmware), A.R.M.O.R. (perimeter security and home automation) and EH3D (a PCB, CNC and laser fabrication suite). This dashboard discovers and lists every one of them live; each ecosystem's own family groups are labeled below, and HYDRA-UMC's own deeper architecture is detailed in the section right after this one.
Platform foundation
Raspberry Pi OS ARM64 remains the operating-system base. The HYDRA-UMC layer adds device profiles, diagnostics and service lifecycle.
- HYDRA-UMC-OS
- CM5 / Linux
- systemd / udev
- MCU / URTC boundary
Contracts and operations
The SDK defines stable data and command contracts; Server, UI and tools use those contracts instead of raw hardware protocols.
- HYDRA-UMC-SDK
- Server / Studio / Suite
- DSI / mobile / CLI
- Job Dispatcher
Perception and intelligence
Vision and AI are optional capabilities. Their output is validated before it can influence a mission; they are never safety authority.
- Vision Streamer / Node
- Detection HEF
- Cognitive / VLA
- Safety Zones
Engineering and industry
Simulation, telemetry and standards make the physical cell observable, testable and interoperable.
- Twin / Physics / HIL
- Telemetry / DataLake
- OPC-UA / MQTT
- MTConnect / Gateway
Contracts and tests
HYDRA-UMC-SDK holds the v1 JSON Schemas, a Python reference validator, Go, TypeScript and Rust clients that enforce them, conformance fixtures, an OpenAPI slice of the server and a job intake that never runs the same idempotency key twice; a compatibility matrix and a differential test keep validator and schemas in step.
- HYDRA-UMC-SDK
- JSON Schema / OpenAPI
- Go / TypeScript / Rust clients
- Job intake
Verified updates
HYDRA-UMC-UPDATER installs and updates a repository by cloning, staging, verifying and only then promoting it, so a failed check leaves the live copy untouched; an optional gate refuses a commit without a trusted signature before anything is built. Firmware images are checked with CRC32 and a shared-key HMAC-SHA256; an asymmetric signature is planned, not implemented.
- HYDRA-UMC-UPDATER
- Staged install
- Signed-commit gate
- CRC32 + HMAC-SHA256
Operations and self-healing
HYDRA-UMC-OPS-AGENT diagnoses and applies only reversible repairs; HYDRA-UMC-NODE-HEALING watches health endpoints and recovers a node through the orchestrator; HYDRA-UMC-LOCAL-TECHNICIAN reads a machine without changing it and assembles evidence bundles; HYDRA-UMC-DEVELOPER-NODE (new, first version) ingests those bundles, runs repro cases a person approved and keeps a cache of checkouts, all closed by default; HYDRA-UMC-DEV-SERVER holds the contracts and limits for development hosts.
- HYDRA-UMC-OPS-AGENT
- HYDRA-UMC-NODE-HEALING
- HYDRA-UMC-LOCAL-TECHNICIAN
- HYDRA-UMC-DEVELOPER-NODE
Safety, twin and data
HYDRA-UMC-SAFETY-ZONES checks zone breaches and only ever requests a stop; it can optionally read the alarm state of A.R.M.O.R. to inhibit new starts, never to stop a machine that runs. HYDRA-UMC-TWIN and HYDRA-UMC-COGNITIVE-NODE check the readiness of their child services and HYDRA-UMC-PHYSICS-REPLICA computes forward kinematics and joint limits. HYDRA-UMC-DATALAKE stores time series and HYDRA-UMC-DASHBOARD-AI summarises them. HYDRA-UMC-STUDIO-KIT (new, first version) shares sessions, history and an HTTP client between web clients. None of it has met a physical machine.
- HYDRA-UMC-SAFETY-ZONES
- HYDRA-UMC-TWIN / COGNITIVE-NODE
- HYDRA-UMC-DATALAKE / DASHBOARD-AI
- HYDRA-UMC-STUDIO-KIT
Tool-head firmware
An STM32F303 runs each tool head over CAN, reading its own 5-bit hardware address to auto-configure power stages, sensors and safety logic for one of 25 built-in tool profiles.
- STM32F303
- CAN bus
- 5-bit ID matrix
- 25 tool profiles
Expansion and motion
A 20-pin expansion connector adds a second stepper axis or sensor board through one of six interchangeable variants, sharing STEP/DIR/EN wiring between a TMC2209 and a TMC5160 driver.
- TMC2209 / TMC5160
- 6 expansion boards
- F-RAM persistence
- OLED diagnostics
Maintenance tools
URTC-FLASHER updates firmware over CAN without removing the board, URTC-TESTER exercises the protocol end to end and URTC-UPDATER keeps every desktop tool current from GitHub.
