This program's measurement doctrine: every number the platform plots must be backed by a stored, leak-free, pre-registered artifact — or it does not ship with the word "validated." Backfilled and prospective cohorts are always reported separately (ruling A6). A bad number honestly measured is the product; a good number laundered through imprecise language is a liability. The scorecards below are the first real walk-forward track record for the Cycle Intelligence suite — 12,787 PIT month-end stamps across US sectors, country ETFs and China Shenwan sectors, 2010–2026, price basis (epoch: price_v1 / zz14_v0). 本程序的测量原则：平台展示的每一个数字，必须有存储的、无前视偏差的、预注册的文物作支撑——否则不得使用"已验证"一词。 回测与实盘队列始终分开报告（规则A6）。 诚实测量出的差数据就是产品；经过模糊语言掩饰的好数据才是负债。 以下记分卡是周期智慧套件的首份真实前向跟踪记录—— 涵盖美国板块、国家ETF及中国申万板块，共12,787个逐点时间戳， 2010–2026年，价格基准（版本：price_v1 / zz14_v0）。
The new cone half-width = quantile(|timing_err|, 0.80) from the realized turn-timing error distribution. Headline multipliers are dominated by chronically overdue projections; the forward-only multiplier is the number to watch once a gated projection model replaces the current median-half-cycle engine (D5). 新锥形半宽 = 实测拐点时间误差分布的 quantile(|时间误差|, 0.80)。 整体倍数受慢性过期推演影响偏大；一旦门控推演模型取代当前中位半周期引擎（D5），请关注仅前向倍数。
BH-FDR applied within each gate family (q=0.10). A gate whose success threshold, judging data, or FDR family is edited after data has been seen is void — the edit itself is the finding. All FAILED gates are rendered as FAILED, not omitted. 本-霍克斯伯格FDR在每个门槛族内应用（q=0.10）。若门槛成功阈值、判据数据或FDR族在数据可见后被修改，则该门槛作废——修改行为本身即为发现。所有失败门槛均如实展示为失败，不予省略。
| Gate门槛 | Family族 | Claim主张 | Status状态 | Result / Notes结果/备注 |
|---|---|---|---|---|
| KG-1 | keystone | Position deciles carry forward drawdown-adjusted signal | FAIL失败 | NO-EDGE — every decile return-gap CI straddles zero at 21/63/126d |
| KG-2 | keystone | Phases carry forward drawdown-adjusted signal | PASS通过 | REFINE — Peak→shallower DD [+1.2%,+5.0%]; Trough→deeper DD [−10.0%,−1.9%] at 63d. Decays post-2018 (walk-forward fragility) |
| KG-3 | keystone | LADDER inversion (low-pos/DECLINE > high-pos/FRESH-BUY on dd-adj) | FAIL失败 | INCONCLUSIVE — all 9 era×horizon cells straddle zero; point estimate leans opposite direction |
| KG-4 | keystone | Signal (BUY/SELL) precedes its promised move | FAIL失败 | NO-EDGE — signal cells straddle 0 on both return and drawdown channels |
| KG-5 | keystone | Walk-forward stability of KG-1/2 effects | FAIL失败 | FAIL — Phase DD signal significant pre-2018, straddles zero post-2018 |
| CC-1 (sector_cycles) | calibration | Sector cycles cone band is calibrated | FAIL失败 | MISCALIBRATED — empirical 0.188 vs nominal 0.80 [CI 0.154,0.227]; recal multiplier 8.19× |
| CC-1 (country_cycles) | calibration | Country cycles cone band is calibrated | FAIL失败 | MISCALIBRATED — empirical 0.283 vs nominal 0.80 [CI 0.259,0.309]; recal multiplier 3.57× |
| CC-1 (china_sector_cycles) | calibration | China sector cycles cone band is calibrated | FAIL失败 | MISCALIBRATED — empirical 0.359 vs nominal 0.80 [CI 0.327,0.391]; recal multiplier 3.12× |
| CC-2 (sector_cycles/signal) | calibration | Sector cycles signal labels are honest (Brier skill > base rate) | FAIL失败 | FALSIFIED — skill −1.538, hit 0.430 vs base 0.659 (n=642) |
| CC-2 (sector_cycles/stance) | calibration | Sector cycles stance is honest (Brier skill > base rate) | FAIL失败 | FALSIFIED — skill −1.309, hit 0.481 vs base 0.659 (n=1257) |
| CC-2 (country_cycles/signal) | calibration | Country cycles signal labels are honest | FAIL失败 | FALSIFIED — skill −1.198, hit 0.460 vs base 0.565 (n=2075) |
| CC-2 (country_cycles/stance) | calibration | Country cycles stance is honest | FAIL失败 | FALSIFIED — skill −1.191, hit 0.462 vs base 0.565 (n=3900) |
| CC-2 (china_sector_cycles/signal) | calibration | China sector cycles signal labels are honest | FAIL失败 | FALSIFIED — skill −1.159, hit 0.460 vs base 0.507 (n=1677) |
| CC-3 (sector_cycles) | turn_pr | Sector cycles turn P/R factual (not circular) | FAIL失败 | FALSIFIED — prec 0.075 [0.054,0.103], rec 0.289 [0.214,0.379]; Wilson-lo far below 0.5 (n_eff=422) |
| CC-3 (country_cycles) | turn_pr | Country cycles turn P/R factual | FAIL失败 | FALSIFIED — prec 0.109 [0.093,0.128], rec 0.255 [0.220,0.294] (n_eff=1191) |
| CC-3 (china_sector_cycles) | turn_pr | China sector cycles turn P/R factual | FAIL失败 | FALSIFIED — prec 0.230 [0.203,0.260], rec 0.369 [0.329,0.411] (n_eff=810) |
| CL-1 | collinearity | state_score and pos_osc are redundant (near-collinear price transforms) | PASS通过 | CONFIRMED — rho(state_score, pos_osc) = −0.968; VIF(pos_osc)=29.8, VIF(state_score)=25.8 |
| CL-2 | collinearity | ≥1 leg carries independent risk-channel information | PASS通过 | CONFIRMED — 4 risk-channel survivors: trend_pass_f, mom_score, rs_63d_f, vol_pctile (63d) |
| CL-3 | collinearity | Dimension reduction justified (5 PCs explain ≥90%) | PASS通过 | CONFIRMED — 5 of 8 PCs explain 91.2% of variance |
| HZ-up-1m | hazard | Peak-hazard 1-month beats Kaplan-Meier baseline | ACCRUING积累中 | — |
| HZ-dn-1m | hazard | Trough-hazard 1-month beats KM | ACCRUING积累中 | — |
| BC-1 | calibration | LADDER_SCORE / tier cuts are earned, not asserted | ACCRUING积累中 | — |
| DL-1 | decision | Hazard cone earns its place over the IQR band | ACCRUING积累中 | — |
| LL-A | leadlag | Some ordered pair's lagged Δphase-position leads | PASS通过 | PASS — 136 of 8,253 pair×lag tests survive BH-FDR; top-20 frozen |
| LL-B | leadlag | Knowing the leader's confirmed turn improves the follower's OOS hazard | FAIL失败 | NO-GO — rel improvement +0.029% (bar ≥2%), CI₉₀ [−0.26%,+0.29%] includes 0, 3/9 year-blocks positive (bar ≥6/9). STOP: interaction layer not built. Sync gauge shipped. |
Pooled panel: 12504 PIT monthly stamps (2010–2026). State_score × pos_osc ρ=−0.968 (VIF~30): confluence was ONE price signal in three coats. The de-duplicated feature set is now binding for all downstream hazard and calibration waves. 合并面板：12504 个逐点月度时间戳（2010–2026）。state_score × pos_osc ρ=−0.968（VIF~30）：汇聚度实为一个价格信号披着三件外衣。去重后特征集现对所有下游风险与校准波次具有约束力。
| Experiment实验 | Status状态 | Come-back date复查日期 | Next step下一步 |
|---|---|---|---|
|
Sector Central Grader (Forward Return + Hit-Rate)
Forward returns + hit-rate + rank-IC of US Sector Central calls (basket + sector, tiered Constructive/Neutral/Reduce/Cautious) at 21/63/126 trading-day horizons.
