{
  "version": "1.0",
  "strategy": "vwap_pull_in",
  "title": "VWAP Pull-In Gate — Validation Audit Summary",
  "generated": "2026-06-09",
  "analysis_type": "read_only_no_db_writes",
  "core_question": "Does the Green/Yellow/Red gate improve decision quality vs the old generic vwap_pull_in label?",
  "headline_verdict": "PARTIALLY SUPPORTED — REVISE BEFORE PROMOTION. The gate CONCEPT improves expectancy, but almost entirely through the RED exclusion (driven by gap>1.5%) and the post-entry checkpoints — NOT through the RVOL thresholds, which do not separate winners from losers when bucketed. Instrument type is a stronger separator than RVOL.",

  "data_basis": {
    "source": "Z:/trading/dev/vwap-band-entry-surface (Liberator minute_bars_raw backfill; VWAP validated 0.0028% vs kti)",
    "files": ["stage_b/trades_full.csv (323 trades)", "stage_c/post_entry_window_param_surface.csv", "stage_d_visuals/data/*.json"],
    "trade_set": "full-window 323 trades; plain vwap-pull-in n=132 (net -$298.13). NOTE: differs from the 2026-06-03 /system/ snapshot (n=127, -$364.47) because this is the full Liberator window, not the snapshot date. Numbers below are from THIS set.",
    "no_db_query": "All computed from on-disk validated artifacts. No database connection used."
  },

  "validation_status": "working_framework_partially_validated_revise_before_promotion",

  "per_question_verdicts": {
    "Q1_rvol_buckets": {
      "verdict": "NOT SUPPORTED as specified",
      "evidence": "Plain-PI bucketed P&L (rvol populated 108/132): <1.0 = -$320.62/32.1%win (WORST); 1.0-1.5(green) = -$23.80/34.6% (~flat); 1.5-2.5(yellow) = +$14.27/43.8% (slightly +); >2.5(red) = -$58.20/30.8%; NA(24) = +$90.22.",
      "interpretation": "Non-monotonic. The 'high RVOL loses' story (loser-mean 4.01 in the old diagnostic) is outlier-driven and does NOT hold in buckets — the calmest <1.0 bucket carries the most loss. RVOL appears confounded by instrument type (low-RVOL crypto-proxies/illiquids sit in <1.0). The green FLOOR of 1.0 is incidentally useful (it excludes the <1.0 trap).",
      "tier": "rejected_as_specified"
    },
    "Q2_gap_buckets": {
      "verdict": "SUPPORTED (strongest pre-trade filter)",
      "evidence": "Plain-PI by |gap| (populated 132/132): <0.5 = -$36.40/40.0%; 0.5-1.5 = +$80.20/47.4%; >1.5 = -$341.93/24.1%. Signed: gap-up>+1.5 = -$200.09/26.5% (n=34); gap-down<-1.5 = -$141.84/20.0% (n=20).",
      "interpretation": "|gap|>1.5% holds essentially the entire loss at 24% win. BOTH large gap-up and large gap-down lose, so the magnitude rule is correct. Gap is fully populated and adds more value than RVOL. The green/yellow split within <1.5% is weak (yellow > green here).",
      "tier": "proven"
    },
    "Q3_first_3min_fade": {
      "verdict": "SUPPORTED — decisively",
      "evidence": "Plain-PI in FADE state: 3min = 6.7%win/median -$7.29 (n=30); 5min = 9.4%/-$6.95; 10min = 10.3%; 15min = 17.5%. Early-expansion: 3min 62.5% / 5min 72.7% / 15min 77.8% win.",
      "interpretation": "A plain pull-in fading at 3 minutes wins ~7%. This is the single cleanest separator. 'Respect the 3-minute failure' should be a HARD scorecard/management rule.",
      "tier": "proven"
    },
    "Q4_10_15min_expansion_confirms_trend": {
      "verdict": "SUPPORTED — directionally",
      "evidence": "Plain-PI early_expansion at 15min = 77.8% win / 100% reached-next-band. PI->trend trades cluster in expansion/favorable (favorable% 73% at 5 and 15 min vs plain-PI ~34%). Flat rotation stays mediocre.",
      "interpretation": "Trend-hold should only be allowed AFTER 10-15 min expansion confirms. Caveat: PI->trend n=15 (tiny); early expansion decays (~42% persist to 30 min per stage C) so trail, don't blind-hold.",
      "tier": "proven_but_n_tiny"
    },
    "Q5_symbol_repeat_offenders": {
      "verdict": "SUPPORTED for IBIT + CVX (robust); WATCH for AMD/ETHU/RGTI (tiny)",
      "evidence": "Plain-PI by symbol: IBIT -$185.51/10/30% (ROBUST, single biggest leak); CVX -$44.63/5/20% (robust); META -$19.13/9/55.6% (marginal). AMD -$69.65/3/0%, ETHU -$69.35/3/0%, RGTI -$38.98/3/0% (TINY n=3). Winners: MSFT +$133.54/4/100%, AMZN +$43.93/9, TSLA +$43.39/6, GOOGL +$32.67/5.",
      "interpretation": "Only IBIT (n=10) and CVX (n=5) are robust-sample losers. AMD/ETHU/RGTI are 0% but n=3 — underpowered warnings, not a proven blacklist.",
      "tier": "proven_for_IBIT_CVX_only"
    },
    "Q6_instrument_type": {
      "verdict": "SUPPORTED — and a STRONGER separator than RVOL",
      "evidence": "Plain-PI by type: megacap_tech +$151.91/51.0% (n=51, POSITIVE); etf_or_crypto_proxy -$232.26/40.7% (n=27); smallcap_highbeta_or_other -$217.78/18.5% (n=54).",
      "interpretation": "Plain pull-in works on megacap tech and fails on crypto-proxies + small/high-beta names. Likely the real driver behind the RVOL paradox (low-RVOL crypto-proxies). Classification is approximate; cells moderate-n.",
      "tier": "proven_directional"
    },
    "Q7_gate_vs_generic": {
      "verdict": "SUPPORTED — via the RED exclusion, not the green/yellow gradation",
      "evidence": "Generic plain-PI -$298.13/35.6%. Gate partition: green(rvol1.0-1.5 & gap<0.5) +$41.38/30%/n10; green_sub1rvol -$229.31/41%/n17; yellow +$124.45/50%/n32; red -$356.72/25.4%/n59; rvol-NA +$122.07/n14. Passing RED flips the book to ~+$58.59.",
      "interpretation": "The RED gate isolates the entire loss (-$356.72) — primarily gap>1.5%. Yellow is net positive (+$124), so half-sizing yellow would have reduced PROFIT by ~$62, NOT reduced drawdown. The green/yellow split is weakly supported; the value is in RED exclusion + the RVOL>=1.0 floor.",
      "tier": "proven_for_red_exclusion"
    },
    "Q8_mistake_tags_explain_leak": {
      "verdict": "SUPPORTED — partially; 2 of 7 tags not computable",
      "evidence": "traded_red_gate (= chased_hot_tape here) = 59 trades / -$356.72 — explains the entire leak. ignored_3min_failure real (fade 6.7% win). NOT computable without new fields: oversized_yellow (no position_size_vs_plan), no_clear_invalidation (not logged), held_rotation_like_trend (needs management intent).",
      "tier": "proven_partial"
    }
  },

