{
  "sft": [
    {
      "id": "S1",
      "path": "paper/distill-fingerprint/tex/main.tex",
      "label": "现稿摘要与章节入口",
      "date": "2026-09-06 现稿",
      "sha256": "033790b2f28e199e122e496621afe54544eafdce2e1981d9920025ca381d2998",
      "excerpts": [
        "We curate 8,137 \\Teacher/\\Harness trajectories from 32,830 captured runs with an explicit rule set, fine-tune two open Qwen families on them for one to three epochs from bare and post-trained bases, and evaluate twelve arms in one harness on \\SWEPro and \\SWEVerified; one extractor fingerprints all 46,542 teacher and student trajectories, and two same-pool anchors put teacher and students on the same tasks.",
        "Fingerprint proximity changes whether or not \\passat changes: the corpus lifts a bare base by $+8.9$\\pp on Pro ($+3.4$\\pp on Verified), every comparison against a post-trained base ($-2.2$ to $+2.8$\\pp) stays below the paired design's detectable effect, and the next Qwen generation gains $+5.6$ to $+5.8$\\pp over the native dense model at $1.35\\times$ its input.",
        "\\input{sections/01_intro}",
        "\\input{sections/02_related}",
        "\\input{sections/03_corpus}",
        "\\input{appendix/A_evaluation_noise}",
        "\\input{appendix/B_fingerprint_robustness}",
        "\\input{appendix/C_extras}"
      ]
    },
    {
      "id": "S2",
      "path": "paper/distill-fingerprint/PAPER_PLAN.md",
      "label": "贡献定位、锚点与关闭实验",
      "date": "2026-09-06 更新",
      "sha256": "d4b96d91203a84a8aacd3cdac2539750773e8aebc34492c58ca5d6dcd7b0697e",
      "excerpts": [
        "**One-sentence contribution**: A curated corpus of 8,137 frontier-agent (Claude Opus 4.8 / Claude Code) trajectories, built by a rule set we describe and measure, moves open students' command vocabulary, error recovery, thinking rhythm and input-token footprint toward the teacher — most of the dense model's convergence after one epoch — while pass@1 response depends on the starting model: +8.9 pp on a bare base (Pro), and −2.2 to +2.8 pp against post-trained baselines with no comparison resolved at this evaluation scale.",
        "- One-sentence contribution (above). Approach: curate → train 1/2/3 epochs on two families → evaluate 12 arms → fingerprint everything → anchor → audit.",
        "- [x] GPU: held-out anchor (must) — landed 2026-09-06: val373 → 345 oracle-passed tasks × 8 arms (`analysis/anchors/val373/`), pro100 grew to 359 Pro tasks run by the teacher (`analysis/anchors/pro359/`); Appendix E (Tables E1–E5) + body Figure 15; \"task-independent\" claimed (composite 0.051 → 0.006 at one epoch on never-trained tasks; read precision 0.56 → 0.78 held-out vs 0.50 → 0.80 trained); corrections applied: teacher makes 0 task-management calls on the benchmark (3.6 % on val, 3.2 % corpus), thinking position within 2× floor on the same tasks",
        "- [x] Same-condition retest (reviewer item 3): STOPPED 2026-09-06 15:50 at 9/345 by user decision (no further experiments unless strictly necessary; temperature twins + split-half floors stand as the reliability reference) — had started 13:20 on inst1 engine + gpu23 sandboxes: n27 and s256 again on val373 at T=0.8 (`chain_val373.sh LINE=C ARMS=\"n27 s256\" TAG_SUFFIX=-retest`); when done, add flip rate and fingerprint test–retest to Appendix A/E and §7",
        "- [x] Qwen3.8-27B + v12 graft: dropped 2026-09-06 (user decision, \"不测了\"); the generation comparison stays descriptive with the native Qwen3.8-27B arm only"
      ]
    },
    {
      "id": "S3",
      "path": "paper/distill-fingerprint/tex/tables/tab1_arms.tex",
      "label": "主表：分数、输入中位数与 Bash-JS",
      "date": "2026-09-06 当前文件",
      "sha256": "49818f7b9d95804c8cbcad1363ee17f1b9f21f0ec202568c5ea7d8027e4e7ac7",
      "excerpts": [
        "\\nTwoSeven & 50.1 & 68.4 & -- & -- & -- & 3.27 & 0.202 \\\\",
