Last status update: November 3, 2026. This is a snapshot, not a claim of live status. Check the source repository for subsequent changes.
What is working
| Subsystem | State | Evidence / limit |
|---|---|---|
| Breadboard geometry + netlist model | Implemented | Existing Python library models known board electrical nodes and components. |
| Circuit checker + repair suggestions | Implemented | Existing JSON examples can trigger specific wiring diagnoses. |
| Image rectifier and hole overlay | Implemented | Previous developers demonstrated real-image normalization; current team reproduced sample output. |
| Arduino LED diagnostics | Prototype demonstrated | Previous team operated USB hardware tests; not yet replicated as a complete classroom deployment. |
| Local core software baseline | Reproduced | On WSL: 425 tests passed and 58 skipped; lint, format, types and tests subsequently passed. |
| GitHub collaboration / CI | Configured | Team fork, protected main, and a successful automated test workflow. |
What is experimental or incomplete
| Work item | State | What would count as progress |
|---|---|---|
| Resistor RF-DETR pose model | Experimental / not yet reproduced by current team | Run the provided checkpoint on a real image; save keypoint predictions and setup details. |
| Clean evaluation set | Not established | Test configurations are distinct from training configurations; evaluation can be rerun. |
| Additional component classes | Not complete | Correct endpoint detections for wires, LEDs, and relevant sensors. |
| Wire-end association | Not complete | Reliably pair two ends belonging to each physical wire. |
| Complete image → netlist | Not complete | Generate the checker-compatible build data automatically from photographs. |
| Classroom browser integration | Not complete | End-to-end camera, model, checker, and hardware workflow in the target deployment. |
Project milestones
Milestone 01 — Inherited foundation: reproduced. The existing software is available through the team fork. The core tests pass locally, and the normalizer and JSON checker have been exercised. Some archived real-photo and hardware-recording tests skip because their source data is not in the public repository.
Milestone 02 — ML baseline: next. Load Hartley’s original RF-DETR checkpoint and reproduce a resistor detection with two endpoint keypoints on a real normalized image. Document the inference version and machine requirements.
Milestone 03 — Evaluation. Establish dataset provenance and a train/test split that does not put multiple views of the same physical wiring into both groups. Report endpoint and hole-level accuracy, not just object detection scores.
Milestone 04 — Integration slice. Convert detections to electrical-node or hole identifiers using the normalizer and create valid component records for the existing checker. Test the mapping deterministically.
Milestone 05 — Expand and validate. Add the missing component and wire association capabilities, then test whole-circuit outputs and abstentions on real boards. This is the larger research-and-engineering objective, not an accomplished result.
Dated notes
November 3, 2026 — Team environment established
The current team established a shared GitHub fork, branch protections, automatic CI checks, a common development process, and school Drive access to seven historical checkpoints and two datasets. The existing application was reproduced on a WSL desktop. The remaining work shifts from infrastructure into verifying Hartley’s resistor model and building the perception layer.
October 27, 2026 — Previous-team handoff
The current team met with Stone Stephenson and Hartley Blakey to understand the project’s architecture and inherited resources. The vision model and labeled evaluation data were identified as the main development priorities.
Earlier 2026 — Previous development
The earlier team implemented circuit modeling, checking, image normalization, hardware testing, and experimental resistor pose detection. Their prototype and research established a starting point, not a fully automated photograph-to-circuit solution.
What would convince us that a milestone is done?
We want reproducible evidence: a checkpoint name, test input, command, output, sample count, measurement, and known failure cases. A polished screenshot without a traceable evaluation is a demonstration, not a benchmark.
Read the computer-vision deep dive for the exact questions the model must answer.