# chase-extended Bucket Audit

**Date:** 2026-06-04
**Scope:** Every trade where `canonical_strategy_slug = 'chase-extended'` in `main.trade_context` (n=58)
**Method:** Layer 1 evidence-only pull + Layer 2 provisional reclassification with source/proxy/heuristic-tagged rules
**Constraints:** Read-only on production. No DB / taxonomy / classifier code / VCET changes.

---

## TL;DR — answer to the key question

> Is `chase-extended` profitable because:
> **A.** true Chase / Extended trades are working,
> **B.** real researched setups are being mislabeled,
> **C.** the bucket contains crypto/hold-time/market-regime artifacts,
> **D.** or there is insufficient data?

**Answer: B + C, supported by D. A is rejected.**

- **A (true chase trades working) — NO.** Only **2 trades** match the canonical chase signature (RVOL > 2.5 AND |gap| > 1.5%), netting **&minus;$4.32 on 1/2 win rate**. The chase mechanism itself is NOT profitable in this sample.
- **B (real setups mislabeled) — YES, dominant.** **9 trades (15.5% of the bucket) generate +$333.28** of net P&L: 7 trend_trade candidates (MTF ≥ 80 in trade direction, +$255.30) and 2 event_driven candidates (catalyst-driven, +$77.98, 2/2 WR). **Without these 9, the bucket nets &minus;$135 across the other 49 trades.**
- **C (hold-time / market-regime artifacts) — YES, strong.** Trades held >60 min net **+$261.22** (43 trades, 67% WR); sub-15min trades net **&minus;$64.87** (7 trades, 1/7 WR). The 10–12 ET entry window is the only consistent loser (&minus;$54.96 on 10 trades). The bucket's profitability is shaped by hold time and time of day, not by a chase setup mechanism. Crypto-adjacent leak persists (ETHU &minus;$60, IBIT &minus;$47).
- **D (insufficient data) — YES, supplementary.** **37 of 58 trades (64%)** cannot be classified to a non-chase mechanism from DB fields alone — they need intraday VWAP distance, VWAP slope, and candle structure to confirm any alternative mechanism. The default chase-extended assignment from Priority 5 stands for these as a residual label.

**The bucket's +$198.28 surface profit hides: 9 mislabeled real-setup trades carrying the edge, 1 huge pre-market NVDA outlier (+$199.07 alone), and a residual mass of unclassifiable trades that net modestly negative.**

---

## 1. Rule discipline — source / proxy / heuristic separation

Per the audit discipline, every rule below is tagged by where it came from. **Rules of different types are NOT mixed in a single classification decision; they are applied in priority order and the lowest-strength tag caps the confidence.**

### Source rules (S) — from gen_scorecard.py and playbook documents

| ID | Source | Rule |
|---|---|---|
| **S1** | `gen_scorecard.py:454-468` Priority 3 | VWAP Pull-In → Trend: pnl>0 AND entry within 1.0 est-SD of session avg AND exit moved ≥2.0 est-SD |
| **S2** | `gen_scorecard.py:484-492` Priority 5 | Chase / Extended: entry > 2.0 est-SD from session avg AND no other priority fired *(THIS is the rule that produced the chase-extended label)* |
| **S3** | VWAP_PULL_IN_QUALITY_GATE §3 | Rubberband / Backside Long: extended ≥2.5 SD BELOW intraday VWAP, double-bottom failure forming, LONG entry |
| **S4** | VWAP_PULL_IN_QUALITY_GATE §3 | Range Break / Hitchhiker: at intraday/HTF range edge, volume building toward edge |
| **S5** | VWAP_PULL_IN_QUALITY_GATE §3 | Pull-In → Trend: HH+HL after +2 SD close, pullback to +1 SD, VWAP slope rising |
| **S6** | Canonical taxonomy — event_driven family | Trade driven by a named catalyst (earnings, news, FDA, M&A) |
| **S7** | Canonical taxonomy — level_breakout family | Entry on a multi-day/52-week level break with consolidation prior |
| **S8** | Canonical taxonomy — gap_fade family | Trade fades a large opening gap (SHORT a gap-up, LONG a gap-down) |