        "\\sOneTwoEight & 49.7 & 67.8 & $-0.41$ & $-0.60$ & $-1.22$ & 1.69 & 0.069 \\\\",
        "\\sTwoFiveSix & 52.8 & 68.6 & $+2.74$ & $+0.20$ & $+2.60$ & 1.52 & 0.053 \\\\",
        "\\sThreeEightOne & 51.0 & 71.2 & $+0.96$ & $+2.80$ & $-0.15$ & 1.75 & 0.050 \\\\",
        "\\aBase & 35.4 & 52.0 & -- & -- & -- & 3.40 & 0.278 \\\\",
        "\\aTwoEp & 44.3 & 55.4 & $+8.89$ & $+3.40$ & $+9.62$ & 2.44 & 0.057 \\\\",
        "\\caption{The twelve evaluation arms, named by base model and corpus dose (\\dose{v12}{2} = two epochs of \\vTwelve; \\dose{v13}{1} = one epoch of \\vThirteenCorpus, 362 updates). \\passat = solved tasks over the pool (731 \\SWEPro, 500 \\SWEVerified; anomalous trials count as unsolved), one rollout. $\\Delta$: percentage points against the line's native arm (\\qThreeEight: against \\nTwoSeven, the generation jump). $\\Delta$Pro$-$tp: Pro without the 76 teleport tasks ($N=655$). Input: median cumulative input tokens per task on Pro, millions. Bash-\\JS: Jensen--Shannon distance ($\\log_2$) of the Bash head-word distribution to the teacher on Pro. The two $T{=}0.6$ rows are the same weights as the rows above them, run a second time (the run-to-run reference, \\S\\ref{sec:measurement}). ``--'': not defined for a native arm.}"
      ]
    },
    {
      "id": "S4",
      "path": "paper/distill-fingerprint/findings.md",
      "label": "run 差异、锚点与评分限制",
      "date": "截至 2026-09-06",
      "sha256": "84f0cc4512a7c02d4047df72e2b9ba1fc352695ed9e3f130bb6115f840f12034",
      "excerpts": [
        "- Full draft: abstract, §1–8, Limitations, Ethics, appendix A (7 tables + endings audit), B (8), C (5) — every appendix table generated from the analysis JSON/MD by `tex/tables/make_tabA/B/C.py`; `references.bib` filtered to the 45 cited entries (`tables/make_refs.py`). Body ends on page 8 of the ACL two-column build (28 pages with appendix), 0 errors, 0 undefined references, no figure fallback text in the PDF.",
        "s256 and s381 come from the same run (`output_dir=/output/qwen36-27b-v12-3ep/v0-20260809-124327`,",
        "`max_length=196608`, `save_steps=64`); **s128 does not** (`output_dir=/v12/ckpt128-hf`,",
        "`max_length=262144`, no `save_steps`). All three start from `/models/Qwen3.6-27B` on corpus v12,",
        "All eight arms finished 345/345 (`bench/results/anchors/val373/summary.md`, regenerated 09-06):",
        "- 🔴 The two MoE arms are **not same-family** with the six 27B dense arms, and not with each other"
      ]
    },
    {
      "id": "S5",
      "path": "paper/distill-fingerprint/tex/sections/09_limitations.tex",
      "label": "现稿 Limitations",
      "date": "当前稿件",
      "sha256": "9b04b60a149f76f17836cda5db7d0ede47a7f2bf1adfe82ad86fc24c1e313e57",
      "excerpts": [
        "Every student learns from trajectories of a single frontier agent captured in a single harness, and every arm is evaluated in that same harness.",
        "The teacher's benchmark trajectories cover 359 of the 731 \\SWEPro tasks and none of \\SWEVerified, and its thinking is returned encrypted there, so same-task comparisons on the benchmark use tool rulers only; the held-out anchor shares repositories, not tasks, with the corpus, and its two MoE arms start from different bases.",
        "The 180 endings (110 regex-negative, 70 regex-positive) and 68 rationalization windows were labeled by two model annotators, which agree on the two quantities the body uses (0 of 110 admissions of failure; regex precision 69 of 70) and differ on the rationalization split (48 versus 49 verified attributions, 12 versus 6 bare dismissals); no human pass was run, and the body states only the agreed quantities.",
        "\\paragraph{No single-rule ablation.}",
        "The MoE dose curve has two points per base and every arm has one rollout per task, so the behavioral rulers are read against split-half floors and the two repeat-sample arms."