|
ACCRUING积累中 | 2026-07-28 | Monitor forward IC stability, flag conviction-return inversions; use for sector rotation validation. |
|
Anticipation Engine Forward Cone (Phase-0 direction skill)
Multi-horizon conditional forward-return quantile cone + drawdown distribution; short-horizon direction measured coin-flip (Brier ~0, TOSS-UP); legs not passing Phase-0 are display-only.
|
PROVEN已证实 | 2026-07-15 | anticipation_phase0.py publishes gate verdicts -> anticipation.py activates/deactivates legs; direction stays TOSS-UP where Brier~0. Output site/anticipation.json. |
|
Cycle PIT Backfill — W2.3 price-basis stamps
Price-basis (W2.2 epoch) PIT month-end stamps for 11 US sectors + 24 country ETFs + 7 aggregates + 31 China Shenwan sectors. Provenance=backfilled; graders join via engine.grading_stats.load_graded_log(). Manifest at data/
|
ACCRUING积累中 | 2027-05-01 | W2.4 graders consume this via load_graded_log(); re-run on basis_version bump. |
|
US Sector Cycles — Turn Precision/Recall (W2.4)
Turn Precision/Recall vs independent realized-extrema truth (ruling A6)
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
US Sector Cycles — Cone Coverage (W2.4)
Cone coverage vs 0.80 nominal via grading_stats.cone_coverage
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
US Sector Cycles — Reliability / Brier (W2.4)
Directional Brier skill (signal/stance) vs per-instrument base rate over 63d
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
Country Cycles — Turn Precision/Recall (W2.4)
Turn Precision/Recall vs independent realized-extrema truth (ruling A6)
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
Country Cycles — Cone Coverage (W2.4)
Cone coverage vs 0.80 nominal via grading_stats.cone_coverage
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
Country Cycles — Reliability / Brier (W2.4)
Directional Brier skill (signal/stance) vs per-instrument base rate over 63d
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
China Sector Cycles — Turn Precision/Recall (W2.4)
Turn Precision/Recall vs independent realized-extrema truth (ruling A6)
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
China Sector Cycles — Cone Coverage (W2.4)
Cone coverage vs 0.80 nominal via grading_stats.cone_coverage
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
China Sector Cycles — Reliability / Brier (W2.4)
Directional Brier skill (signal/stance) vs per-instrument base rate over 63d
|
ACCRUING积累中 | 2027-01-01 | Monthly re-grade via scripts/grade_promises.py in cycle-calibration.yml. LIVE cohort promotes when n_eff ≥ 40. If gate clears, badge moves from ACCRUING to MEASURED. |
|
Market State — hand weights vs equal-weight Brier (h21 drawdown-concordance)
Pre-registered measurement: do market_state's hand weights (trend .24/risk .18/vol .16/breadth .16/liquidity .14/stress .12) beat equal-weight-renormalized on h21 drawdown-concordance (any_dd5_within_h21 Brier skill vs intercept-only base rate)? Primary metric: Brier skill score for hand-weighted raw_score vs equal-weight counterfactual recomputed from the SAME logged per-component scores. N floor: 250 trading days of component-logged forward_log.jsonl entries. Verdict printed either way — do-no-harm gate: if hand weights fail (Brier_hand < Brier_equal OR fp_hand > fp_equal), print 'equal-weight baseline meets or beats hand weights', no reweighting. Component history does not exist pre-2026-07-04 (4 entries, no components logged); logger patched in W3 PR2 to accrue per-component scores nightly from today.
|
ACCRUING积累中 | 2027-05-01 | Return 2027-05-01: load forward_log.jsonl rows with 'components' key + graded, compute Brier skill hand vs equal-weight, print verdict either way, update this registry note with the outcome. |
|
CN-SYS Falsifier Ledger — first calibration read (90d phase prints)
Each nightly china_cycle_phase print embeds machine-checkable falsifiers (CN-SYS-R5); the next run grades them into data/china_cycle_phase/falsifier_ledger.parquet. This is not a study — auto-grading accrues calibration statistics that inform a future pre-registered promotion review. First meaningful read requires ~90 trading days of phase prints.
|
ACCRUING积累中 | 2026-10-08 | Return 2026-10-08: check falsifier_ledger.parquet row count and calibration split; report held/fired/indeterminate rates per falsifier_id; no promotion authority at this read. |
|
CN-SYS Limit-tape x Lifecycle-board Confluence — detection print (no verdicts)
Limit-state tape (data/china_microstructure/limit_tape.parquet, 2011→ backfill + nightly appends) cross-referenced with the lifecycle-board (china_alpha RIPENING/ENTRY/RAN shelves + W8 microstructure chips). Detection print only — no promotion verdicts at this read. Checks whether limit-up breadth regime and lifecycle shelf co-occur in patterns worth registering for a future pre-registered study.
|
ACCRUING积累中 | 2026-09-15 | Return 2026-09-15: read limit_tape post-backfill nightly append count; cross-tab limit_up_breadth_pct regimes vs lifecycle shelves; flag candidate pattern for potential pre-registration; no promotion at this read. |
Each row is an adjudicated finding stored in data/cycle_pattern/truths.jsonl.