### Database proxies (P) — DB-derived approximations of source rules

| ID | Approximates | Proxy rule | Notes |
|---|---|---|---|
| **P1** | S2 / Quality Gate chase definition | `rvol_pre_entry > 2.5 AND \|gap_pct\| > 1.5%` | Matches Quality Gate "RED tape" definition |
| **P2** | S3 Rubberband extension (intraday VWAP missing) | `\|atrs_from_prior_close\| ≥ 2.0` | Prior-close ATR is daily, NOT intraday; weak proxy for VWAP distance |
| **P3** | S8 gap_fade | `\|gap_pct\| ≥ 3% AND side opposite to gap direction` | Direction + magnitude detectable from DB |
| **P4** | S7 level_breakout | `is_52wk_breakout = TRUE OR consolidation_days ≥ 5` | Direct DB fields |
| **P5** | S6 event_driven | `catalyst_type IS NOT NULL AND catalyst_score ≥ 3` | Direct DB fields |
| **P6** | trend-trade family | `\|mtf_grand_total\| ≥ 80 AND side matches MTF direction` | MTF totals are 6-TF × 4-tier confluence score |

### Provisional heuristics (H) — mine, lowest confidence

| ID | Rule | Notes |
|---|---|---|
| **H1** | Side opposes MTF grand sign by ≥80 pts (LONG with MTF ≤ -80; SHORT with MTF ≥ 80) | Catches counter-trend trades that may be unintentional fades |
| **H2** | Hold time < 5 min AND `\|pnl_per_share\| < $0.10` | Quick-scratch artifact, mechanism unclear |

### Confidence ceilings enforced

Because the database does NOT contain intraday VWAP geometry (VWAP distance, band width, VWAP slope) or candle structure (HH+HL, double bottom, range position), the following confidence caps apply:

| Setup requiring... | Max confidence |
|---|---|
| Intraday VWAP distance at entry | MEDIUM |
| Intraday band width / slope | MEDIUM |
| Candle structure (HH+HL, double-bottom) | MEDIUM |
| Just catalyst fields (in DB) | HIGH |
| Just 52wk-breakout + consolidation_days (in DB) | HIGH |
| Just RVOL + gap (in DB) | MEDIUM (because TRUE chase requires VWAP distance confirmation) |

**No setup in this audit reached HIGH confidence except event_driven** (where the catalyst fields directly populate the source rule with no proxy gap).

---

## 2. Layer 1 — Evidence-only audit

**Output file:** `Z:\public_html\system\CHASE_EXTENDED_AUDIT_evidence.csv` (38 columns × 58 rows)

Columns exported per trade (no classification, just raw evidence):
- Identity: `trade_key`, `trade_date`, `symbol`, `side`, `entry_time`, `exit_time`
- Outcome: `net_pnl`, `pnl_per_share`, `shares`, `realized_r`
- Classifier audit: `strategy`, `raw_classifier_output`
- Pre-trade context: `rvol_pre_entry`, `gap_pct`, `pre_mkt_pct_adv`, `atrs_from_20d_sma`, `atrs_from_prior_close`, `consolidation_days`, `is_52wk_breakout`
- Stock MA stack: `stock_above_5d` / `_10d` / `_20d` / `_50d` / `_200d`
- MTF totals: `mtf_grand_total`, `mtf_position_total`, `mtf_swing_total`, `mtf_intraday_total`
- Catalyst: `catalyst_type`, `catalyst_text`, `catalyst_score`
- Index regime: `tickq_at_entry`, `qqq_above_*`, `spy_above_*`