  "proven_rules": [
    "Pass when |gap| > 1.5% (both directions): -$341.93 / 24.1% win over n=54 — the strongest single pre-trade filter (gap 100% populated).",
    "Respect the 3-minute fade: a plain pull-in in a fade state at 3 min wins 6.7% (median -$7.29). Cut it.",
    "Trend-hold only after 10-15 min expansion confirms: early-expansion = 77.8% win / 100% reached-next-band at 15 min.",
    "Instrument type separates: megacap tech +$152/51%; crypto-proxy -$232; small/high-beta -$218. Plain PI is a megacap-tech setup.",
    "IBIT and CVX are robust repeat offenders (IBIT -$185/10/30%; CVX -$44/5/20%).",
    "The RED exclusion improves expectancy: removing RED-gate trades flips plain-PI from -$298 to ~+$59."
  ],

  "rejected_rules": [
    "RVOL green/yellow/red thresholds (1.0-1.5 / 1.5-2.5 / >2.5) as a winner/loser separator: bucketed P&L is non-monotonic and the calmest <1.0 bucket is the WORST (-$320). The 'high RVOL = chase = loss' story is outlier-driven, not bucket-supported.",
    "'Hot tape (high RVOL) is the main leak': FALSE in bucketed data. The leak is big gaps + crypto-proxy/small-cap names + uncut fades. High-RVOL (>2.5) is only -$58 on n=13.",
    "Half-sizing YELLOW to reduce drawdown: yellow was net +$124; half-size would have cut profit, not drawdown."
  ],

  "working_rules": [
    {"rule": "RVOL >= 1.0 floor for GREEN (exclude the <1.0 trap)", "status": "working_rule_needs_backtest", "note": "incidentally useful but confounded with instrument type"},
    {"rule": "green/yellow gradation within |gap|<1.5%", "status": "working_rule_needs_backtest", "note": "weak separation; yellow >= green in this sample"},
    {"rule": "AMD/ETHU/RGTI avoid", "status": "working_rule_needs_backtest", "note": "0% win but n=3 each — underpowered"},
    {"rule": "exact instrument-type thresholds", "status": "working_rule_needs_backtest", "note": "direction proven; classification approximate, moderate n"}
  ],

