效度 · 冻结先于结果——每条主张的密码学顺序链，可自行验证
预注册证明链 Proof Chain
每个主张携带一条四环链：①预注册文档（先入库）→②config_committed 账本行（append-only，先于任何样本外指标观测）→③冻结结果目录（同签名重跑逐位一致）→④裁决。数据逐位来自 committed 面板（evidence_matrix / knowledge_shelf / ledger_audit），零手抄。
Phase B
② ledger
#28 · 2026-07-28
confirmatory:first · sig 17245a75
③ frozen
17245a75d2d4cd17…
runs/results/<sig>/ · bit-identical reruns
④ verdict
-0.0008
95% CI [-0.0106, 0.0090]
Phase C
② ledger
#30 · 2026-07-28
confirmatory:first · sig a7fdb48f
③ frozen
a7fdb48f5942fae1…
runs/results/<sig>/ · bit-identical reruns
④ verdict
-0.0065
95% CI [-0.0195, 0.0066]
Phase D
② ledger
#34 · 2026-07-29
confirmatory:first · sig d3158063
③ frozen
d31580630ff35326…
runs/results/<sig>/ · bit-identical reruns
④ verdict
-0.0030
95% CI [-0.0137, 0.0078]
Phase E1
② ledger
#37 · 2026-07-29
confirmatory:first · sig ef321e9e
③ frozen
ef321e9ee8088046…
runs/results/<sig>/ · bit-identical reruns
④ verdict
-0.0028
95% CI [-0.0115, 0.0059]
Track C（时序 confirmatory）
② ledger
#49 · 2026-08-05
confirmatory:first · sig e14b9d44
③ frozen
walk-forward
71-month walk-forward OOS
④ verdict
-0.0088
95% CI [-0.0336, 0.0159]
自行验证（四环各一条命令）
# ① prereg doc predates the run — check its commit history
git log --follow --format='%ci %h' -- docs/phase-b-preregistration.md | tail -1
# ② the append-only ledger row (Phase B example, row #28)
sed -n '28p' runs/ledger.jsonl | python -m json.tool
# ③ the frozen config sha in that row must equal the results dir name
ls runs/results/ | grep 17245a75
# ④ evidence artifact hashes are pinned in the committed matrix
python -c "import json;print(json.load(open('web/src/data/aionis/evidence_matrix.json'))['artifacts'][0])"账本 runs/ledger.jsonl 随仓库入库且 append-only：git 历史本身可证明行未被回改；证据工件 sha256 在 committed 矩阵中钉死。
one command: uv run python scripts/verify_claim.py
这正是临床试验预注册文化在量化研究里的工程化对应物：终点冻结先于揭盲。同签名换配置=新账本行，绝无静默覆盖。