Null findings are guardrails: they rule out approaches that have been tried and found to carry no edge,
so they should not be rebuilt. Displaying them prominently is the honesty contract.
每一行均为存储于 data/cycle_pattern/truths.jsonl 的已裁定发现。
零效应发现是护栏：它们排除了已尝试且无优势的路径，不应被重建。
将其显著呈现正是诚实契约的体现。
| ID编号 | Statement陈述 | Status状态 | Effect class效应类别 | PIT classPIT类别 | Next review下次复查 |
|---|---|---|---|---|---|
| CPI-001 | Cycle position deciles carry no forward drawdown-adjusted return signal at 21-, 63-, or 126-day horizons on the membership-free US-sector and country-ETF universe (8,344 PIT stamps, 2005–2026): every decile return-gap CI straddles zero. | NULL | null | pit_pure | 2026-10-06 |
| CPI-002 | Cycle phase carries a risk-only, sign-inverted drawdown signal: Peak phase precedes shallower forward 63-day drawdowns (gap-CI [+1.2%, +5.0%]) and Trough phase precedes deeper drawdowns (CI [−10.0%, −1.9%]); this signal decays post-2018 and must be treated as regime-fragile, never as a return lever. | DISPLAY | risk_only | pit_pure | 2026-10-06 |
| CPI-003 | The ladder inversion hypothesis — that low-position/DECLINE states outperform high-position/FRESH-BUY states on the drawdown-adjusted lens — is not confirmed on PIT data: all 9 era × horizon cells are INCONCLUSIVE and the full-sample point estimate leans the opposite direction. | NULL | null | pit_pure | 2026-10-06 |
| CPI-004 | The 1-month up-leg peak-hazard model beats the family-stratified KM prior on Brier score (gap +0.014, 90% CI [+0.007, +0.021], p=0.001) across 17 OOS year-blocks (14/17 positive); this is the only robust cell in the 6-cell hazard family. | DISPLAY | risk_only | mixed | 2026-10-06 |
| CPI-005 | Down-leg trough-hazard beats KM at 1m (gap +0.014, 90% CI [+0.003, +0.025]) and marginally at 3m (gap +0.008, CI [+0.001, +0.016]) and 6m (gap +0.004, CI [+0.001, +0.008]); down-side hazard is horizon-persistent but weakens at longer windows and must not be wired to position sizing. | DISPLAY | risk_only | mixed | 2026-10-06 |
| CPI-006 | Up-leg hazard at 3-month and 6-month horizons shows no skill beyond the family-stratified KM prior: up/3m CI [−0.000, +0.014] touches zero (p=0.061); up/6m gap ≈ 0 (p=0.52). Both ship PRIOR. | NULL | null | mixed | 2026-10-06 |
| CPI-007 | After vol-residualizing forward max-drawdowns, no ladder state × family × horizon cell survives BH-FDR (q=0.10): 2 nominal hits of 48 cells are consistent with chance; every state ships risk_size_mult=1.0. | NULL | null | pit_pure | 2026-10-06 |
| CPI-008 | 136 in-sample FDR-surviving cross-lagged pair correlations yield zero out-of-sample Brier improvement (rel improvement 0.029%, 90% CI spanning zero, 3/9 year-blocks positive): the lead-lag interaction layer is a NO-GO and must not be built on this evidence. | NULL | null | pit_pure | 2026-10-06 |
| CPI-009 | 39 of 240 phase × quad × family cell combinations show forward-return or vol-residualized drawdown CIs excluding the phase-pooled baseline, with the strongest signal at cn_sector|Peak|Q1 (shrunk 63d return +14.3%, CI [+7.3%, +19.3%]); ALL are revision_optimistic until a PIT-correct macro-regime spine is built. | CANDIDATE | positive | revision_optimistic | 2026-10-06 |
| CPI-010 | Turn-projection precision is falsified against the pre-registered 0.5 bar: sector_cycles 0.075 [0.05, 0.10], country_cycles 0.109 [0.09, 0.13], china_sector_cycles 0.230 [0.20, 0.26] — all Wilson lower bounds far below 0.5. | NULL | null | pit_pure | 2026-10-06 |
| CPI-011 | Cycle projection cones are severely miscalibrated toward too-tight: empirical coverage 0.188/0.283/0.359 vs 0.80 nominal; recalibration multipliers are 8.19× (sector), 3.57× (country), 3.12× (china sector). | DISPLAY | structural | pit_pure | 2026-10-06 |
| CPI-012 | Directional signal and stance labels carry negative Brier skill vs the instrument base rate over 63 days: sector −1.538 / −1.309, country −1.198 / −1.191, china −1.159 — the labels do not beat an always-predict-majority-direction coin on the backfill cohort. | NULL | null | pit_pure | 2026-10-06 |
| CPI-013 | Basket cycle histories are hindsight-curated (membership frozen at a single curated date, pit=False per sector_cycles.py:409): any backtest or grading that uses basket series is survivorship-biased and MUST be labeled descriptive only until a membership-freeze audit creates a point-in-time membership archive. | DISPLAY | structural | pit_pure | 2026-10-06 |
| CPI-014 | Forward-log cohorts are too young to overrule backfill verdicts: sector_cycles and country_cycles have 2 unique live dates, china_sector_cycles has 5; a minimum of 40 matured observations is required before any live badge can displace a backfill finding. | DISPLAY | structural | pit_pure | 2026-10-06 |
| CPI-015 | Cycle narrative and DNA files (cycle_dna.json, narratives.json, leg_context.json) are mechanism-description material and CANNOT serve as statistical evidence for graded truths until they are normalized into a machine-readable format and joined to outcome-labeled forward-return data. | DISPLAY | structural | pit_pure | 2026-10-06 |
| CPI-016 | PHASE-CLOCK-1 (§18): The 6-state phase-clock classifier (capitulation/basing/early_expansion/late_expansion/rolling_over/early_contraction, FROZEN thresholds 20/80) does NOT provide a consistent ΔBrier improvement over the age-only family KM prior at h=6m on the 2024+ embargoed OOS window: only 1 of 3 families showed positive dBrier on post-2010 OOS (falsifier phase_clock_no_lift fires at <2/3 threshold). Phase-clock states remain available as descriptive display; all cards ship the KM prior (no phase-conditioned default). | NULL | null | pit_pure | 2026-10-01 |
