Rank213 Evidence Map：先看这页，避免误读
这页从 manifest 和回测 summary 生成，用来区分：运行口径、无未来函数历史证据、候选研究、as-of 修正、退役证据和 live audit。默认先看这里，再进入具体页面。
更新时间：2026-05-06 · manifest: reports/artifacts/rank213_evidence_map/manifests
一句话结论
Rank213 不能再被简写成”历史滚动 Top30 已长周期验证通过”。 当前更严谨的读法是：frozen30 只说明当前 runner / paper lane 怎么跑；真正用来检查”去掉选池未来函数以后还行不行”的，是 monthly_volume_causal。这条证据明显削弱旧结论。
如果你只想判断是否应该继续实盘扩张：先看下面”无未来函数版本”的表。它显示旧母策略在上月 K 线选池下很弱，固定 gate 后仍是负收益。
当前活跃策略：rank213_age90_14d_skip1d_voladj 已于 2026-05-06 接入真钱 canary。Phase 3 验证未通过正式 promotion，但用户选择先跑 tiny-live 进行实盘 falsification。详见下方 Live Canary 卡片。
三分钟读懂
- 问题在哪里：如果拿 4 月才确定的 frozen30 名单去跑 3 月或更早历史，等于把未来才知道的标的带回过去，历史收益容易虚高。
- 修正怎么做：每个月只用上一完整自然月 Binance UM perpetual
1dK 线里的quote_volume总和选 Top30，当月只交易这批标的。 - 这不是完美 market cap 真值：它是成交额 / 热门度 proxy，但它至少满足“当月选池不看当月未来表现”。
- 读法优先级：讨论历史有效性先看
monthly_volume_causal；讨论当前执行/残仓才看live_audit_shadow；不要用asof或frozen30冒充滚动选池证明。
无未来函数版本：上月 K 线选池回测
样本 2020-02-01T00:00:00Z → 2026-04-10T18:00:00Z，rebalance 18087 次。选池覆盖：候选币 534，任一月份入选过 247 个，月均和 frozen30 重叠约 12.37 个。
| 版本 | 它在测什么 | 开仓率 | 单次均值 | 累计净收益 | 最大回撤 | 人话结论 |
|---|---|---|---|---|---|---|
| 1) plain baseline plain_baseline | 原始 15m Rank213 排名逻辑，不加 veto，不加 gate；每次都交易。 | 100.00% | -1.58 bps | -98.09% | -99.62% | 亏损或弱化明显，不支持继续用旧故事解释 Rank213。 |
| 2) baseline + veto baseline_plus_veto | 在 baseline 上加入 short-leg jump veto；仍然每次都交易。 | 100.00% | -2.53 bps | -99.60% | -99.79% | 亏损或弱化明显，不支持继续用旧故事解释 Rank213。 |
| 3a) veto + fixed gate baseline_plus_veto_plus_fixed_gate | 沿用 frozen30 研究里的固定 gate；gate OFF 时空仓。 | 3.54% | -0.23 bps | -37.13% | -48.63% | 亏损或弱化明显，不支持继续用旧故事解释 Rank213。 |
| 3b) veto + percentile gate q60 baseline_plus_veto_plus_percentile_gate_q60 | 用 monthly-volume 历史自身的 expanding percentile q60 gate；这是研究候选，不是当前 live 规则。 | 16.44% | 0.18 bps | 6.47% | -71.22% | 全样本略正，但回撤很深，只能当候选研究，不能当已过关。 |
重点不是某个单月，而是同一套 Rank213 规则一旦换成 causal rolling universe，baseline 和 veto 都接近归零式亏损；fixed gate 只把亏损缩小到 -37.13%，不能证明策略已过关。
为什么旧结果看起来更好
下面是同一长样本、同一套 Rank213 规则、同一成本口径下的关键差异：只换 universe 选池方式，结果就从正收益变成负收益。
| 对比项 | frozen30 固定名单 | 上月 K 线选池 / monthly-volume causal |
|---|---|---|
| 选池方式 | 固定 admission 30 币；历史回看时容易把后面才知道的标的带回过去。 | 每个月只用上一完整自然月 Binance UM perpetual 1d K 线的 quote_volume 总和选 Top30，再交易当月。 |
| 累计净收益 | 100.72% | -37.13% |
| 单次均值 | 0.45 bps | -0.23 bps |
| 最大回撤 | -32.01% | -48.63% |
| 读法 | 这解释了为什么旧页面看起来强。 | 这才是当前更该优先看的历史 sanity check；结果明显弱化。 |
这不是说 frozen30 页面没有用；它有运行和执行定义价值。但它不能回答“历史上每个月重新选 Top30 是否有效”。
证据分层：每个页面该怎么用
下面的卡片按证据面分组。进入具体页面前，先看它的 universe、causality 和 do-not-use，避免把执行审计误读成历史 alpha。
Current Runtimepaused_runtime
Frozen admission 30-symbol runner
current_runtime_frozen30 · paper/213a_paper_runner.html
| Universe | frozen30 |
|---|---|
| Causality | static admission seed; not a rolling historical top30 |
| Cadence | 3h non-overlapping frozen seed baskets |
| Sample | 2026-02-09T10:15:00Z -> 2026-03-28T07:00:00Z |
| Verdict | This is the frozen paper/live reference lane definition. Live automation is paused; this lane is not evidence that historical rolling top30 selection works. |
closed_baskets=369mean_net_bps=22.03lifetime_total_return=113.47%
Use / do-not-use / artifacts
Use for
- current runner status
- frozen runtime definition
- dedicated paper lane plumbing
Do not use for
- claiming long-history rolling top30 validation
- claiming universe selection is free of survivorship bias
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_status.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_state.json
Live Auditpaused_audit
Raw-bar shadow / dryrun / live canary audit
live_audit_shadow · paper/213a_shadow_runner.html
| Universe | frozen30 execution universe |
|---|---|
| Causality | execution audit; consumes current frozen universe and latest raw-bar recompute |
| Cadence | 15m official-close audit plus canary pending manager |
| Sample | runtime -> runtime |