      ]
    },
    {
      "id": "S6",
      "path": "paper/distill-fingerprint/CLAIMS_FROM_RESULTS.md",
      "label": "早期 result-to-claim（历史版本）",
      "date": "2026-09-02",
      "sha256": "415be75a7d8c48ee154b3ad6afc41715c608a2b517b85993c8b2614f84df36c4",
      "excerpts": [
        "verdict: PARTIAL",
        "integrity audit: not run (`EXPERIMENT_AUDIT.json` absent) → this verdict is provisional per the skill's rules.",
        "| H2 fingerprint vs capability | partial | high | \"Fingerprint proximity changes substantially whether or not pass@1 changes: bare-A3B SFT improves Pro by 8.89 pp but is inconclusive on Verified, while no baseline-relative post-trained comparison is significant and its intervals still permit several-point effects.\" | Holm over the 111 valid paired tests; MDE re-derivation on 731; teleport-excluded tables; equivalence-style inference instead of \"≈0\"; v12 on Qwen3.8-27B; drop causal \"score comes from the base\" wording |",
        "| H3 cost | partial | high | \"Under the Claude Code Pro harness, v12/v13 SFT reproducibly reduces cumulative input tokens through shorter runs and more bounded observations, while increasing output tokens; this is a token-use signature, not a general monetary or wall-clock saving.\" | paired CIs per task; mini-SWE-agent harness check (nice to have); rename to \"input-token footprint is part of the fingerprint\" |"
      ]
    },
    {
      "id": "S7",
      "path": "paper/distill-fingerprint/CITATION_AUDIT.md",
      "label": "引用审计与已应用修订",
      "date": "2026-09-06",
      "sha256": "c22361b34fe7aa618d098c42b813c85211be459548b704fe4996e8ae7f3edec1",
      "excerpts": [
        "**Date**: 2026-09-06  **Bib file**: tex/references.bib (61 cited entries, generated from lit/refs.bib + lit/refs_manual.bib)  **Reviewer**: gpt-5.6-sol (xhigh), four fresh threads; web access blocked in the sandbox, existence checked against the arXiv API record of 2026-09-02/03, vendor URLs re-checked by the orchestrator.",
        "## Priority fixes (all applied 2026-09-06)",
        "- **advani2026from / cloud2025subliminal** — unverified venues (FAGEN@ICML 2026; Nature) dropped -> arXiv-only entries",
        "- **singh2025from** — clause removed; entry no longer cited",
        "| singh2025from | REPLACE | context | clause removed; entry no longer cited |"
      ]
    }
  ],
  "rl": [
    {
      "id": "R1",
      "path": "paper-rl/paper/sections/5_evaluation.tex",
      "label": "现稿评测：pilot、成本与效果Pending",
      "date": "当前稿件",
      "sha256": "90a6a8574b27466b6e876253200ff9e7762f5747fc8c62cfe4cb0e71484bc811",
      "excerpts": [
        "\\paragraph{Policy effectiveness: pending.}",
        "The base-model development pilot generated eight trajectories on each of the original 40 tasks. Of these 320 planned trajectories, 244 met the training query rules and all 244 intervention trials returned usable reports. Here, usability follows the training adapter's fixed-list utility contract. The exclusions were 41 format-error trajectories, 34 without an eligible block, and one without an original utility after the agent deleted installation files. This last trajectory is retained as a model failure; it is not resampled.",
        "The contrast is nonzero for 92 queries (37.7\\%), including 17 from failed original trajectories: 12 positive and five negative, spanning 14 tasks and six repositories. These are 17 of the 111 queried failed trajectories, within 150 original trajectories with a known failed verdict. The twelve positive contrasts split evenly between a positive fail-to-pass (F2P) component and the regression indicator alone (\\cref{app:signal-origin}). These observations establish information availability on failures, rather than a benefit from learning with that information. \\Cref{tab:pilot} preserves the original pilot denominator and separately exposes the 39-task stratum admitted to training.",
        "The cap activates for 83 of the 92 nonzero contrasts (90.2\\%), placing most observed nonzero cases in the sign-and-support regime of \\cref{eq:coefficients}. Among these 92 cases, the median selected-token fraction $m/L$ is 2.10\\% and median shrinkage $s$ is 0.181; \\cref{app:geometry} gives their distributions. Before format-error and original-utility query exclusions, 282 trajectories have a constructed intervention artifact; of these, 47 become an empty patch. The median fraction of final changed paths touched by the selected block is 0.5. One of the 92 nonzero cases instead selects paths absent from the final patch: the block repairs earlier agent damage. This illustrates why last-touch eligibility is defined relative to the pre-block tree rather than only to the final net diff. These structural measurements do not certify semantic attribution accuracy.",
        "All 120 prescribed cross-machine factual replays match their originals at the level of reported test labels. The initial 244 intervention trials take 237.3 seconds on average, with a 416.9-second empirical 95th percentile. A separate deployment repeat of the same 244 artifacts on the designated verifier host, at concurrency 24 with images retained between trials, yields identical per-test reports for all cases and takes 115.1 seconds on average, with a 189.9-second 95th percentile. Service time includes setup, verification, and cleanup, and excludes semaphore waiting. Host placement, concurrency, and cleanup differ together; the comparison does not isolate a cache effect. Production queueing and total training cost remain unmeasured.",
        "All 120 prescribed cross-machine factual replays match their originals at the level of reported test labels. The initial 244 intervention trials take 237.3 seconds on average, with a 416.9-second empirical 95th percentile. A separate deployment repeat of the same 244 artifacts on the designated verifier host, at concurrency 24 with images retained between trials, yields identical per-test reports for all cases and takes 115.1 seconds on average, with a 189.9-second 95th percentile. Service time includes setup, verification, and cleanup, and excludes semaphore waiting. Host placement, concurrency, and cleanup differ together; the comparison does not isolate a cache effect. Production queueing and total training cost remain unmeasured."
