You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

FullStackArena Auction: GPT-5.6-sol, one instance per template

This private evidence release contains the complete September 7, 2026 Auction qualification run: one deterministic task from each of all 25 released templates, executed with five isolated website/database/clock replicas.

Result interpretation

  • Raw framework result: 18/25 pass (72.0%).
  • Manual trace-audited result: 17/25 pass (68.0%).
  • Confirmed strict false positive: auction_454.
  • Confirmed strict false negatives: none.
  • Diagnostic LLM-judge disagreements are preserved but are not authoritative.

The raw files have not been rewritten to hide the disagreement. The authoritative review is recorded in release_evidence/manual_audit.json, including the task-level rationale and integrity hashes.

Layout

The task artifact shape follows FullStackArena's valid_runs/<site>/<mode>/<task> convention. Because this was one five-worker run and every worker had an independent deployment, database, clock, transport, run manifest, and signed content manifest, the worker boundary is retained explicitly:

valid_runs/auction/gpt-5.6-sol_uncapped_no-include-time/worker_0/
valid_runs/auction/gpt-5.6-sol_uncapped_no-include-time/worker_1/
valid_runs/auction/gpt-5.6-sol_uncapped_no-include-time/worker_2/
valid_runs/auction/gpt-5.6-sol_uncapped_no-include-time/worker_3/
valid_runs/auction/gpt-5.6-sol_uncapped_no-include-time/worker_4/

Each worker root independently validates against fullstackarena-run-artifacts/v1. Each task directory includes the trace, structured steps, every screenshot (including the final state), conversations, legacy request/response pairs, the complete model-call request/response ledger, task log, and token summary. Run-level configuration, results, provenance, validation, content manifests, and logs are stored at the corresponding worker root.

Aggregate evidence is in release_evidence/; task_index.json maps every task to its template, worker, and audited classification. release_manifest.json hashes every release payload file except itself. The immutable Hub revision is the checksum for the manifest itself.

Privacy and access

This repository is intentionally private. Full trajectories contain screenshots, complete model/browser records, and deployment metadata. Automated secret scans found no API keys, bearer credentials, private keys, or runtime-auth files. Do not make this raw repository public without a separate sanitization and review.

Task index

Template Task Worker Strict Diagnostic Audited Classification
auction_f01 auction_409 0 pass pass pass true_positive
auction_f02 auction_405 1 pass fail pass diagnostic_false_negative
auction_f03 auction_519 2 pass pass pass true_positive
auction_f04 auction_411 3 pass pass pass true_positive
auction_f05 auction_420 4 pass pass pass true_positive
auction_f06 auction_507 0 pass pass pass true_positive
auction_f07 auction_500 1 pass pass pass true_positive
auction_f08 auction_514 2 pass pass pass true_positive
auction_f09 auction_510 3 pass pass pass true_positive
auction_f10 auction_513 4 fail fail fail true_negative
auction_f11 auction_535 0 fail fail fail true_negative
auction_f12 auction_443 1 pass pass pass true_positive
auction_f13 auction_223 2 pass pass pass true_positive
auction_f14 auction_161 3 pass pass pass true_positive
auction_f15 auction_460 4 fail pass fail diagnostic_false_positive
auction_f16 auction_273 0 pass pass pass true_positive
auction_f17 auction_303 1 pass pass pass true_positive
auction_f18 auction_467 2 fail pass fail diagnostic_false_positive
auction_f19 auction_454 3 pass fail fail strict_false_positive
auction_f20 auction_540 4 pass pass pass true_positive
auction_f21 auction_543 0 fail fail fail true_negative
auction_f22 auction_493 1 fail fail fail true_negative
auction_f23 auction_526 2 fail fail fail true_negative
auction_f24 auction_504 3 pass pass pass true_positive
auction_f25 auction_537 4 pass fail pass diagnostic_false_negative

Reproducibility anchors

  • FullStackArena commit: de65f7683886255c7df149e850c00c25e17aad52
  • Auction source commit: 58bc388aba6e15f9a0306a1a4051d24a0ca8c6ef
  • Task pack SHA-256: beb8afc469f4a0b50dfe624b9da21e7d85f2dec550a32d0814579a07b056602d
  • Template pack SHA-256: cc66517a30c2d83d5b9853c19b2517f46c3df96a549b9b66604eb701616d8943
  • Selection digest: 0174737e7a503a11bf531925489e9aebd40bdd298f1a72b4b8334c6db542b472
  • Experiment wall clock: 50m 12.9s (manual audit excluded)
  • Model calls: 460
  • Total tokens: 5645360
  • Recorded experiment cost: $7.3989168
Downloads last month
16