  "recommended_threshold_adjustments": [
    "Demote RVOL from a primary gate axis to a secondary/contextual flag; investigate the RVOL x instrument-type confound before re-using it.",
    "Promote |gap|>1.5% (magnitude, both directions) to the PRIMARY hard RED condition.",
    "Add INSTRUMENT TYPE as a gate axis: megacap_tech = eligible; etf/crypto_proxy + small/high-beta = stricter (size down or red) until backtested.",
    "Keep the RVOL>=1.0 floor for green (it excludes the losing <1.0 bucket) but label it working_rule_needs_backtest.",
    "Make the 3-minute fade-exit a HARD management/scorecard rule (not advisory).",
    "Reduce the robust symbol watch to IBIT + CVX (proven); keep AMD/ETHU/RGTI as tiny-sample warnings only."
  ],

  "sample_size_warnings": [
    "vwap-pull-in -> trend n=15 (entire 'edge' side); green-proper n=10; per-instrument-type and per-symbol cells often n<10.",
    "RVOL populated on only 108/132 (82%) of plain-PI trades; 24 NA.",
    "PI->trend post-entry cells (Q4) frequently n=1-6 — directional only.",
    "Instrument-type classification is heuristic (symbol->type map), not authoritative.",
    "This is in-sample on ~2 months of one trader's trades; no out-of-sample / walk-forward validation. Treat all as hypotheses to forward-test (CMT Ch.35/38 data-mining caution)."
  ],

  "data_gaps_missing_fields": [
    "position_size_vs_plan / intended_size — REQUIRED to validate 'oversized_yellow'. Not stored.",
    "pre_trade_invalidation_defined — REQUIRED to validate 'no_clear_invalidation'. Not logged.",
    "intended_management_mode / planned_hold — REQUIRED to fully validate 'held_rotation_like_trend'. Not stored.",
    "consistent RVOL definition + full backfill — current rvol_pre_entry is 82% populated and may differ from the trade_context.rvol used in the original diagnostic (which produced the divergent loser-mean 4.01).",
    "live VWAP/SD/band + first-N-min state at entry (VCET v1.1 + 1-min feed) — needed to compute the proven post-entry checkpoints LIVE rather than from backfill."
  ],

  "scorecard_fields_to_prioritize": [
    "gate_color (with gap as the primary driver)",
    "setup_classification (plain_vwap_pull_in / pull_in_to_trend / chase_extended / pass_no_trade)",
    "gap_pct (proven separator, fully populated)",
    "instrument_type (NEW — proven separator; add megacap_tech / etf_crypto_proxy / smallcap_highbeta)",
    "first_3min_result (proven; the fade flag)",
    "first_10_15min_confirmation (proven; gates trend-hold)",
    "mistake_type: traded_red_gate + chased_hot_tape + ignored_3min_failure (computable & proven NOW)",
    "DEFERRED until logged: oversized_yellow, no_clear_invalidation (need new pre-trade fields)"
  ],

  "promotion_readiness": {
    "strategy_library": "NOT READY. Structure (4 slugs) is ready; thresholds are not. Rework RVOL, add instrument-type, confirm gap as primary RED, then promote. Keep validation_status=working_rule_needs_backtest on every threshold until a walk-forward pass.",
    "scorecards": "PARTIALLY READY. The proven fields (gate_color, gap_pct, instrument_type, first_3min_result, first_10_15min_confirmation, setup_classification) and tags (traded_red_gate, ignored_3min_failure) can be captured now. oversized_yellow + no_clear_invalidation need new logged fields first. Design-only until approved — no gen_scorecard.py / _scientific_overlay.js edits.",
    "vcet_live_classifier": "PARTIALLY READY. gate_color (gap-driven) + instrument_type are computable live; RVOL is 82% available and unreliable (do not lead with it); the proven post-entry checkpoints need the live 1-min/VCET-v1.1 feed that is not yet wired. Surface gap + instrument-type now; add post-entry state when the feed lands."
  },

  "recommendation": "REVISE, then promote in stages. Promote NOW (after a confirmatory walk-forward): the gap>1.5% RED rule, the 3-min fade-exit, the 10-15 min trend-confirmation, instrument-type filtering, and the IBIT/CVX watch. REWORK: the RVOL axis. DEFER: green/yellow gradation, AMD/ETHU/RGTI blacklist, and the two non-computable mistake tags until data is logged."
}