      ]
    },
    {
      "id": "R2",
      "path": "paper-rl/PAPER_PLAN.md",
      "label": "方法与O/B/M/E实验计划",
      "date": "2026-09-27",
      "sha256": "f345575600d0c77bdd22b735a141a30a97a0513e3d65dbe0061cc37063e8cbe1",
      "excerpts": [
        "| M/B在同一轨迹上具有同带符号/L1原始辅助系数质量 | 公式代数、CPU实际loss接入检查 | 技术支持；不含梯度范数/裁剪等价 | §3 |",
        "| 每条准入轨迹至多一个额外verifier query、零策略续写 | 已实现查询路径；实际端到端成本仍在测 | 查询界受支持；可接受成本及相对成本尚未证实 | §3/§5 |",
        "主终点80。M−B是主要比较；M−O、M−E次之。独立seed和task配对统计放附录，主表不写CI，但不删实验范围/局限。给出真实成本字段，待实测后填值。",
        "E的辅助项为`C_E(t)=clip(lambda*A*L/m,-c,c)*I_e(t)`，加到原终局组优势A上；不访问D，m=0/选择不可用时为0。",
        "主终点80。M−B是主要比较；M−O、M−E次之。独立seed和task配对统计放附录，主表不写CI，但不删实验范围/局限。给出真实成本字段，待实测后填值。"
      ]
    },
    {
      "id": "R3",
      "path": "paper-rl/paper/sections/6_limitations.tex",
      "label": "现稿干预、优化与判分限制",
      "date": "当前稿件",
      "sha256": "c0d214dcd0f06f9efd49d21e934252f7c53c7e32cdfd1c949ca2e73967d72adc",
      "excerpts": [
        "Final-artifact removal conditions on the other completed files. Later edits on other paths may depend on the removed block, even when no later turn overwrites its paths. The resulting contrast is not an unbiased action value, a Shapley allocation, or the outcome of a naturally continued trajectory. Last-touch and test-path eligibility also restrict which edits can receive this feedback. Snapshot attribution is temporal: background writes captured after a turn can be assigned to that turn, and restoring a last-touch block can reintroduce damage from an earlier block.",
        "Equal signed and absolute coefficient mass does not isolate semantic placement from every optimization effect. Localization changes concentration; adding the terminal advantage can change signs and DPPO masking. O/B/M/E evaluate a specific practical allocation rule. Claims about attribution accuracy require a suitably matched relocation control, and claims that removal information is necessary require controls using simpler factual test feedback. Those conclusions are outside the present comparison.",
        "The planned study uses one base model and a limited RL task pool. It does not establish model-family-wide effectiveness, and base-model pretraining overlap is not measured here. An earlier adapter admitted zeros without grading reports; the affected run was excluded from the comparison. Successor training requires complete grading evidence, but this can also withhold negative feedback for model-broken installation; retry and masking rates must be reported by arm (\\cref{app:grading_recovery}). The official evaluation's expected-test lists can omit behavior relevant to an issue; retaining its labels and denominator preserves the benchmark definition rather than repairing that limitation. Matching reports establish reproducibility of the recorded score. Missing test observations can be classified as failures, and reused descriptions can overwrite earlier test statuses. Final evaluation therefore retains raw observations and separates observed failures from outcomes requiring attribution."
      ]
    },
    {
      "id": "R4",
      "path": "paper-rl/results/O42_retry1_external_termination_20260928.json",
      "label": "O42-retry1外部终止收据",
      "date": "2026-09-28 15:35",
      "sha256": "5b9ce35674fd8114d171fb00d8b0336ffdd278b5f6e6bc9161e12d06244577f9",
      "excerpts": [
        "\"observed_at\": \"2026-09-28T15:35:17.342206+08:00\",",
        "\"quality_observed_at\": \"2026-09-28T15:34:35.821420+08:00\",",