| CPI-017 | HAR-1 (§18): Historical-analog kNN retrieval over normalized completed half-cycles (k=6, era-cap=2) evaluated against the BACKTEST cohort OOS window (end_date>=2024-01-01). Overall verdict: promoted_null. Per-family: us_sector: promoted_null, gap_vs_km=0.877415, CI90=[-0.693663, 2.881158], gap_vs_shuffle=1.848006, coverage=0.5, n_turns=26; country: promoted_null, gap_vs_km=1.548107, CI90=[0.875633, 2.165431], gap_vs_shuffle=0.138679, coverage=0.3924, n_turns=79; cn_sector: FAIL, gap_vs_km=1.905695, CI90=[1.377511, 2.530548], gap_vs_shuffle=1.15374, coverage=0.4453, n_turns=137. Analog-shuffle null (within-era permutation of retrieved analogs, 200 draws) is the primary gate: if HAR does not beat its own shuffle, verdict=promoted_null (retrieval theater). TRAJECTORY-SHAPE INERTNESS: at query time (age_m=0.5) n_pts=1, so HAR retrieves on fingerprint+family only — not trajectory shape. Any KM/shuffle gap reflects fingerprint+family matching, not shape similarity. Macro fingerprint revision_optimistic (P-D5-1, quad/liquidity not PIT-vintaged). | NULL | null | revision_optimistic | 2027-01-07 |
| cycle_truth_cn_downturn_broken_trend_tail_null_v1 | The CN Downturn × broken-trend deep vol-adjusted 63d tail claim (batch-1 lead, ex-CPI-020) does NOT survive the §15 within-family re-test: gap vs CN's own Downturn baseline −0.0350 (CI95 [−0.0688, −0.0021], boot p=0.04, n=145 months, era-stable) — directionally persistent at ~60% smaller magnitude than the cross-family estimate (−0.0597; §14 baseline confound), and killed by BH-FDR q=0.10 across the declared 135-cell family. The risk-relevant survivor is the SAME cell on turn_event_3m: broken-trend CN downturns show a turn DEFICIT (−0.086, BH-pass) — longer downturns, unproven deeper tails. | NULL | null | pit_pure | 2027-01-06 |
| cycle_truth_falsosc_osc_covariate_null_v1 | The monthly oscillator covariate family (MACD hist/sign/slope + StochRSI K&D computed on monthly-resampled close) adds NO out-of-sample Brier skill beyond the shipped W4.2 feature set at the 6-month horizon for EITHER direction that satisfies the §18 kill-switch gate. Kill-switch FALS-OSC fired: up/6m CI₉₀ includes 0 (ΔBrier +0.0037, CI₉₀ [−0.0037, +0.0104], boot_p=0.211). NOTE: down/6m separately shows PASS under §18 criteria (ΔBrier +0.00276, CI₉₀ [+0.00138, +0.00445], boot_p=0.0012, bh_pass=True) — a genuine incremental signal in the down regime — but the kill fires on EITHER direction failing. Columns remain in state_monthly lake; model design reverts. | NULL | null | pit_pure | 2027-01-07 |
| cycle_truth_ft1_breadth_hazard_null_v1 | The FT-1 family-breadth block (fam_pct_above_200d, fam_pct_above_50d, breadth_div_own, breadth_thrust_3m) does not add out-of-sample turn-hazard skill beyond the shipped W2.5-bound feature set under the W4.2 harness (0/6 cells pass; the block significantly DEGRADES down/1m, dBrier -0.0056 CI90 [-0.0099,-0.0016]). Up-direction point estimates lean positive but are unearned. | NULL | null | pit_pure | 2027-01-06 |
| cycle_truth_ft2_credit_hazard_null_v1 | The FT-2 credit/curve block (hy_oas_pctile, hy_oas_d63, curve_10y3m) does not add out-of-sample turn-hazard skill beyond the shipped W2.5-bound feature set — 0/6 cells pass and 4/6 are SIGNIFICANTLY HARMFUL (all up-cells: dBrier -0.0098..-0.0118 with CI90 fully below 0). Additive time-only macro covariates overfit regime co-occurrence in the pooled hazard logistic. Program synthesis after 2 FT batches (18 cells, 0 passes, 5 harmful): the shipped hazard model's parsimony is load-bearing; additive-feature FT trials on the pooled hazard are suspended. | NULL | null | pit_pure | 2027-01-06 |
| cycle_truth_ft4_structure_hazard_null_v1 | The FT-4 cross-entity structure block (sync_family, phase_breadth_late, phase_breadth_early, pos_dispersion) as features on MEMBER-level turn hazard adds no out-of-sample skill beyond the shipped feature set (0/6 cells pass). Does NOT cover the index-level target: IX-1 (index turn hazard from constituent structure) remains an open, separately-gated question. | NULL | null | pit_pure | 2027-01-06 |
| cycle_truth_ix1_index_transfer_null_v1 | IX-1 (§17): the member-trained W4.2 hazard model does NOT transfer to index-level entities (SPY + 7 blocs) against each index's own age-pooled KM under the frozen gate — 0/4 cells (CI90 positive AND BH q=0.10 AND years+ >= 9). Up cells show no earned skill (1m dBrier +0.0099, CI90 [-0.0056,+0.0247], 8/14 years; 3m -0.0018, straddles). Down cells show REAL pooled improvement (1m +0.0335, CI90 [+0.0084,+0.0590], p 0.011; 3m +0.0290, CI90 [+0.0129,+0.0455], p 0.001; both BH-pass) that FAILS the sign-stability leg (5/13 and 7/13 vs bar >=9): the gain is year-concentrated (2021 year-mean +0.31/+0.25 dominates; 2020 harmful -0.16) — episodic, not reliable. Scoped to THIS transfer recipe: member-fit L2 logistic + member train-fold standardization + member-fit per-fold PAV, no index-row fitting, no index covariates (sync/phase-breadth/dispersion RESERVED, unused). NOT a claim that index-level turn hazards are unpredictable. | NULL | null | pit_pure | 2027-01-07 |
| cycle_truth_lattice1_confirmatory_and_baseline_confound_v1 | Lattice batch 1 (135 PIT-pure shrunken cells, no quad) independently reproduces era-stable phase-conditional structure in turn arrival, persistence and vol-adjusted drawdown (48/135 cells clear the frozen §14 gate; KG-2 raw-DD direction reproduced). HOWEVER the cross-family phase-pooled baseline conflates family base-rate offsets with phase effects (CN cells mirror-signed vs US/country in the same phase). Per-cell gaps are confirmatory context only, never per-cell edges; lattice batch 2 must use within-family baselines (to be frozen as §15). | DISPLAY | structural | pit_pure | 2026-10-06 |