| Verdict | Use this lane to understand latest decision, gate state, basket completeness, residuals, and execution drift when automation is enabled. Live automation is currently paused; do not read this lane as long-history alpha evidence. |
latest_shadow_decision=runtimebasket_parity=execution_auditresidual_reconciliation=execution_audit
Use / do-not-use / artifacts
Use for
- latest official-close decision
- gate on/off state
- basket parity and execution drift
- residual or conflict blocker audit
Do not use for
- long-history performance validation
- universe selection validation
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_shadow_status.json
- reports/artifacts/rank213_live_canary/live_status.json
- reports/artifacts/rank213_live_canary_shell/live_status.json
Causal Historicalprimary_historical_evidence
Monthly-volume causal universe rebuild
monthly_volume_causal · paper/213a_volume_universe.html
| Universe | monthly_volume_top30 |
|---|---|
| Causality | select each month by previous full calendar month's Binance UM perpetual 1d quote volume |
| Cadence | 15m rank213 rules on monthly rebuilt universe |
| Sample | 2020-02-01T00:00:00Z -> 2026-04-10T18:00:00Z |
| Verdict | This is the current primary historical sanity check for rolling selection. It materially weakens the old Rank213 story. |
plain_baseline=-98.09%baseline_plus_veto=-99.60%baseline_plus_veto_plus_fixed_gate=-37.13%percentile_q60=slightly_positive_but_deep_drawdown
Use / do-not-use / artifacts
Use for
- historical rolling-selection discussion
- baseline weakness assessment
- deciding whether old 15m mother strategy deserves more live expansion
Do not use for
- exact official historical market-cap reconstruction
- claiming the strategy passed long-history validation
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_monthly_volume_universe_rebuild_summary.json
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_monthly_volume_segment_stability_summary.json
Candidate Researchresearch_only
Monthly-volume baseline refresh
baseline_refresh · paper/213a_volume_baseline.html
| Universe | monthly_volume_top30 |
|---|---|
| Causality | same causal monthly volume universe as rebuild; different daily baseline candidates |
| Cadence | daily rebalance / next-day close-to-close hold |
| Sample | 2020-02-01T00:00:00Z -> 2026-04-10T00:00:00Z |
| Verdict | Useful as a first falsification of new mother baselines, not an apples-to-apples replacement for original 15m Rank213. |
best_candidate=age90_14d_skip1d_voladjbest_mean_net_bps=10.92best_net_cum_pct=455.65%best_max_drawdown=-60.08%
Use / do-not-use / artifacts
Use for
- candidate baseline triage
- designing a cleaner successor research track
Do not use for
- switching current live strategy directly
- claiming original Rank213 has been repaired
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_monthly_volume_baseline_refresh_summary.json
Candidate Researchresearch_only
Baseline V2 four-direction initial review
baseline_v2_four_direction · paper/rank213_baseline_v2_four_direction_review.html
| Universe | monthly_volume_top30 |
|---|---|
| Causality | same causal monthly volume universe; first three directions use daily next-day backtest, fourth needs historical perp data |
| Cadence | daily rebalance / next-day close-to-close hold for price baselines |
| Sample | 2020-02-01T00:00:00Z -> 2026-04-10T00:00:00Z |
| Verdict | Useful as the next baseline research map. The first direction is strongest so far, but drawdown remains too high; the perp overlay needs historical funding/basis/OI data before evaluation. |
age90_14d_skip1d_voladj=455.65% cum / -60.08% DDresidual_14d_skip1d_voladj=200.24% cum / -63.63% DDresidual_blowoffpen=93.21% cum / -74.56% DDperp_funding_basis_overlay=data_gap_no_valid_backtest
Use / do-not-use / artifacts
Use for
- deciding which replacement baseline deserves second-round testing