**Bucket-level stats:**

| Metric | Value |
|---|---:|
| n total | 58 |
| Total P&L | +$198.28 |
| Wins / Losses / Scratches | 33 / 25 / 0 |
| Win rate | 56.9% |
| Catalysts recorded | 2 / 58 (3.4%) |
| 52wk breakouts | 2 / 58 (3.4%) |

**Symbol concentration (top 12):**

| Symbol | n | Total P&L | Wins | Note |
|---|---:|---:|---:|---|
| CRWV | 9 | &minus;$13.38 | 4 | Most-traded; modestly negative |
| IBIT | 6 | &minus;$46.64 | 3 | Crypto-adjacent leak |
| NVDA | 4 | **+$139.58** | 3 | Driven by 6/2 pre-market +$199 |
| ETHU | 3 | **&minus;$60.22** | 1 | Crypto-adjacent leak |
| SOXL | 3 | +$59.58 | 3 | Clean winner |
| MSFT | 3 | +$28.56 | 2 | |
| TSLA | 3 | &minus;$2.04 | 2 | |
| AAPL | 2 | +$64.94 | 2 | Both catalyst-driven |
| CAR | 2 | &minus;$26.61 | 1 | |
| BIRD | 2 | +$5.24 | 1 | Note: 4/15 gap = 333.7% — likely data anomaly |
| WULF | 2 | &minus;$2.73 | 1 | |
| SNOW | 1 | +$36.27 | 1 | |

---

## 3. Layer 2 — Provisional reclassification

**Output file:** `Z:\public_html\system\CHASE_EXTENDED_AUDIT_reclassification.csv` (21 columns × 58 rows)

Columns: `trade_key, trade_date, symbol, side, entry_time, exit_time, net_pnl, rvol_pre_entry, gap_pct, atrs_from_prior_close, mtf_grand_total, is_52wk_breakout, catalyst_type, current_label, proposed_label, confidence, evidence_matched, evidence_missing, why_not_chase, why_it_may_still_be_true_chase, required_missing_fields_to_confirm`

### Breakdown by proposed mechanism

| Proposed label | n | Total P&L | Win rate | Confidence | % of bucket P&L |
|---|---:|---:|---:|:---:|---:|
| **trend_trade (candidate)** | 7 | **+$255.30** | 4/7 (57%) | MEDIUM | 128.7% |
| **event_driven (candidate)** | 2 | **+$77.98** | 2/2 (100%) | **HIGH** | 39.3% |
| gap_fade (candidate) | 7 | &minus;$3.06 | 3/7 (43%) | MEDIUM | &minus;1.5% |
| chase_extended (confirmed) | 2 | &minus;$4.32 | 1/2 (50%) | MEDIUM | &minus;2.2% |
| level_breakout (candidate) | 2 | &minus;$31.51 | 1/2 (50%) | MEDIUM | &minus;15.9% |
| counter_trend_scalp (candidate) | 1 | &minus;$22.43 | 0/1 (0%) | LOW (heuristic) | &minus;11.3% |
| insufficient_data | 37 | &minus;$73.68 | 22/37 (59%) | INSUFFICIENT | &minus;37.2% |
| **TOTAL** | **58** | **+$198.28** | **33/58 (57%)** | | **100.0%** |

### Confidence distribution

| Confidence | n |
|---|---:|
| HIGH | 2 |
| MEDIUM | 18 |
| LOW | 1 |
| INSUFFICIENT | 37 |

