Trusted final: infrastructure valid; geometry checks executed; gates not passed. Official J withheld; reporting S withheld.
Verified full run archiveTask · Design discovery · v2
MZI-v2
Broadband 6x6 Rectangular MZI Mesh
Design a broadband 6×6 rectangular MZI mesh with 36 quantized controls and trusted optical, electrical, and electrothermal verification.
Published runCodex · GPT-5.6-sol
No valid designReporting score S · provisional calibration—
Official score J · lower is better—
J_acc = 0.457 · calibration design-jacc-2026-09-05-v2. S ≥ 1 meets the aggregate reporting reference, not every optical or resource specification.
Saved-run scores and gate outcomes are unchanged; this reporting update is not a new evaluation against current task code.
Published runClaude Code · Claude Opus 5.5
ValidReporting score S · provisional calibration0.7345
Official score J · lower is better0.7874
J_acc = 0.457 · calibration design-jacc-2026-09-05-v2. S ≥ 1 meets the aggregate reporting reference, not every optical or resource specification.
Saved-run scores and gate outcomes are unchanged; this reporting update is not a new evaluation against current task code.
2 published runs are compared on one elapsed-time chart. Their final layouts and score evidence remain separate.
Evidence chain
Why these results are reported this way
Trusted final: infrastructure valid; geometry checks executed; gates passed. Official J 0.7874; reporting S 0.7345.
Verified retained run archiveThe saved ZIP contains the retained source, probes, final evaluator evidence and layout, but the five raw final thermal HDF5 files were not captured. Private authorization and Claude session identity files were excluded. No simulation was rerun for this archive.
How to interpret this result
A valid outcome means the submitted artifact cleared the evaluator gates recorded for this archived run. It does not imply that every possible physical specification, fabrication condition, or later task revision has been independently re-tested.
Run trajectory · model comparison
Reporting score S vs. time
Each run starts at its own elapsed time zero. Hollow markers are gate-ineligible probes, not measured S = 0.
Codex · GPT-5.6-solcircle probes · solid bestClaude Code · Claude Opus 5.5triangle probes · dashed best
Codex · GPT-5.6-sol
No valid design · 40 probes · reporting score S ineligible
Final artifact · Codex · GPT-5.6-sol
Codex · GPT-5.6-sol GDS layout
The exact final submission, rendered from its evaluator-produced GDS. It failed the final gates and has no official score.
Claude Code · Claude Opus 5.5
Valid · 10 probes · reporting score S 0.7345
Final artifact · Claude Code · Claude Opus 5.5
Claude Code · Claude Opus 5.5 GDS layout
The exact submission the evaluator scored, rendered from its GDS