        "\"host_observed_at\": \"2026-09-28T15:34:54.658129+08:00\",",
        "\"training_call_completed\": 34,",
        "\"termination_class\": \"external_or_host_termination_unresolved\",",
        "\"automatic_restart\": false,",
        "\"oom_killed\": false,",
        "\"next_action\": \"Do not resume until a separately reviewed recovery plan binds checkpoint30 and termination evidence\""
      ]
    },
    {
      "id": "R5",
      "path": "paper-rl/results/common_profile_runtime_coverage_final_20260930.json",
      "label": "500共同环境最终汇总（纠正后）",
      "date": "2026-10-01 纠正",
      "sha256": "45ef82e6cc1980428d679e28ad6f29ab4ec72fa843521448f25a5172a2385a48",
      "excerpts": [
        "\"scope\": \"All500 common-profile tasks have been exercised across retained waves; unresolved and mismatches remain explicit. No model-effect claim.\",",
        "\"controls\": 1000,",
        "\"controls\": 1000,",
        "\"raw_pass\": 995,",
        "\"strict_pass\": 862,",
        "\"report_binary\": 887,",
        "\"unresolved\": 113,",
        "\"partial_zeros\": 20,",
        "\"control_expectation_mismatches\": 5,",
        "\"control_expectation_mismatches\": 5,",
        "\"report_binary\": \"Binary report diagnosis includes unexpected control scores; not a count of passed controls or formal model slots\""
      ]
    },
    {
      "id": "R6",
      "path": "paper-rl/PAPER_ACCEPTANCE_CONTRACT.md",
      "label": "草稿验收约定与结果版本缺口",
      "date": "2026-09-27",
      "sha256": "3d9ae3c2667e2c4a0bc56dbf24e83b279b8267c122956181e230d585242208cd",
      "excerpts": [
        "status: accepted",
        "assurance: draft",
        "9. 数据表明确实际训练149、2384别名、dev39；官方500及holdout100分母不变；原150映射与未用于后继训练的351不混用。",
        "10. 80步、三种子计划、E42、n=2/n=16及当前执行状态明确区分；成功率按每条rollout求均值，绝不取best-of-n，计划的holdout100 n=16明确包含B；不声称未运行的重复实验已经完成。",
        "11. 所有主结果/成本空缺使用Pending或破折号并解释未测量，不使用模拟数字、臆测曲线或旧无效O/B/M成绩。全文任何位置都不能给出尚未测得的本方法胜负/泛化提升或更低总成本结论。",
        "这份约定针对当前草稿，不把“方法有效”作为可凭空满足的写作要求。最终提交阶段必须在真实终评后重新建立结果版本的约定；不能把本草稿验收等同投稿就绪。"
      ]
    },
    {
      "id": "R7",
      "path": "paper-rl/STATUS_20260928.md",
      "label": "当前交接：汇总更正与运行边界",
      "date": "2026-10-01 更新",
      "sha256": "824026dba6207f23184e0c9dd09e37f9e8ea21b534a21ce4b6a878a03db40b2f",
      "excerpts": [
        "**2026-10-01汇总更正：** `common_profile_runtime_coverage_final_20260930.json` 已按23个保留cohort重算：500题/1000控制，raw 995，strict 862，report-binary 887，未决113，部分观测20，参考/环境差异5。原文件中的999/779/798/101/18为聚合错误，已由提交 `fdd83033` 修正；这些是环境验证覆盖数字，不是模型效果。",
        "- **O42-retry1在第34次调用后随父容器外部终止。** 质量观察器最后看到34/80，训练父 `skyrl-rl` 曾 exit137、Docker `OOMKilled=false`；driver控制读取进入 review。step20与step30 HF导出/参数审计已完成；step30为427张量、329个变化张量、823171568个变化元素、无非有限值。`skyrl-rl` 后来被外部重新启动，但没有形成O42恢复证据，不把容器重启当自动恢复，也不把存储变化写成效果。收据 `results/O42_retry1_external_termination_20260928.json`、`O42_retry1_step30_parameter_audit_20260928.json`。",
        "旧O40仍作废于主比较，所有旧分数/存点保留。新runtime、数据、launch和release都保持冻结。gpu08 GPU6/7及GPU0原小服务不动；监督不自动重启。",
        "- 后继入口F0–F3/两轮复核已完成，真实O80、终点参数核验及CF适用性release仍欠；B/M/E未启动。共同候选仍87题/174控制、16未决、413题未新跑，不能用新诊断改变分母。论文仍15页、效果未测，当前没有等待自动重试的CPU控制任务。",
        "下一步：核验首个原生存点的参数；实现共同评测的可信参考/完整成员映射/准入规则；沿已完成后继入口准备真实O80与CF证据绑定的release。旧CPU控制/原始未决和论文数字均未修改。",
        "- 后继入口的CPU核验明确：当前恢复make_plan只收O42，前驱校验只认旧O40停训且拒绝retry1名称。要另建支持固定矩阵和正常80调用前驱的新入口；9份配置计划不是release。见 `results/successor_launch_boundary_20260928.json`。",
