| { |
| "schema": "mt11-olmo3-think-evalscope-exact240-contract/1", |
| "contract_id": "OLMO3_THINK_EXACT240_BATCH32", |
| "eval_path": "A", |
| "framework": "EvalScope 1.9.1 CLI via frozen OLMo capture proxy", |
| "score_authority": "EvalScope 1.9.1 generated review rows", |
| "datasets": { |
| "aime24_plus_aime25_unique_items": 60, |
| "seeds": [ |
| 42, |
| 43, |
| 44, |
| 45 |
| ], |
| "review_identities": 240, |
| "identity_basis": "seed+dataset+index" |
| }, |
| "sampling": { |
| "temperature": 0.6, |
| "top_p": 0.95, |
| "top_k": 20, |
| "n_per_request": 1 |
| }, |
| "runtime": { |
| "dynamic_lora": true, |
| "merged": false, |
| "formal_eval_batch_size": 32, |
| "max_num_seqs": 32, |
| "enable_chunked_prefill": true, |
| "max_model_len": 32768, |
| "reserve_tokens": 64, |
| "max_tokens_rule": "32768 - exact_prompt_tokens - 64", |
| "stop_token_ids": [ |
| 100265, |
| 100257 |
| ], |
| "template_sha256": "6d549883b5ed12879e191845c256a30c7dfd4eced0a4f160060a3ea0199d9e3a" |
| }, |
| "formal_result": { |
| "candidate_correct": 46, |
| "candidate_total": 240, |
| "candidate_accuracy": 0.19166666666666668, |
| "fresh_base_correct": 41, |
| "fresh_base_total": 240, |
| "paired": { |
| "wins": 19, |
| "losses": 14, |
| "ties": 207 |
| }, |
| "status": "PASS_POSITIVE_SIGNAL" |
| }, |
| "contract_checks": { |
| "candidate_identity_240": true, |
| "fresh_base_identity_240": true, |
| "cursor_identity_pass": true, |
| "captures_240": true, |
| "dynamic_prompt_budget": true, |
| "double_stop": true, |
| "official_template": true, |
| "dynamic_lora": true, |
| "fresh_base_independently_recomputed_41": true |
| }, |
| "corrected_termination_diagnostics": { |
| "cap_definition": "prediction choice stop_reason/finish_reason in {max_tokens, length}", |
| "natural_stop_definition": "prediction choice stop_reason/finish_reason == stop", |
| "candidate": { |
| "cap": 162, |
| "cap_correct": 3, |
| "natural_stop": 78, |
| "natural_stop_correct": 43 |
| }, |
| "fresh_base": { |
| "cap": 159, |
| "cap_correct": 0, |
| "natural_stop": 81, |
| "natural_stop_correct": 41 |
| }, |
| "historical_diagnostic_correction": "The matched120 helper counted only the literal value length, while this runtime emitted max_tokens. Its zero-cap field is therefore not reused. This correction changes termination diagnostics only and does not alter formal scores." |
| }, |
| "boxed_diagnostics": { |
| "candidate_literal_boxed_outputs": 66, |
| "candidate_literal_boxed_and_formally_correct": 44, |
| "fresh_base_literal_boxed_outputs": 63, |
| "fresh_base_literal_boxed_and_formally_correct": 40, |
| "candidate_independent_numeric_scoreable": 101, |
| "candidate_independent_numeric_correct": 26, |
| "candidate_last_boxed_numeric_scoreable": 33, |
| "candidate_last_boxed_numeric_correct": 22, |
| "policy": "Box presence and independent numeric extraction are diagnostics only. No post-hoc boxed correction is added to or removed from the authoritative EvalScope 46/240 score." |
| }, |
| "completed_at_utc": "2026-08-23T17:47:51Z", |
| "source_gate_receipt_sha256": "3d6daff73b0b9fba0c66093f5e5ca1ad3f35d535c57d5144722c7e8a707fe7b2" |
| } |
|
|