| cycle_truth_lattice2_within_family_structure_v1 | Lattice batch 2 (same 135 PIT-pure cells as batch 1, WITHIN-FAMILY baselines): 27/135 promote under the frozen §15 gate. Phase-conditional turn-arrival and persistence structure is real WITHIN every family (not a family-composition artifact): turn arrival is phase-graded (Trough/Recovery/Downturn elevated, Expansion/Peak deferred), Recovery is fragile in all three families (persist −0.17 us / −0.17 country / −0.25 cn), Peak persists. Two vol-adjusted risk cells: Peak carries SHALLOWER rdd_63d tails within country (+0.049) and CN (+0.071) — KG-2/CPI-002 direction in stricter form. Trend integrity conditions turn arrival within family-phase: broken-trend CN downturns grind (turn deficit −0.086) while intact-trend ones resolve (+0.160). Confirmatory/structural — phase labels share mechanics with turn labels; these are NOT tradeable edges. | DISPLAY | structural | pit_pure | 2026-10-06 |
| cycle_truth_tr1_next_phase_softmax_skill_v1 | TR-1 (§16, the program's first gate-passing cells): a pure-numpy L2 softmax over EXISTING PIT-pure hazard-panel columns (current-phase one-hot + the shipped W2.5-bound feature set; no new covariates, no calibration layer) beats the family-stratified Laplace(α=1) transition-matrix baseline on OOS multiclass Brier in 4/6 frozen cells: all three families at 1m (ΔBrier +0.0030 us_sector / +0.0055 country / +0.0066 cn_sector — 2.7%/4.8%/6.1% of a strong baseline, CI90 excl 0, years+ 11/13/14 of 14, BH q=0.10) and country at 3m (+0.0027, 12/14). NOT established at 3m for us_sector (+0.0005, CI straddles, 8/14) or cn_sector (+0.0014, CI straddles). Mechanism is model capacity, not new information: continuous within-phase state (pos_osc/slope/momentum/age) anticipates phase-boundary crossings the discrete matrix cannot see — hence the 1m concentration. A display-class next-phase probability (masterplan C4), NOT a tradeable edge. | DISPLAY | structural | pit_pure | 2026-10-06 |
These findings reached promoted_null status because the evidence — at the sample sizes and horizons listed — ruled out an edge.
The evidence file paths below are citations, not promises. A null holds until the falsifier conditions in the truth record are met.
以下发现已达到 promoted_null 状态，因为在所列样本量与时间窗口下，证据排除了优势的存在。
下方文件路径为引用，非承诺。零效应结论在其真值记录中的证伪条件成立前持续有效。
These are accrual clocks, not promises. The projected date to 40 unique stamps is a mechanical extrapolation of the observed cadence — it assumes the ledger continues filling at the same pace, which may not hold. 40 unique stamp dates is the minimum for the live cohort maturity gate (n_eff threshold for meaningful turn P/R). No edge is implied by reaching this milestone; it only means the gate can be re-evaluated. 这些是积累时钟，非承诺。达到40个唯一时间戳的预测日期，是对观测节奏的机械外推——假设台账以相同速度持续填充， 实际情况可能并非如此。40个唯一时间戳是实盘队列成熟度门槛的最低要求（拐点精准率/召回率的有效n阈值）。 达到此里程碑并不意味着存在优势；仅意味着该门槛可以重新评估。
| Ledger台账 | Rows行数 | IDs标的数 | Stamp dates时间戳数 | First / Last首/末日期 | Cadence节奏 | → 40 stamps→ 40时间戳 |
|---|---|---|---|---|---|---|
|
US Sector Cycles — forward log美国板块周期 · 前向日志
|
627 | 57 | 11 |
2026-07-02 2026-07-17 |
1.5d | 2026-08-29 |
|
Country Cycles — forward log国家周期 · 前向日志
|
341 | 31 | 11 |
2026-07-02 2026-07-17 |
1.5d | 2026-08-29 |
|
China Sector Cycles — forward log中国板块周期 · 前向日志
|
689 | 53 | 13 |
2026-06-26 2026-07-17 |
1.8d | 2026-09-03 |
|
US Sector Central — calls美国板块中枢 · 研判记录
|
795 | 57 | 14 |
2026-06-28 2026-07-17 |
1.5d | 2026-08-25 |
|
China Sector Central — calls中国板块中枢 · 研判记录
|
689 | 53 | 13 |
2026-06-26 2026-07-17 |
1.8d | 2026-09-03 |
The hazard model scores each cycle entity's probability of a turn within 1/3/6 months. Per-cell criterion: OOS Brier(model) < Brier(KM baseline), 90% CI excludes zero, BH-FDR q=0.10. Failing cells ship the KM prior — no model output, no invented confidence. Quad-conditioned cells are revision-optimistic (macro vintage fix pending, P-D5-1). 风险模型评估每个周期实体在1/3/6个月内发生拐点的概率。 单格标准：OOS Brier（模型）< Brier（KM基准），90% CI不含零，BH-FDR q=0.10。 未通过的格子使用KM先验——无模型输出，无捏造的置信度。 四象限条件格子处于修正乐观状态（宏观数据时间点修复待处理，P-D5-1）。
Where each artifact is produced vs consumed. Machine-detectable gaps are derived from parquet non-null counts. Gaps marked "curated check" were verified by manual inspection and are not machine-detectable. 每个文物的生产地与消费地。机器可检测的差距源自 parquet 非空计数。 标注"人工核查"的差距经人工审核，无法由机器自动检测。
Each cell shows whether the page currently uses this capability: yes / partial / no.
Values are sourced from a committed Python dict in scripts/build_measurement.py with a comment per cell citing the evidence file.
This matrix is hand-maintained until machine detection exists — it is honest about what exists today, not a roadmap.
每格显示该页面是否当前使用此能力：是 / 部分 / 否。
值来源于 scripts/build_measurement.py 中已提交的 Python 字典，每格附有引用证据文件的注释。
该矩阵为人工维护，直至机器检测可用——仅反映当前实际情况，非路线图。
| Page页面 | State export状态导出 | Outcome join结果关联 | Hazard adoption风险采纳 | Truth badge真值徽章 | NW export神经网络导出 | Live grader实盘评分器 |
|---|---|---|---|---|---|---|
| cycle.html | no | no | no | no | no | no |
| sector_cycles.html | partial forward_log | no | no col-not-rendered | no | no | no |
| country_cycles.html | partial forward_log | no | no col-not-rendered | no | no | no |
| markets.html | no | no | no | no | no | no |
| sector_central.html | no | no | no | no | no | no |
| sector_central_china.html | no | no | no | no | no | no |
| measurement.html | partial accrual-clocks | no | partial model-json | yes truth-ledger | no | no |
COVERAGE_MATRIX dict in scripts/build_measurement.py — curated, audited 2026-07-06. Each cell cites the evidence file in a Python comment.