        "剩余9个预定训练条件的比较计划已经生成：B/M各3种子、O另2种子、E42。全部通过实际SkyRL配置解析，与活跃O仅臂、种子和输出路径不同。数据、原base、runtime、采样、优化器及14卡布局保持相同；没有启动后继或申请新GPU。入口 `staged/credit-after-recovery-20260928/`，计划 `rl/runs/rl-paper-20260927/credit-after-recovery-plans-v2/`。v1清单相对路径在runtime cwd下无法定位，原失败保留，v2改绝对路径。计划不是release；后继仍要核对新O80实际结束、清理、源、CF与资源证据。"
      ]
    },
    {
      "id": "R8",
      "path": "paper-rl/results/grading_recovery_paper_update_20260928.json",
      "label": "恢复限制稿局部审查与编译回执",
      "date": "2026-09-28",
      "sha256": "7756cd4c431ab0056467e333e00fcdd6feafe4e39f6a71e6b0fbe4c31fc05b8e",
      "excerpts": [
        "\"scope\": \"Limitations and appendix recovery/resource-control changes only, no efficacy claim or full160-claim reaudit.\",",
        "\"review_independence\": \"same-family\",",
        "\"acceptance_status\": \"provisional\",",
        "\"pdf_pages\": 15,",
        "\"main_body_end_page\": 9,",
        "\"submission_ready\": false,",
        "\"policy_effectiveness_measured\": false,"
      ]
    }
  ],
  "jev": [
    {
      "id": "J1",
      "path": "paper-jev/HANDOFF_20260930_CURRENT.md",
      "label": "当前checkpoint权威交接与结果",
      "date": "2026-09-30",
      "sha256": "10a11e81d52647b713221bf1c800a9802586e14db7114c4c1520603cc527f7f4",
      "excerpts": [
        "项目仍未有可以诚实写成主方法的创新结果。test checkpoint 标签没有读取，`runs/ckpt-policy-freeze/` 尚未建立；因此不能调 test 阈值、不能做 test 结果选择，也不能开始最终论文定稿。GPU6/7 归本项目长期使用，GPU0–5 属于 RL，不能碰。",
        "- dev：40 任务、160 次尝试、4,091 个决策标签。",
        "- ext：300 任务、599 次尝试捕获；其中 580 次/296 任务可用于训练，13,601 个训练决策点。",
        "- test：60 任务、239 次尝试捕获；官方 checkpoint 标签继续封存。",
        "九个 0.6B 文本模型（joint / independent-now / independent-final，三 seed，含 final1 recovery4 与 final2 successor）均有独立 readout。总体审查结论为 `PARTIAL_DESCRIPTIVE_LABEL_ABLATION_ONLY`：当前标签相对最终标签在部分 resolve-loss 菜单上省输入 token，但解决率和 harm 存在 trade-off，没有建立相对 counts 的安全增益、部署保证或总体保证。",
        "- counts failure-termination：Q=77/160，输入省 17.7944%，但 H 的条件区间不能支持总体 H≤1%。",
        "test 允许做的事情只有不含标签的特征计算，例如文本缓存或 9B 自评。所有任务、特征、模型、阈值、成本规则和冻结文件写入 `runs/ckpt-policy-freeze/` 之前，禁止读取 test 标签。",
        "- 9B 自评：AUC 约 0.78；单轨迹在允许约 1 个点损失时只能省约 6% 输入。",
        "- 0.6B 文本模型：AUC now 约 0.77、final 约 0.76；gain 头单轨迹约省 11.2%（零点损失菜单）到 13.2%（允许一点损失），但预算化 submit/continue/restart 没有超过原生策略。",
        "- 13,601 个决策点，1,830 个正例，300 个任务；",
        "- 任务分组 5 折 HistGradientBoosting，AUC=0.7892185481；",
        "2. **未来修复标签是否值得训练？** ext AUC=0.789 是正面信号，但还没有显示决策收益。需要一个预登记的控制器、严格的 task-group cross-fit、未知标签处理和 reviewer gate。",
        "随后做了一个严格的 train_ext→dev policy screen：future-repair AUC=0.8079，但把 `p_future_repair` 作为“低于阈值即可停止”的唯一控制信号时，解决率损失不超过 1 个点的最优点是不停止，输入节省 0%。因此该监督目前只能算可预测性证据，不能算方法效果。",
        "当前正确性头与 future-repair 头的组合网格（`p_now` 高或 `p_future_repair` 低即停止）在解决率不降时最多省 4.42% 输入，仍低于已有 0.6B 文本基线约 11%；作为负对照保留，不升级为主方法。",
        "1. **Crossed provenance**：20 任务、80/80 视图有效，但 17 个任务四视图全通过、3 个全失败，没有 code/test 分化。",
        "2. **Persistent failure witness**：8 任务、16 视图；7 个失败仍失败，1 个 rc1→rc4 仍失败，没有 failure→pass。",
        "3. **Repair-response**：4 任务、8 视图；current 与 code-reversed 公共命令返回码完全一致。",
        "同任务 group-wise patch stopping 也查到与 R4P 的直接重叠，不能作为新方法。P3 typed fixed instance 因可训练机会不足而关闭，效果标记为 `NOT_EVALUATED`，不能包装成负结果。",
        "初始 ACRB（artifact ledger + current/final + risk budget + abandon）经独立 review 判为 RETHINK：与 hazard / optimal stopping / Fail-Fast 的既有结构重叠，risk budget 没有可识别定义。",
        "同任务 group-wise patch stopping 也查到与 R4P 的直接重叠，不能作为新方法。P3 typed fixed instance 因可训练机会不足而关闭，效果标记为 `NOT_EVALUATED`，不能包装成负结果。",
        "5. **论文什么时候写？** 用户允许等待约十个月，但要求方法创新。当前不应调用最终论文改稿循环；只有在 result-to-claim 明确支持主方法后，才调用 `aris:auto-paper-improvement-loop` 做两轮 review–edit–compile，并编译 PDF。"
