This page records measured local evidence. It is not a publication, contest submission, customer claim, or proof of performance on non-synthetic data.
pytest: 56 passed, with two deprecation warnings.ruff: clean.mypy: 16 source files clean.- Node syntax check: clean.
The accepted native local receipt completed on 2026-09-06. It used the pinned local qwen3.5:9b-q4_K_M model, synthetic data only, and the explicit human UI routes exercised by an automated test harness. It passed in 324.67 seconds. The receipt covers baseline, repair, diagnosis, and simulation workflow evidence. An earlier native evaluation was rejected for a status-handling bug; the regression was fixed before the accepted receipt.
The accepted synthetic uploaded-note probe completed on 2026-09-06 in 40.694 seconds with a real local model. It inspected four provider calls and confirmed that raw names, titles, and the instruction-like note were excluded from model input. The resulting schedule had eight preferred assignments. This is evidence of the tested probe only.
media-work/ShiftProof-Demo.mp4 is a 208.25-second rendered visualization using synthetic operators and recorded results. Inference waits were shortened for presentation. It is not a live recording.
The sanitized structured receipts are native workflow and uploaded-note probe. They include actual tool activity, measured durations, and source-receipt checksums.
The public source repository is https://github.com/YusefSyed/shiftproof. The public video is published. The signed-in Builder ID was verified and the entry was submitted September 6, 2026. Devpost displayed Project submitted and Submitted to: Agents for Humans Hackathon. This proves receipt, not judging eligibility, selection, a prize, or payment. See submission checklist.
The final interface was exercised with the live local provider: input review, baseline scheduling, explicit approval, a simulated absence, exact-change confirmation, approval revocation, and repair. The repaired view showed six preferred assignments, two replacements, unchanged helper assignments, and both leads within their 240-minute/two-shift limits.