来源：scripts/build_measurement.py 中的 COVERAGE_MATRIX 字典 — curated, audited 2026-07-06。每格在 Python 注释中引用了证据文件。
TIME vs BUILD: "accruing (time-starved)" means the grader is wired and correct — rows are simply too recent to have matured outcomes. These ledgers need clocks, not engineers. "needs grader (build)" means no grader exists yet. Do not conflate the two. 时间 vs 构建："积累中（时间不足）"表示评分器已接入且正确——行数据只是太新，尚未到期。这些账本需要的是时钟，而非工程师。"需要评分器（待构建）"表示尚无评分器。请勿混淆二者。
| Ledger key账本键 | n_logged已记录 | n_graded已评分 | Verdict裁定 | Starvation (TIME vs BUILD)缺失类型（时间 vs 构建） |
|---|---|---|---|---|
| qledger_claims | 19561 | 15151 | CLOSED已关闭 | closed / partial |
| qledger_grades | 15151 | 15151 | CLOSED已关闭 | closed / partial |
| board_ledger_hk | 71 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-08 (n=9, weekly cadence) |
| board_ledger_ca | 127 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-08 (n=40) |
| us_board_ledger | 1210 | 1210 | CLOSED已关闭 | closed / partial |
| signal_archive_track_record | 56378 | 56240 | CLOSED已关闭 | closed / partial |
| china_standout_track | 662 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-08 (n=240) |
| sector_cycles_forward_log | 627 | 1937 | CLOSED已关闭 | closed / partial |
| china_sector_cycles_forward_log | 636 | 5081 | CLOSED已关闭 | closed / partial |
| country_cycles_forward_log | 341 | 5769 | CLOSED已关闭 | closed / partial |
| breadth_divergence_forward_log | 96 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_forward_log | 16 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_intl_cn | 12 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-07-10 (n=5 at ~weekly cadence) |
| risk_radar_intl_hk | 11 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-07-10 (n=4) |
| risk_radar_intl_ca | 9 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-07-10 (n=4) |
| risk_radar_intl_kr | 1 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_intl_jp | 1 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_intl_tw | 1 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_intl_in | 1 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_intl_au | 1 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_intl_gb | 1 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| risk_radar_intl_ez | 1 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
| market_state_forward_log | 14 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-07-10 (n=5) |
| oracle_forward_ledger | 173 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-07-30 (n=173, quarterly grading window) |
| oracle_compounds_live_ledger | 53 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-07-30 (n=53) |
| oracle_reversion_forward | 0 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — seeded only after first reversion compound passes gauntlet |
| btc_override_ledger | 13 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-08-25 (n=5, monthly cadence) |
| foresight_log | 69 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-08-25 (n=34) |
| foresight_policy_calendar_ledger | 193 | 131 | CLOSED已关闭 | closed / partial |
| froth_fragility_log | 13 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-08-25 (n=5) |
| species_registry | 27 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — earliest maturity ~2026-09 (n=21, accruing) |
| species_antichase_shadow_ledger | 439 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — seeded when F3_ANTICHASE accrues first row |
| species_f1d_shadow_ledger | 439 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) — seeded when EI-F1D-RW accrues first row |
| bd_avoid1_ledger | 0 | 0 | GRADER-STARVEDGRADER-STARVED | accruing (time-starved) |
data/governance/grading_closure.json.
生成时间：2026-07-17T09:11:12.503081+00:00。来源：data/governance/grading_closure.json。
| Family族 | n rows行数 | Latest timestamp最新时间戳 |
|---|---|---|
| active_commodity_lev_phase0 | 1 | 2026-07-02T03:11:22 |
| alpha_grammar_confluence_v1 | 565 | 2026-07-05T13:53:25 |
| anticipation_phase0 | 1 | 2026-07-02T03:11:22 |
| baskets_abs_escape | 4 | 2026-07-03T11:25:25 |
| baskets_breadth_divergence | 2 | 2026-07-02T05:07:21 |
| baskets_conviction_demotion | 2 | 2026-07-03T11:25:27 |
| baskets_rank_dd_mod | 4 | 2026-07-03T11:25:28 |
| btc_onchain_dd_phase0 | 1 | 2026-07-02T03:11:22 |
| btc_vector_optimal_phase0 | 1 | 2026-07-02T03:11:22 |
| business_cycle_recession_signal | 2 | 2026-07-02T12:25:07 |
| c1b_commodity_flip_protective_phase0 | 1 | 2026-07-03T17:48:13 |
| capitulation_overlay_phase0 | 1 | 2026-07-02T03:11:22 |
| causal_scan | 110 | 2026-07-12T08:42:03 |
| china_validation | 3 | 2026-06-22T02:36:20 |
| coiled_ca_phase0 | 1 | 2026-07-03T19:37:01 |
| commodity_carry_phase0 | 1 | 2026-07-02T03:11:22 |
| commodity_tsmom_phase0 | 1 | 2026-07-02T03:11:22 |
| commodity_xsec_carry_phase0 | 1 | 2026-07-02T03:11:22 |
| commodity_xsec_mom_phase0 | 1 | 2026-07-02T03:11:22 |
| context_scan | 1 | 2026-07-08T06:02:12 |
| cortex | 5 | 2026-07-13T10:30:39 |