      ]
    },
    {
      "id": "J2",
      "path": "paper-jev/RESEARCH_BRIEF.md",
      "label": "当前研究目标、边界与收口流程",
      "date": "2026-09-28 更新",
      "sha256": "ee40fa2170f38d82c7a942512ddbeb17119f3a27e12431a38423878b07acf6f5",
      "excerpts": [
        "用户目标：完成一篇有可辩护方法创新、有真实实验增益的 JEV 论文；不接受以测量或诊断为最终贡献。时间允许持续研究（此前授权可等约十个月），不是立即包装现有结果。",
        "- 原 final-state G001/F1 路线维持 STOP；当前使用 checkpoint 线。",
        "- test 的 checkpoint 标签在方法、特征、阈值和评估代码冻结进 `runs/ckpt-policy-freeze/` 之前保持封存；自动判分及无标签特征提取允许。",
        "2026-09-28追加：修复数据上的counts-only GBDT三固定seed已完成、预测完全一致，独立复核仅支持开发集描述性点：failure-termination在160尝试上保持77次成功，输入省17.7944%、输出省12.4584%，一次误停与一次挽救；H的95%上界3.125%，没有总体安全/独立泛化保证。因此不能继续把旧40任务实验解读为“廉价计数信号学不出来”的一般结论。它是现有基线，不是新方法贡献。自评补充基线没有通过主H门槛，历史查询费用亦不完整；不得用generator-only节省冒充净节省。权威结论见`refine-logs/CLAIMS_FROM_COUNTS_BASELINE_20260928.*`和`refine-logs/CLAIMS_FROM_PROBE_BASELINES_20260928.*`。这些baseline-only判定不推进P3、test或论文门槛。P3 typed固定96任务实例已独立复核关闭为UNTRAINABLE_FIXED_READOUT：必需读出支持不足，停止效果未测，counters两头确有拟合。下一步完成已登记0.6B标签消融，再独立打磨新候选；不增补任务或修改提示救援本实例。",
        "2026-09-28追加：修复数据上的counts-only GBDT三固定seed已完成、预测完全一致，独立复核仅支持开发集描述性点：failure-termination在160尝试上保持77次成功，输入省17.7944%、输出省12.4584%，一次误停与一次挽救；H的95%上界3.125%，没有总体安全/独立泛化保证。因此不能继续把旧40任务实验解读为“廉价计数信号学不出来”的一般结论。它是现有基线，不是新方法贡献。自评补充基线没有通过主H门槛，历史查询费用亦不完整；不得用generator-only节省冒充净节省。权威结论见`refine-logs/CLAIMS_FROM_COUNTS_BASELINE_20260928.*`和`refine-logs/CLAIMS_FROM_PROBE_BASELINES_20260928.*`。这些baseline-only判定不推进P3、test或论文门槛。P3 typed固定96任务实例已独立复核关闭为UNTRAINABLE_FIXED_READOUT：必需读出支持不足，停止效果未测，counters两头确有拟合。下一步完成已登记0.6B标签消融，再独立打磨新候选；不增补任务或修改提示救援本实例。",
        "6. 方法与效果成立后更新论文主体、图表、引用和 PDF，并执行 ARIS 论断/引用/论文审查。用户本轮明确要求最后调用论文修改技能：必须执行 `aris:auto-paper-improvement-loop` 两轮评审—修改—重编译，保存改稿日志，不能仅交付结果摘要。"
      ]
    },
    {
      "id": "J3",
      "path": "paper-jev/refine-logs/NEW_METHOD_IDEAS_20260930_163221.md",
      "label": "尚未执行的新方法候选与评估建议",
      "date": "2026-09-30",
      "sha256": "cb8c064435ef468d7aa4f15d51f6b57e99de40a8a00c804c4383ab5796f605bd",
      "excerpts": [
        "**所有卡片中的 pilot 均未执行。** 数字是建议上限/预算，不是覆盖率或结果。建议进入任何效果 pilot 前共同冻结一个实际增益门槛，例如相对较强基线多省 **3 个百分点的 9B 输入**，输出不劣，满足同一个观察风险菜单，并公开 observer CPU/GPU/输入/输出成本。3 点是本轮建议，尚不是已授权修改的既有门槛。只改善 AUC、控制器成本、eligible 子集准确率、或事后挑一个阈值均不通过。",
        "**所有卡片中的 pilot 均未执行。** 数字是建议上限/预算，不是覆盖率或结果。建议进入任何效果 pilot 前共同冻结一个实际增益门槛，例如相对较强基线多省 **3 个百分点的 9B 输入**，输出不劣，满足同一个观察风险菜单，并公开 observer CPU/GPU/输入/输出成本。3 点是本轮建议，尚不是已授权修改的既有门槛。只改善 AUC、控制器成本、eligible 子集准确率、或事后挑一个阈值均不通过。",
        "40 个反复使用的 dev 任务不能被称作新的独立确认。小 pilot 也不能给总体 1% 安全保证。若希望使用 ext 内新任务留出，所有学习型对照都必须按同样的 task 划分重训；不能拿已经见过这些任务的旧 baseline 权重作为公平留出比较。",
        "## 1. 合约支持的变形关系组合执行",
        "## 2. 将已有参数化断言提升为有界符号义务",
        "## 3. 用官方确认的局部语义反例训练停止器",
        "- `d_i(τ)=Y_iT-Y_iτ`；`D=mean_i d_i`。挽救可以抵消 D，不能抵消 H。",
        "**结论：没有可宣布成立的新主方法。** 本轮给出八个可精确定义的机制候选，不把八个名字冒充八个已通过查新的创新。其中两个是尚可做极小可用性盘点的新执行假设；两个是高重合风险的训练数据协议；其余四个暴露出数据不可识别或直接先验重合。没有运行 pilot，没有读取 test 标签或任何 raw result 文件，没有改代码或启动任务。所有判断为 **same-family / provisional**。",
        "生成：`/root/new_method_ideation_20260930`；独立 execution lens 为 `gpt-6-astra / xhigh`，rule/data lens 为 `gpt-5.5 / xhigh`。两者都只 brainstorm，不看项目数据。fresh-context `gpt-6-astra / xhigh` jury 已完成八候选检查；其完整 verdict 与 metadata 位于 `.aris/traces/idea-creator/2026-09-30_new-method-ideas/`。全程 **same-family / provisional**。"
      ]
    },
    {
      "id": "J4",
      "path": "paper-jev/paper/sections/0_abstract.tex",
      "label": "旧final-state稿摘要（不是当前方法）",