- documenting why baseline_v2 should not be confused with original 15m Rank213
- tracking the funding/basis data gap
Do not use for
- switching current live strategy directly
- claiming baseline_v2 has passed execution-grade validation
- judging perp funding/basis overlay without historical data
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_baseline_v2_four_direction_review_summary.json
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_baseline_v2_four_direction_review_overall.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_baseline_v2_four_direction_review_annual.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_baseline_v2_four_direction_review_monthly.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_baseline_v2_four_direction_review_monthly_summary.csv
Candidate Researchsecond_round_research_only
age90_14d_skip1d_voladj second-round validation
age90_14d_second_round_validation · paper/213b_round2_validation.html
| Universe | monthly_volume_top30 |
|---|---|
| Causality | previous full-month quote-volume universe; age >= 90d; score uses t-15d to t-1d only |
| Cadence | daily rebalance / top3-bottom3 / 4bps per-basket research cost |
| Sample | 2020-02-01T00:00:00Z -> 2026-04-13T00:00:00Z |
| Verdict | Strongest current replacement-baseline candidate, but still research-only. Cost sensitivity, 2022-2023 weakness, and lack of minute/orderbook execution validation prevent live promotion. |
paper close-to-close: 455.65% cum / -60.08% DD / 10.92 bps meancost sensitivity: 8bps remains +124.78%, 12bps drops to -9.10%2022-2023 fold is weak: -18.42% cum / -50.41% DDraw 1d open/close timing stress: same-day open-to-next-open 2983.62%, delayed next-open 464.11%simple prior-60d positive gate: 382.29% cum / -33.68% DD / 53.69% active
Use / do-not-use / artifacts
Use for
- second-round discussion of whether age90_14d deserves deeper walk-forward testing
- checking cost, monthly stability, long/short attribution, and execution timing sensitivity
- separating this candidate from the original weak Rank213 baseline
Do not use for
- claiming execution-grade validation
- deploying live without minute/TWAP/slippage tests
- assuming the high open-to-open result is directly tradeable at scale
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_second_round_validation_summary.json
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_second_round_validation_cost_sensitivity.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_second_round_validation_time_folds.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_second_round_validation_monthly.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_second_round_validation_rolling_windows.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_second_round_validation_symbol_attribution.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_second_round_validation_execution_timing.csv
Live Canarylive_canary_active
age90_14d_skip1d_voladj — current live canary (2026-05-06)
age90_14d_phase3_validation · paper/213b_phase3_validation.html
| Universe | monthly_volume_top30 |
|---|---|
| Causality | previous full-month quote-volume universe; age >= 90d; score uses t-15d to t-1d only; signal skips most recent day for causality |
| Cadence | daily rebalance / top3-bottom3 / $20 per leg / maker-first entry + horizon exit |
| Sample | 2020-02-01T00:00:00Z -> 2026-04-13T00:00:00Z |
| Verdict | Phase 3 did not pass formal live promotion (drawdown, cost sensitivity, walk-forward fragility, weak short-side). User chose to proceed with tiny-live canary for real-money falsification since 2026-05-06. Live engine signal formula cross-validated against backtest — identical output. |
Phase 3: 15m TWAP at 12bps 272.55% cum / -63.68% DDPhase 3: walk-forward 2023 weak: -13.35% cum / -38.25% DDPhase 3: short-only minus 4bps: -96.97% cum / -99.52% DDLive engine cross-validated vs backtest (April 12 identical)Live canary: $20/leg, 3L+3S, maker-first + horizon exitGate: eligible universe >= 6 symbols