| credit_duration_verify_phase0 | 1 | 2026-07-02T03:11:22 |
| cross_asset_confirmation_phase0 | 1 | 2026-07-02T03:11:22 |
| cross_asset_phase0 | 1 | 2026-07-02T03:11:22 |
| crypto_voltarget_phase0 | 1 | 2026-07-02T03:11:22 |
| cycle_pattern_analog | 1 | 2026-07-07T02:00:02 |
| cycle_pattern_ft | 3 | 2026-07-07T09:49:57 |
| d2_cn_export_share_nowcast | 16 | 2026-07-07T14:04:02 |
| d2_cn_holder_sale_calendar | 12 | 2026-07-07T14:04:02 |
| d2_comment_letter_release | 8 | 2026-07-07T14:04:02 |
| d2_rates_calendar_flows | 14 | 2026-07-08T08:56:45 |
| d2_russell_deletion_overshoot | 1 | 2026-07-07T14:04:02 |
| dt_replication | 8 | 2026-07-06T23:35:47 |
| dt_replication_64y | 4 | 2026-07-07T00:02:10 |
| esx_appendix | 1 | 2026-07-05T07:36:12 |
| esx_decline_geometry | 5 | 2026-07-06T07:48:08 |
| esx_ev_blackout | 10 | 2026-07-05T11:40:59 |
| esx_fund_repair | 2 | 2026-07-06T07:15:43 |
| esx_htf_turn | 13 | 2026-07-06T07:44:15 |
| esx_htf_turn_dose | 3 | 2026-07-06T07:44:15 |
| esx_insider_sponsor | 12 | 2026-07-06T07:15:43 |
| esx_lq_bands | 1 | 2026-07-05T07:36:12 |
| esx_macro_release | 10 | 2026-07-06T08:15:24 |
| esx_null_competitors | 7 | 2026-07-05T10:16:03 |
| esx_pos_reset | 6 | 2026-07-06T08:15:24 |
| esx_ql_overlay | 11 | 2026-07-05T12:48:18 |
| esx_sponsorship | 2 | 2026-07-06T07:15:43 |
| esx_sq_phase0 | 1 | 2026-07-05T07:36:12 |
| esx_sub_x_turn | 3 | 2026-07-06T07:44:15 |
| esx_support_dose | 2 | 2026-07-06T07:15:43 |
| esx_ts_adx | 5 | 2026-07-05T11:40:03 |
| esx_underwater | 5 | 2026-07-06T07:48:08 |
| esx_ur_phase0 | 1 | 2026-07-05T07:36:12 |
| esx_vol_transition | 5 | 2026-07-06T07:48:08 |
| esx_washout_x_turn | 9 | 2026-07-06T07:44:15 |
| eth_vector_phase0 | 1 | 2026-07-02T03:11:22 |
| hk_canada_h3 | 11 | 2026-07-06T11:09:22 |
| hk_southbound_phase0 | 1 | 2026-07-02T03:11:22 |
| hyoas_z_timer_phase0 | 1 | 2026-07-02T03:11:22 |
| incremental_ic:fip_continuity | 4 | 2026-06-21T10:36:19 |
| incremental_ic:mom_12_1 | 4 | 2026-06-21T10:36:16 |
| incremental_ic:near_52w_high | 4 | 2026-06-21T10:36:18 |
| index_direction_phase0 | 1 | 2026-07-02T03:11:22 |
| insider_phase0 | 1 | 2026-07-02T03:11:22 |
| insider_phase1 | 1 | 2026-07-02T03:11:22 |
| intl_bridge | 23 | 2026-07-17T07:05:30 |
| intl_macro_sleeve_phase0 | 1 | 2026-07-02T03:11:22 |
| intl_tr_trend_phase0 | 1 | 2026-07-02T03:11:22 |
| intl_trend_overlay_phase0 | 1 | 2026-07-02T03:11:22 |
| long_hold.expect_drift | 1 | 2026-07-06T09:13:16 |
| long_hold.insider_sponsor_lh | 1 | 2026-07-06T08:45:01 |
| macro_tx | 1 | 2026-07-06T13:36:06 |
| mastermind_moderate_phase0 | 1 | 2026-07-02T03:11:22 |
| naaim_overlay_phase0 | 1 | 2026-07-02T03:11:22 |
| okx_retail_phase0 | 1 | 2026-07-02T03:11:22 |
| provisional_cn_blend | 1 | 2026-07-02T05:23:29 |
| provisional_fresh_ticks_cn | 5 | 2026-07-02T05:26:40 |
| provisional_fresh_ticks_us | 5 | 2026-07-02T05:17:32 |
| quad_nfci_phase0 | 1 | 2026-07-02T03:11:22 |
| regime | 23 | 2026-07-02T05:09:29 |
| replay | 5 | 2026-07-06T22:05:26 |
| residual_alpha_phase0 | 1 | 2026-07-02T03:11:22 |
| rf.cycle_pattern.ft_v0 | 1 | 2026-07-06T12:13:06 |
| rf.cycle_pattern.ft_v1 | 1 | 2026-07-06T22:04:11 |
| rf.cycle_pattern.ix_v0 | 1 | 2026-07-07T01:04:24 |
| rf.cycle_pattern.lattice_v0 | 1 | 2026-07-06T22:19:21 |
| rf.cycle_pattern.lattice_v1 | 1 | 2026-07-06T23:58:26 |
| rf.cycle_pattern.tr_v0 | 1 | 2026-07-07T00:27:45 |
| s10_margin_inflection | 3 | 2026-07-07T00:59:47 |
| s11_buyback_floor | 1 | 2026-07-07T01:21:46 |
| s13_reversal_sleeve | 2 | 2026-07-04T01:55:21 |
| s6_failed_fire_fuel | 2 | 2026-07-04T07:20:46 |
| sector_pulse_heat | 13 | 2026-07-04T00:50:42 |
| short_side | 24 | 2026-07-06T15:14:57 |
| signal_factory | 5 | 2026-06-21T15:04:20 |
| signal_foundry | 17 | 2026-07-17T10:01:09 |
| slf001_ftd_pressure | 6 | 2026-07-06T11:31:17 |
| slf006_auction_absorption | 12 | 2026-07-06T11:06:06 |
| slf048_wiki_attention | 24 | 2026-07-06T11:09:50 |
| slf051_cn_margin_impulse | 8 | 2026-07-06T11:07:33 |
| slf055_dealer_stress | 15 | 2026-07-06T11:37:27 |
| slf056_funding_tail | 14 | 2026-07-06T11:06:52 |
| stock_conviction_phase0 | 1 | 2026-07-02T03:11:22 |
| stock_personality_compat | 1 | 2026-07-07T17:56:55 |
| sue_insider_deep_phase0 | 1 | 2026-07-02T03:11:22 |
| thematic_rotation_phase0 | 1 | 2026-07-02T03:11:22 |
| top_picks_phase0 | 1 | 2026-07-02T03:11:22 |
| turn_of_month_phase0 | 1 | 2026-07-02T03:11:22 |
| unknown | 2 | 2026-07-07T13:10:00 |
| value_growth_phase0 | 1 | 2026-07-02T03:11:22 |
| vector | 5 | 2026-07-02T03:43:12 |
| w2031_kev_vendor_shock | 7 | 2026-07-06T11:32:57 |
| w2051_housing_hf | 8 | 2026-07-07T14:04:02 |
| w2096_nhtsa_defect | 6 | 2026-07-06T13:05:04 |
| w2104_cmdi_conditioning | 12 | 2026-07-07T14:04:02 |
| w2153_itc337 | 4 | 2026-07-07T14:04:02 |
| w3_bank_callreport_stress | 7 | 2026-07-08T07:59:44 |
| w4_multistate_gaming_tape | 6 | 2026-07-07T14:04:02 |
| w5_trade_size_capitulation | 8 | 2026-07-08T00:00:00 |
data/trial_ledger.jsonl (1220 rows total).