      "date": "2026-09-24 PDF版本",
      "sha256": "883c4cbafe2a72e4b2c1964f78384f7b67bf660fe4c4f6802dd3662a7cd17081",
      "excerpts": [
        "Choosing the next action after a coding agent produces a patch requires distinguishing current correctness from the outcome of further work. We investigate whether matched executions can train a compact, non-generative controller under a fixed repair model. A three-seed CPU study trains a 0.6B current-correctness predictor, reaching 77.1\\% mean accuracy on 32 development states from eight tasks. A 108-execution pilot on 12 final-candidate training states yields no repair of the three initially unsuccessful candidates; capped restart frequently damages correct patches. In a separately frozen diagnostic enriched for 12 initially unsuccessful training tasks, 83 of 84 executions are judged. Public-test feedback repairs one task at both seeds, but no pair of actions exhibits replicated opposite preferences across tasks. The prespecified feasibility gate fails. These results establish an executable collection protocol but do not establish sufficiently task-diverse intervention gains for the planned utility training. An action-utility model and independent test-set evaluation remain outstanding.",
        "Choosing the next action after a coding agent produces a patch requires distinguishing current correctness from the outcome of further work. We investigate whether matched executions can train a compact, non-generative controller under a fixed repair model. A three-seed CPU study trains a 0.6B current-correctness predictor, reaching 77.1\\% mean accuracy on 32 development states from eight tasks. A 108-execution pilot on 12 final-candidate training states yields no repair of the three initially unsuccessful candidates; capped restart frequently damages correct patches. In a separately frozen diagnostic enriched for 12 initially unsuccessful training tasks, 83 of 84 executions are judged. Public-test feedback repairs one task at both seeds, but no pair of actions exhibits replicated opposite preferences across tasks. The prespecified feasibility gate fails. These results establish an executable collection protocol but do not establish sufficiently task-diverse intervention gains for the planned utility training. An action-utility model and independent test-set evaluation remain outstanding.",
        "Choosing the next action after a coding agent produces a patch requires distinguishing current correctness from the outcome of further work. We investigate whether matched executions can train a compact, non-generative controller under a fixed repair model. A three-seed CPU study trains a 0.6B current-correctness predictor, reaching 77.1\\% mean accuracy on 32 development states from eight tasks. A 108-execution pilot on 12 final-candidate training states yields no repair of the three initially unsuccessful candidates; capped restart frequently damages correct patches. In a separately frozen diagnostic enriched for 12 initially unsuccessful training tasks, 83 of 84 executions are judged. Public-test feedback repairs one task at both seeds, but no pair of actions exhibits replicated opposite preferences across tasks. The prespecified feasibility gate fails. These results establish an executable collection protocol but do not establish sufficiently task-diverse intervention gains for the planned utility training. An action-utility model and independent test-set evaluation remain outstanding."
      ]
    }
  ]
}