### The 7 trend_trade candidates (carrying +$255.30)

| Date | Symbol | Side | P&L | MTF grand | RVOL | Gap % | Note |
|---|---|---|---:|---:|---:|---:|---|
| 2026-05-04 09:32 | CRWV | LONG | +$26.05 | 84 | 2.23 | +3.5% | Clean trend confluence |
| 2026-05-04 09:35 | DXYZ | LONG | +$31.31 | **103** | 2.62 | +2.9% | Strong MTF |
| 2026-05-07 09:41 | DDOG | LONG | &minus;$1.90 | 90 | **7.23** | **+32.3%** | Earnings-blowoff territory; barely scratched |
| 2026-05-11 09:33 | NVDA | LONG | +$33.32 | 81 | 1.79 | +0.2% | |
| 2026-05-29 09:25 | RGTI | LONG | &minus;$23.20 | 97 | 0.04 | &minus;1.2% | RVOL near zero — MTF was up but tape was dead |
| 2026-05-29 10:02 | MASK | LONG | &minus;$9.35 | 106 | 0.85 | **+30.5%** | Crypto-name with huge gap; thin tape |
| 2026-06-02 05:58 | NVDA | LONG | **+$199.07** | 81 | 3.27 | +2.0% | **Pre-market — dominant outlier** |

**4 of 7 winners; +$255 total but +$199 comes from one pre-market NVDA. Median per-trade is +$26.05.** Even if we accept all 7 as trend_trade, the mechanism is not uniformly profitable — DDOG and MASK with huge gaps stalled, RGTI with dead RVOL lost.

### The 2 event_driven candidates (both winners, HIGH confidence)

| Date | Symbol | Side | P&L | Catalyst | Score |
|---|---|---|---:|---|---:|
| 2026-05-05 09:20 | INTC | LONG | +$27.61 | minor_news | (≥3) |
| 2026-05-06 09:32 | AAPL | LONG | +$50.37 | minor_news | (≥3) |

Both pre-9:45 open entries on named catalysts. Only HIGH-confidence calls in the audit.

### The 2 confirmed chase-extended (canonical chase signature met)

| Date | Symbol | Side | P&L | RVOL | Gap % |
|---|---|---|---:|---:|---:|
| 2026-04-13 09:39 | IBIT | LONG | +$1.08 | 3.84 | +2.4% |
| 2026-06-04 10:08 | IBIT | LONG | &minus;$5.40 | 2.74 | +1.9% |

Net &minus;$4.32 over 2 trades. **The true chase pattern is unprofitable in this sample** — consistent with the Quality Gate's read.

### The 37 insufficient_data residuals (net &minus;$73.68)

These trades met NO candidate rule (no strong MTF, no catalyst, no 52wk breakout, no large gap-fade signature, no extreme prior-close extension, not a hot-tape chase, not a quick-scratch). The Priority 5 chase-extended assignment stands as the residual default. Top winners and losers within this subset:

**Top winners in insufficient_data:**
- CRWV 2026-05-14 09:26 LONG +$44.69 (RVOL 0.01, gap +0.6%, MTF 2 — odd profile)
- CRWV 2026-05-28 09:35 LONG +$39.14
- SNOW 2026-04-13 09:40 LONG +$36.27
- ETHU 2026-05-14 09:09 LONG +$35.84

**Top losers in insufficient_data:**
- NVDA 2026-05-27 08:21 LONG **&minus;$100.51** (pre-market, RVOL 1.46, MTF 2 — likely a real chase)
- ETHU 2026-05-06 09:09 LONG **&minus;$96.01** (pre-market, RVOL 0.47 — odd)
- CAR 2026-04-10 09:35 SHORT &minus;$56.86
- IBIT 2026-04-27 09:39 LONG &minus;$45.27

**The 37 insufficient_data trades net 22 winners / 15 losers — modest positive raw rate but P&L distribution is fat-tailed by the few large losers above.**

---

## 4. The hold-time / time-of-day artifact (mechanism C)

This is where the bucket's profitability really lives, and it's NOT a setup mechanism.

