| { |
| "variant": "shared_drafts", |
| "audit_status": "passed", |
| "source_protocol_sha256": "d99323c7d332f60f842f5960eaa47dad2b0ec2a79f5a34c115b93cee9605dc3b", |
| "protocol": { |
| "development_only": true, |
| "population": [ |
| "scibench_atkins_065", |
| "scibench_atkins_042", |
| "scibench_atkins_015", |
| "scibench_calculus_023", |
| "scibench_calculus_036", |
| "scibench_calculus_037", |
| "scibench_chemmc_018", |
| "scibench_chemmc_006", |
| "scibench_chemmc_002", |
| "scibench_class_049", |
| "scibench_class_051", |
| "scibench_class_010", |
| "scibench_diff_021", |
| "scibench_diff_010", |
| "scibench_diff_032", |
| "scibench_fund_017", |
| "scibench_fund_004", |
| "scibench_fund_055", |
| "scibench_matter_015", |
| "scibench_matter_023", |
| "scibench_matter_021", |
| "scibench_quan_013", |
| "scibench_quan_032", |
| "scibench_quan_003", |
| "scibench_stat_003", |
| "scibench_stat_058", |
| "scibench_stat_016", |
| "scibench_thermo_014", |
| "scibench_thermo_055", |
| "scibench_thermo_054" |
| ], |
| "stages": [ |
| "paired_scibench_30" |
| ], |
| "intervention": "True code weights versus zero-delta sham; otherwise identical frozen inputs and harness.", |
| "completion": "Ordinary Python stdout, no report/quote gate; same-adapter reserved final call. Accept one visible boxed number or a whole bare finite number. No draft extraction or reference access.", |
| "stop_rule": "Offload/bare completion; identical successful code+feedback repeats; two consecutive display-only executions; 4 successful or 10 total executions; 3 consecutive failures; existing resource reserves.", |
| "execution_limits": { |
| "total": 10, |
| "successful": 4, |
| "consecutive_failures": 3, |
| "sandbox_creations": 3, |
| "walltime_reserve_seconds": 240, |
| "consecutive_display_only": 2 |
| }, |
| "startup_gate": "Both arms must pass four independent native canaries: arithmetic, LaTeX and SI units, correction of a faulty draft, ordinary stdout. Same first requests, correct final results, real execution, code routing and cleanup required. Caught exceptions and uncaught exceptions are both allowed. Uncaught error propagation is covered by deterministic real-sandbox preflight, not gated on model coding style.", |
| "budgets": { |
| "requests": 40, |
| "completion_tokens": 32768, |
| "turns": 40, |
| "seconds": 1200, |
| "final_completion_reserve": 1024, |
| "normal_call_max": 8192, |
| "tool_output_max_tokens": 4096, |
| "context_length": 32768, |
| "context_margin": 1536 |
| }, |
| "sandbox_timeout": "120-second HTTP execution read timeout; kill owned sandbox before replacing it; tell agent all state was lost.", |
| "task_concurrency_per_arm": 1, |
| "sandbox_slots_across_arms": 2, |
| "sampling": { |
| "temperature": 0, |
| "seed": 42 |
| }, |
| "scoring": "Unchanged SciBench numeric scorer, 5% relative tolerance, one actual terminal artifact only.", |
| "diagnostic_only": { |
| "scibench_diff_032": "Missing numeric damping coefficient; report both original 30 and predeclared 29." |
| }, |
| "interpretation": "Known development tasks; recovery and cost diagnostic. Generalization requires fresh questions.", |
| "bounded_delivery": true, |
| "evidence_control": false, |
| "harness_strategy": "shared_base_draft_v1", |
| "hypothesis": "Separate initial scientific modeling from code-adapter execution by supplying one shared base-model draft. No reference answers, terminal submissions, execution outputs, or per-task correctness selection enter the plans." |
| } |
| } |
|
|