- URTC-FLASHER
- URTC-TESTER
- URTC-UPDATER
- CAN-OTA
Vision and operation
URTC-VISION-TOOL adds camera-based tool inspection, URTC-WEB-STUDIO gives it a browser console and URTC-SMART-RACK stores and identifies tool heads between changes.
- URTC-VISION-TOOL
- URTC-WEB-STUDIO
- URTC-SMART-RACK
Contracts and tests
URTC-COMMON writes the CAN protocol once as data, generates a C header, TypeScript and Python modules and a protocol table from it, and keeps the conformance vectors every implementation must reproduce. The firmware and the desktop tools still keep their own copies until each one migrates, so today it is compared against them rather than used by them.
- URTC-COMMON
- CAN spec + generators
- Conformance vectors
- Migration plan
Release assets and signing
Each firmware release is published as GitHub release assets with SHA-256 sidecars and a manifest, and the release tool refuses to publish images signed with the public development key as an official release. The bootloaders check CRC32 and a shared-key HMAC-SHA256; asymmetric signatures are planned, and nothing has been flashed to a real board yet.
- GitHub release assets
- SHA-256 + manifest
- Dev-key release gate
- Asymmetric signing planned
Gateway, shared tools and documentation
URTC-GATEWAY sits in front of the CAN bus for an installation that cannot keep it closed: only identifiers made from the specification cross, dangerous ones are refused unless unlocked, with a payload check, a rate limit and a session key, each one switchable. URTC-DEVTOOLS keeps the shared repository checks as one versioned set and URTC-DOCS holds the architecture, the security baseline and the capability matrix. All three are tested on a computer only.
- URTC-GATEWAY
- URTC-DEVTOOLS
- URTC-DOCS
Network segmentation
Field nodes and cameras sit on their own VLAN with no internet access; the MQTT broker and server live on a core VLAN; Studio and the Android app reach the server from a client VLAN.
- VLAN 10 field
- VLAN 20 core
- VLAN 30 clients
- MQTT, one identity per node
Field nodes
ESP32-S3 radar, solar, electrical, alarm and wall-panel nodes publish sensor and equipment readings or their state; the solar and electrical ones only read, never write to an inverter, battery or meter.
- ARMOR-RADAR
- ARMOR-SOLAR
- ARMOR-ELECTRICAL
- ARMOR-ALARM
- ARMOR-HMI
Perception, never authority
Visual and voice AI recommend a severity or an intent with reasons attached; authorizes_action is always false until the server itself authenticates and confirms.
- ARMOR-SERVER-AI
- ARMOR-VOICE-AI
- Recommend, never act
State and operator consoles
ARMOR-SERVER is the only place that ever touches a camera password or an RTSP address; Studio and the Android app are its authenticated clients, and a phone can set a new node up over Bluetooth alone.
- ARMOR-SERVER
- ARMOR-STUDIO
- ARMOR-ANDROID-CONTROL
- ARMOR-HMI
- BLE set-up only
Contracts and tests
ARMOR-COMMON defines every node message once as a JSON Schema, with a validator that refuses what it does not implement, accepted and rejected conformance vectors that each implementation must pass, the OpenAPI file of the server's routes and the firmware base shared by the five nodes, which a tool keeps identical in each of them.
- ARMOR-COMMON
- JSON Schema / OpenAPI
- Conformance vectors
- Shared firmware base
Signed releases and updates
Node firmware is published with a SHA-256 and an Ed25519 signature, and both a node and ARMOR-SERVER refuse a release without a valid one. The maintenance helper of ARMOR-DEVOPS checks an Ed25519 signature on each service package before unpacking it, which is how ARMOR-STUDIO updates the central machine; the private key never goes to GitHub. A firmware file uploaded by hand is not checked.
- Ed25519 firmware signatures
- ARMOR-DEVOPS signed packages
- ARMOR-STUDIO
- Key never on GitHub
Simulator, network watch and the iPhone client
ARMOR-SIMULATOR produces repeatable telemetry and faults without real nodes; ARMOR-NETWORK watches the network of the test bench and only observes; ARMOR-HARDWARE keeps the enclosure concepts, which need a physical prototype to be validated; ARMOR-IOS-CONTROL is the iPhone port of the Android client, its rules tested without a phone and its screens not yet run on a device. The alarm node can publish its state for HYDRA-UMC-SAFETY-ZONES, which may read it as an optional input.
- ARMOR-SIMULATOR
- ARMOR-NETWORK
- ARMOR-HARDWARE
- ARMOR-IOS-CONTROL
One project file and shared crates
EH3D-COMMON defines the .eh3dproj container (a ZIP archive with a hash for every entry and limits that refuse unsafe names), the telemetry types of a machine and the crates the programs share: the scene of a job, the words of the interface, archives, the Bungard file formats, the machine controllers, the height map of the table and the file windows.