Use / do-not-use / artifacts
Use for
- understanding current live canary setup and signal formula
- reviewing Phase 3 validation details (TWAP/VWAP stress, walk-forward, side decomposition)
- monitoring live shadow runner and live canary artifacts
- evaluating whether Phase 3 issues survive real-money conditions
Do not use for
- claiming the strategy has passed formal live promotion
- expanding notional before accumulating sufficient live samples
- ignoring short-side weakness and walk-forward instability findings
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_phase3_validation_summary.json
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_phase3_execution_cost_grid.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_phase3_liquidity_capacity.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_phase3_gate_grid.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_phase3_walk_forward.csv
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_age90_14d_phase3_side_decomposition.csv
- reports/site/paper/213c_shadow_runner.html
- reports/site/paper/213c_live_launch.html
Partial Causalitylimited_historical_evidence
As-of universe long-history review
asof_frozen_seed · paper/213a_asof_universe.html
| Universe | frozen seed with onboard-time availability |
|---|---|
| Causality | symbols only participate after onboard time; no rolling top30 reselection |
| Cadence | 15m rolling research cadence |
| Sample | 2020-02-01T00:00:00Z -> 2026-04-10T18:00:00Z |
| Verdict | Useful to prevent symbols from trading before listing, but insufficient to validate historical rolling universe selection. |
fixes_onboard_time_visibilitydoes_not_roll_select_top30full_period_veto_negative
Use / do-not-use / artifacts
Use for
- onboard-time causality check
- separating listing-time visibility from universe-selection causality
Do not use for
- claiming rolling top30 passed
- claiming current frozen30 selection was historically available
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_asof_universe_long_history_review_summary.json
Formal Frozen Evidencesuperseded_by_monthly_volume_for_historical_claims
Formal frozen baseline / veto / gate review
formal_frozen_gate · paper/213a_formal_review.html
| Universe | frozen seed with as-of availability |
|---|---|
| Causality | formal gate is causal; original frozen universe selection remains survivorship-risky |
| Cadence | 15m rolling research cadence |
| Sample | 2020-02-01T00:00:00Z -> 2026-04-10T18:00:00Z |
| Verdict | Keep as formal definition and frozen-gate reference, but do not use it alone as long-history proof after monthly-volume rebuild weakened the result. |
formal_baseline_plus_veto_plus_gate=100.72%gate_on_rate=5.04%selection_risk=survivorship_bias
Use / do-not-use / artifacts
Use for
- formal rule definition
- gate rule reference
- same-formula comparison
Do not use for
- standalone historical validation
- rolling universe proof
Source artifacts
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_formal_threeway_backtest_summary.json
- reports/artifacts/paper_rank213_largecap_xs_jump_veto/rank213_formal_strategy_freeze_summary.json
Deprecated / Riskretired
Monthly marketcap universe rebuild
retired_marketcap_rebuild · paper/213a_monthly_marketcap.html
| Universe | retired marketcap proxy |
|---|---|
| Causality | retired; replaced by monthly volume proxy due proxy distortions |
| Cadence | historical exploration |
| Sample | retired -> retired |
| Verdict | Do not use for current decisions. Kept only as audit trail. |
retired=truereplacement=monthly_volume_causal
Use / do-not-use / artifacts
Use for
- historical audit trail
Do not use for
- current conclusion
- strategy validation
- universe selection validation
Source artifacts
- 无