### Hold-time distribution

| Hold bucket | n | Total P&L | Wins | Note |
|---|---:|---:|---:|---|
| Sub-5 min | 4 | **&minus;$45.90** | 0/4 (0%) | Quick-exit losers — H2 heuristic flagged some |
| 5–15 min | 3 | &minus;$18.97 | 1/3 (33%) | |
| 15–60 min | 8 | +$1.93 | 3/8 (38%) | Near break-even |
| **Over 60 min** | **43** | **+$261.22** | **29/43 (67%)** | **All the profit lives here** |

**74% of trades held >60 min** carry **132% of the bucket's net profit**. Quick exits (sub-15min, n=7) are 1/7 WR with combined &minus;$64.87. The bucket "works" because long holds resolved positively — not because the chase setup worked.

### Time-of-entry distribution

| Window | n | Total P&L | Wins |
|---|---:|---:|---:|
| Pre-market (< 09:00) | 9 | **+$92.06** | 6/9 (67%) |
| Open 09:30–09:45 | 27 | +$58.75 | 15/27 (56%) |
| 09:45–10:00 | 7 | +$55.53 | 5/7 (71%) |
| **10:00–12:00** | **10** | **&minus;$54.96** | **3/10 (30%)** |
| 12:00–14:00 | 3 | +$39.62 | 3/3 (100%) |
| After 14:00 | 2 | +$7.28 | 1/2 (50%) |

**The 10–12 ET mid-morning window is the only consistent loser.** Pre-market and open windows are profitable; later windows are scratchy but positive. This points to a **time-of-day artifact** — chase-extended trades in mid-morning lose; the same label in pre-market or early-open windows wins. Setup label doesn't change; mechanism does.

### Crypto-adjacent concentration (mechanism C continued)

ETHU (&minus;$60.22), IBIT (&minus;$46.64), and crypto-name MASK (&minus;$9.35) carry &minus;$116.21 of bucket P&L across 10 trades. Same Quality-Gate finding as the plain Pull-In bucket: **crypto-adjacent leveraged products do not behave like clean-tape Pull-In or chase**, regardless of slug.

---

## 5. Answering the key question definitively

| Option | Verdict | Evidence |
|---|---|---|
| **A.** True chase trades working | ❌ **NO** | Only 2 trades match canonical chase signature; they net &minus;$4.32 (1/2 WR). The chase mechanism itself is not profitable. |
| **B.** Real researched setups mislabeled | ✅ **YES — dominant** | 9 trades (15.5% of bucket) = +$333.28 in trend_trade + event_driven candidates. Without them, bucket is &minus;$135. The bucket's "profit" is mechanistically these 9 trades, not the residual 49. |
| **C.** Crypto / hold-time / market-regime artifacts | ✅ **YES — strong** | 132% of net profit comes from >60 min holds (43 trades); sub-15min trades are 1/7 WR. Mid-morning (10–12 ET) is the only loser window. Crypto-adjacent ETHU/IBIT/MASK = &minus;$116. NVDA pre-market +$199 alone = 100% of bucket P&L. |
| **D.** Insufficient data | ✅ **YES — supplementary** | 37/58 trades (64%) cannot be classified to a non-chase mechanism from DB fields alone. Need intraday VWAP distance, slope, candle structure to confirm. Until then, the chase-extended residual assignment stands honestly. |

**Bottom line:** `chase-extended` shows surface profitability of +$198 because (1) it absorbed 9 mislabeled real-setup trades worth +$333, (2) hold-time/time-of-day artifacts concentrated long-hold profits and pushed out short-hold losses, and (3) one outsized pre-market NVDA trade (+$199) skews the total. **The chase setup itself is NOT a working edge** — consistent with the Quality Gate's earlier read of `chase-extended` as "behavioral chase tag, not an edge."