- EH3D-COMMON
- .eh3dproj
- SHA-256 integrity
- Shared crates
Readers and documentation
EH3D-CORE-LIBS reads Gerber, Excellon and G-code files from other people, never panics on a damaged one and reports what it could not understand. EH3D-DOCS is the hub: the map of the suite, the architecture and the manuals.
- EH3D-CORE-LIBS
- Gerber / Excellon / G-code
- EH3D-DOCS
CAM and CNC programs
The programs turn PCB layers and images into toolpaths and drive the machine: ISOCAM, DRILLPRO, ROUTEPRO, STENCILCAM, PHOENIX-LASER, AUTOLEVEL and CAM-DEBUGGER, GERBERVIEWER to look at the layers, MOTIONCNC to stream G-code, and BUNGARD-GUI, PNP-MANAGER, VISION-STUDIO and TOOL-MANAGER around them. They depend on the libraries and exchange work through .eh3dproj files, never on each other.
- EH3D-ISOCAM
- EH3D-DRILLPRO
- EH3D-ROUTEPRO
- EH3D-PHOENIX-CNC
- EH3D-GERBERVIEWER
Server and Studio
EH3D-SERVER is the center: it takes the jobs of the programs and the clients, plans them, checks them against the limits of each machine and runs them through its node, a simulator or a GRBL controller. EH3D-STUDIO is the web panel it serves, to set up machines and to prepare and follow jobs.
- EH3D-SERVER
- EH3D-STUDIO
- Simulator / GRBL
Machine nodes
EH3D-NODE-BUNGARD is an ESP32-S3 node that puts the serial port of a Bungard machine on the network; EH3D-NODE-BUNGARD-AUX sits on the drill head (table height, spindle speed and temperature, camera); EH3D-NODE-BUNGARD-HMI and EH3D-NODE-HMI-GUI are the touch screen of the machine. They share the firmware base kept in EH3D-COMMON.
- EH3D-NODE-BUNGARD
- EH3D-NODE-BUNGARD-AUX
- EH3D-NODE-BUNGARD-HMI
- EH3D-NODE-HMI-GUI
Mobile clients, launcher and updater
EH3D-ANDROID-CONTROL and EH3D-IOS-CONTROL are clients of EH3D-SERVER only: they read the fleet and the jobs, prepare jobs and stop them, and never talk to a machine. EH3D-SUITE-LAUNCHER finds the programs by their manifests and starts them, and EH3D-UPDATER keeps every repository of the suite current on the PC.
- EH3D-ANDROID-CONTROL
- EH3D-IOS-CONTROL
- EH3D-SUITE-LAUNCHER
- EH3D-UPDATER
Watch
TOOLS-GITHUB-ACTIONS-MONITOR watches the workflows of every repository of an account and shows what is red, stuck or waiting on one local page in eight languages. It only reads unless an action is allowed, and its Complete edition adds the sentinels.
- TOOLS-GITHUB-ACTIONS-MONITOR
- Complete and Lite editions
- Sentinels
Update
ELECTRO-HOBBY-3D-UPDATER is one window and one command line that hand over to the updater of each ecosystem: HYDRA-UMC-UPDATER, URTC-UPDATER, ARMOR-UPDATER and EH3D-UPDATER. It has no updater of its own.
- ELECTRO-HOBBY-3D-UPDATER
- HYDRA-UMC-UPDATER
- URTC-UPDATER
- ARMOR-UPDATER
- EH3D-UPDATER
Report
JuanenRac, the profile repository, reads the manifest of every repository, builds this dashboard in GitHub Actions and publishes it as a static page that makes no request of its own. What could not be read is shown as not verified.
- JuanenRac
- Manifests
- Static page
Generation of this page
Generated 2026-11-22 08:09 UTC in 272 s (the latest generation that changed the page).
- Versions resolved: 135 of 135 repositories
- Valid manifests: 135 of 135 repositories
- Last-commit lookups answered: 135 of 135
- CI and release records read: 135 of 135 repositories
Reads of GitHub: 154; saved by the ETag or because nothing was pushed: 139; answered from the saved copy: 0 (oldest 0 h)
The exact reason of every manifest that could not be read, of every lookup that failed and of every GitHub failure is listed by repository below.
What counts as an error or unknown 0
The number counts repositories the README catalogue promises that the dashboard does not show, manifests that could not be read, maturity or evidence values that are missing or not recognised, and failures of the GitHub API while this page was generated.
135 of 137 public repositories are listed: those with the manifest of an ecosystem. Without a manifest: .github, JuanenRac.
Nothing is missing, unreadable or unknown: every repository of the catalogue is listed and every manifest declares a known maturity and evidence.