来源：data/trial_ledger.jsonl（共 1220 行）。
Pooled SUM is the declared_budget total across all registered experiments (SUM semantic). The TrialLedger uses a per-family max() basis — the two diverge by design. See RUL-5. 合并SUM为所有已注册实验的declared_budget总和（SUM语义）。TrialLedger使用每族max()基准——两者按设计存在差异。见RUL-5。
| Exp ID实验ID | Status状态 | Budget预算 | Registered注册时间 | Question (truncated)问题（截断） |
|---|---|---|---|---|
| exit_grid_v1 | reported已报告 | 15 | 2026-07-06 | For the production fire cohort (verdict_type=fire AND verdict_grade=True, n~49939), what did each frozen exit policy cost in foregone MFE versus save in avoided MAE relative to hold(126), and why did … |
| wait_grid_v1 | reported已报告 | 10 | 2026-07-06 | For the production fire cohort, what does waiting cost or save — how do entry outcomes change as fill delay rises 1→10 bars, at the ratified hold(21) anchor and hold(63)? Grid granularity: the 5-step … |
| disp_gate_1 | reported已报告 | 6 | 2026-07-06 | DISP-GATE-1 (L3_PREREG.md, frozen 2026-07-05): For the production fire cohort, do fires opened when the broad-universe cross-sectional dispersion regime is lean_out (low dispersion / high pairwise cor… |
| trim_grid_v1 | reported已报告 | 6 | 2026-07-06 | TRIM-GRID-1 (RUL-F3.5, PR-F3.3): On the production fire cohort (same as exit_grid_v1: verdict_type=fire AND verdict_grade=True, n~49939), do partial-trim scaled policies preserve more right-tail retur… |
data/rule_experiments/registry.jsonl. Pooled declared budget SUM = 37 (SUM across registrations; TrialLedger max()-basis is a separate number — see RUL-5).
来源：data/rule_experiments/registry.jsonl。合并声明预算SUM = 37（各注册之和；TrialLedger max()基准为另一数字——见RUL-5）。
Wilson CI is computed on independent date clusters (n_dates), not on the larger correlated n_obs. The pooled hit rate is projected onto n_dates: hits_cluster = round(hit_rate × n_dates), n = n_dates. This is the cluster-honest convention per the ticker-cluster time-confound law and the altdata_brain.py article3 convention. n_dates is shown beside every CI. Max n_dates across all families is currently 18 of the 25-date floor. Not yet calibrated — trust/accrual surface, not authority. 威尔逊置信区间基于独立日期簇（n_dates）计算，而非较大的相关n_obs。汇总命中率投影到n_dates：hits_cluster = round(hit_rate × n_dates)，n = n_dates。这是依据票据簇时间混淆法则和altdata_brain.py article3惯例的簇诚实约定。每个置信区间旁均显示n_dates。当前所有族中n_dates最大值为18，floor为25。尚未校准——信任/积累面板，非权威结论。
| Family族 | Horizon周期 | n_obs观测数 | n_dates (of 25 floor)n_dates（floor=25） | Hit rate命中率 | Wilson CI low (overlapping obs)威尔逊下界（重叠观测） | State状态 |
|---|---|---|---|---|---|---|
| altdata | 21d | 91 | 4 / 25 | 48.4% | 0.1500 (n_dates=4) | ACCRUING积累中 |
| altdata | 5d | 155 | 14 / 25 | 51.6% | 0.2680 (n_dates=14) | ACCRUING积累中 |
| china_news | 21d | 191 | 5 / 25 | — | — (n_dates=5) | ACCRUING积累中 |
| china_news | 5d | 806 | 12 / 25 | — | — (n_dates=12) | ACCRUING积累中 |
| cn_importance_v0 | 21d | 176 | 6 / 25 | — | — (n_dates=6) | ACCRUING积累中 |
| cn_importance_v0 | 5d | 856 | 18 / 25 | — | — (n_dates=18) | ACCRUING积累中 |
| cn_importance_v0_pit | 21d | 176 | 6 / 25 | — | — (n_dates=6) | ACCRUING积累中 |
| cn_importance_v0_pit | 5d | 856 | 18 / 25 | — | — (n_dates=18) | ACCRUING积累中 |
| cn_special_sits | 5d | 6 | 4 / 25 | — | — (n_dates=4) | ACCRUING积累中 |
| policy | 21d | 2 | 1 / 25 | 50.0% | 0.0000 (n_dates=1) | ACCRUING积累中 |
| policy | 5d | 11 | 3 / 25 | 40.0% | 0.0615 (n_dates=3) | ACCRUING积累中 |
| radar | 21d | 667 | 5 / 25 | 38.7% | 0.1176 (n_dates=5) | ACCRUING积累中 |
| radar | 5d | 3486 | 17 / 25 | 51.3% | 0.3096 (n_dates=17) | ACCRUING积累中 |
| us_importance_v0 | 21d | 4 | 1 / 25 | — | — (n_dates=1) | ACCRUING积累中 |
| us_importance_v0 | 5d | 3600 | 12 / 25 | — | — (n_dates=12) | ACCRUING积累中 |
| us_importance_v0_pit | 21d | 4 | 1 / 25 | — | — (n_dates=1) | ACCRUING积累中 |
| us_importance_v0_pit | 5d | 3600 | 12 / 25 | — | — (n_dates=12) | ACCRUING积累中 |
| whitehouse | 5d | 2 | 1 / 25 | 0.0% | 0.0000 (n_dates=1) | ACCRUING积累中 |
site/qledger/track_record.json (generated 2026-07-17T09:10:50.353695+00:00). RUL-6: only trust/accrual fields exposed; no composite, no escalation-eligible score. The 21d and 63d horizons have zero grades today.
来源：site/qledger/track_record.json（生成时间：2026-07-17T09:10:50.353695+00:00）。RUL-6：仅展示信任/积累字段；无复合分数，无可升级评分。当前21d和63d周期零评分。