---

## 6. What this audit recommends (no implementation yet)

1. **Do NOT promote chase-extended to a "working" or recommended setup.** The bucket-level +$198 is artifact; the canonical-chase subset (n=2) is net negative.
2. **The 7 trend_trade candidates and 2 event_driven candidates deserve a closer look** — these may be Priority-3-tightening targets when the classifier refinement happens (per `VWAP_PULL_IN_CLASSIFIER_REFINEMENT_PLAN.md` §8 Option A path). Specifically, `gen_scorecard.py` Priority 5 should NOT absorb trades that meet a strong P5_catalyst or P6_trend_confluence rule.
3. **Add intraday VWAP distance to `main.trade_context`** before any further reclassification work. Without it, 64% of this bucket is unclassifiable — and that's the structural blocker behind all confidence ceilings here.
4. **Mid-morning (10–12 ET) "chase" entries should be reviewed at the trader-behavior level**, not the classifier level. Time-of-day skew this strong likely reflects an attention/tilt window rather than a setup category.
5. **Keep the canonical slug as-is.** Like `vwap-pull-in -> trend`, `chase-extended` is an outcome-based residual label. Renaming or splitting it without VWAP geometry would be premature.

---

## 7. Honest caveats

1. **Sample size is small.** n=58 total; n=21 with any candidate rule firing; n=2 in the HIGH-confidence event_driven cell. Single-trade categories cannot establish a base rate. Trend_trade (n=7) is the largest non-residual category and still small.
2. **64% of the bucket is insufficient_data.** Until intraday VWAP geometry is in the DB, this audit is a partial picture.
3. **One trade (NVDA 6/2 +$199.07) accounts for 100% of bucket P&L.** That's a single-trade dominance problem — the bucket would be net &minus;$1 without it. Re-examining this trade's actual mechanism (was it pre-market trend continuation? earnings catalyst? insider news?) would substantially recolor the audit.
4. **BIRD 2026-04-15 has gap_pct = 333.735%** — almost certainly a data quality issue (corporate action not adjusted) or a thinly-traded small-cap freak open. Worth flagging for data-quality review, but I left it in to avoid silent filtering.
5. **The "trend_trade candidate" label is provisional.** All 7 trades have MTF ≥ 80 in trade direction, but 3 of 7 still lost. MTF is a necessary-but-not-sufficient signal; without VWAP slope confirmation the MEDIUM cap on confidence is correct.
6. **Heuristics H1 and H2 fired only once each (LOW confidence).** Their thresholds are tuned to this small sample and should not be trusted beyond it.

---

## 8. File paths (deliverables)

| File | Purpose |
|---|---|
| `Z:\public_html\system\CHASE_EXTENDED_AUDIT.md` | This summary report (Markdown) |
| `Z:\public_html\system\CHASE_EXTENDED_AUDIT_evidence.csv` | Layer 1 raw evidence (38 cols × 58 rows) |
| `Z:\public_html\system\CHASE_EXTENDED_AUDIT_reclassification.csv` | Layer 2 provisional reclassification (21 cols × 58 rows) |
| `Z:\public_html\system\CHASE_EXTENDED_AUDIT_queries.sql` | All SQL queries used (Q1–Q12, read-only) |
| `C:\Users\jtdom\Downloads\vcet\python\chase_extended_audit.py` | Audit source script (read-only) |
| `C:\Users\jtdom\AppData\Local\Temp\chase_audit_summary.json` | Intermediate summary JSON (rules + tallies) |

URLs:
- https://trsportsacademy.com/system/CHASE_EXTENDED_AUDIT.md
- https://trsportsacademy.com/system/CHASE_EXTENDED_AUDIT_evidence.csv
- https://trsportsacademy.com/system/CHASE_EXTENDED_AUDIT_reclassification.csv
- https://trsportsacademy.com/system/CHASE_EXTENDED_AUDIT_queries.sql

---

## 9. What did NOT change

- ❌ No DB writes &middot; no schema changes &middot; no row updates
- ❌ No `gen_scorecard.py` edits &middot; no classifier logic changes
- ❌ No `main.user_strategies` / canonical taxonomy changes
- ❌ No VCET engine changes &middot; no Scorecards refresh
- ✅ Three CSV/SQL artifacts + one Markdown summary in `/system/`
- ✅ One read-only Python script in `vcet/python/`

**Phase board unchanged.**
