Pull-In → Trend Mechanism Audit

Generated 2026-06-04 · Scope: every trade where canonical_strategy_slug = 'vwap-pull-in -> trend' (n=15) · Method: VCET v1.0 in-memory at slope_trend_thresh=2.0, snap at offsets {−5, 0, +5, +10, +15, +25} min · Rules C > A > B > D from VWAP_PULL_IN_CLASSIFIER_REFINEMENT_PLAN §4 · Constraints: Read-only. No DB / taxonomy / classifier / VCET / Scorecards changes.

⚠ Headline finding

The vwap-pull-in -> trend bucket is a MIXED bucket — and within the classifiable subset, the dollar edge comes from rotation scalps, NOT trend conversions.

Of 6 classifiable trades (bars + rules fit): 1 true Pull-In → Trend conversion (JOBY 5/28, +$12.45), 3 VWAP rotation scalps (DAL 5/26, 5/27, 5/28; net +$96), 2 wrong-side escapes that profited anyway (JOBY 5/29 +$0.70, HL 6/3 +$12.25).

Even when the rule fits cleanly, only 1 of 6 trades (16.7%) is mechanically what the slug name claims. The bucket's profit pattern is dominated by something other than trend conversion.

1. Mechanism breakdown

Mechanism n Total P&L Win rate % of classifiable P&L
vwap_rotation_scalp 3 +$96.34 3/3 (100%) 75.5%
true_pull_in_to_trend 1 +$12.45 1/1 (100%) 9.8%
wrong_side_escape 2 +$12.95 2/2 (100%) 10.2%
insufficient_data (NO_BAR) 6 +$235.24 6/6 —
insufficient_data (trend-continuation) 3 +$10.69 2/3 —
TOTAL 15 +$367.67 14/15 (93%)

2. The 9 insufficient_data cases (matter for interpretation)

Sub-A: 6 NO_BAR — older than kti window (P&L +$235.24)

TradeSideNet P&LGap %Notes
AAOI 4/10 10:09LONG+$136.29+4.2%Largest winner in dataset; mechanism unknown
CAR 4/16 10:06SHORT+$53.76−0.7%Only SHORT in the bucket
WULF 4/13 09:37LONG+$18.48−1.7%
CRWV 4/13 10:17LONG+$9.21+7.4%
RGTI 4/15 10:07LONG+$8.92+9.7%
SPY 5/1 10:43LONG+$8.58−0.9%

These predate the kti 14-day window. Cannot classify mechanism. They represent 64% of the bucket's total P&L ($235 / $368) — any conclusion about the bucket's mechanism is necessarily limited by the historical-bar gap.

Sub-B: 3 trades where bars exist but entry state was already TREND_* (P&L +$10.69)

TradeNet P&LEntry stateTrajectoryObservation
LUNR 5/26 09:49 LONG −$11.76 TREND_UP TREND_UP throughout, slope decelerating (+70° → +18°) Bucket's only LOSER. Entered already-extended in a fading trend. Priority 3 shouldn't have produced this for a loser; came via Phase 2 backfill from manual trader label.
BB 5/28 11:56 LONG +$18.40 TREND_UP TREND_UP whole window, slope flat (+10° → +4°), bw flat at 8.7 Entered into established TREND_UP slowly decelerating — trend continuation entry, not a conversion
INTC 6/2 10:44 LONG +$4.05 TREND_UP TREND_UP whole window, slope mildly accelerating (+1.2° → +5°) Similar — trend continuation, not Pull-In

The 4th mechanism the rule set doesn't model

Sub-B reveals a category not in rules A/B/C: trades where entry state was already TREND_* in trade direction. That's not a Pull-In conversion — it's a trend continuation entry. The classifier's "entry near VWAP" condition does NOT mean "state was ROTATION/COMPRESSION." Trades can enter near the rolling VWAP price while VCET is already calling TREND_UP/DOWN. The classifier is blind to state at entry.

3. Per-trade detail (the 6 classifiable trades)

1. JOBY 2026-05-28 09:31 LONG +$12.45 → true_pull_in_to_trend HIGH

RVOL 1.59 · gap +0.74% · ATR +0.11 · MTF 62 · no catalyst

tstateslope°bw%expclose
−5ROTATION+1.611.542011.43
0ROTATION+17.992.987511.59
+5ROTATION+31.375.037511.76
+10ROTATION+42.926.167511.87
+15TREND_UP+49.066.423511.83
+25TREND_UP+44.386.722011.98

Evidence: bw(+5)=5.03 ≥ 1.5 × bw(0)=2.98 ✓ · expansion_score(+5)=75 ≥ 50 ✓ · Δslope(+5) = +13.4° ≥ +5° ✓ · state=TREND_UP at +15min ✓ (4 of 5 sub-checks → HIGH).

Verdict: Textbook ROTATION → TREND_UP conversion in 15 minutes. Bands doubled, slope exploded, state transitioned cleanly. This IS what the slug name claims.

2. DAL 2026-05-26 13:56 LONG +$68.25 → vwap_rotation_scalp HIGH

RVOL 0.73 · gap −1.31% · ATR −0.39 · MTF 83 · no catalyst

tstateslope°bw%expclose
−5ROTATION−0.391.49078.32
0ROTATION−0.351.49078.42
+5ROTATION−0.351.491578.54
+10ROTATION−0.341.49078.53
+15ROTATION−0.291.48078.50
+25ROTATION−0.201.48078.52

Verdict: Pure TD VWAP rotation. Slope ≈ 0 throughout, bands flat at 1.49, no expansion ever fired. Trade made +$68 from rotating around VWAP — not from trend.

3. DAL 2026-05-27 13:56 LONG +$22.50 → vwap_rotation_scalp HIGH

RVOL 1.31 · gap −5.35% · ATR −1.62 · MTF 82 · no catalyst

tstateslope°bw%expclose
−5ROTATION−1.152.823581.73
0ROTATION−1.342.861581.71
+5ROTATION−1.632.923581.49
+10ROTATION−1.762.971581.46
+15ROTATION−1.923.011581.50
+25ROTATION−1.403.05081.56

Verdict: Counter-trend buy on −5.35% gap-down. Stayed in ROTATION 25 min. Price went AGAINST the LONG entry (81.71 → 81.56). +$22 was tight sizing / fast scratch / partial scale near entry.

4. DAL 2026-05-28 13:56 LONG +$5.59 → vwap_rotation_scalp HIGH

RVOL 0.94 · gap −8.14% · ATR −2.50 · MTF 74 · no catalyst

tstateslope°bw%expclose
−5ROTATION+0.212.841581.97
0ROTATION+0.272.823581.93
+5ROTATION+0.202.81081.89
+10ROTATION+0.202.79081.95
+15ROTATION+0.172.78081.88
+25ROTATION+0.112.75082.01

Verdict: Compression scalp. Bands CONTRACTING (2.84 → 2.75), slope flat near 0, expansion never fired. Modest +$0.08 drift over 25 min. Compression existed but never resolved into expansion.

5. JOBY 2026-05-29 09:31 LONG +$0.70 → wrong_side_escape HIGH

RVOL 1.76 · gap −5.61% · ATR −0.88 · MTF 22 (weak) · no catalyst

tstateslope°bw%expclose
−5ROTATION−3.942.057512.12
0ROTATION−36.595.547511.92
+5TREND_DOWN−52.747.027511.73
+10TREND_DOWN−62.227.593511.74
+15TREND_DOWN−66.917.513511.77
+25TREND_DOWN−37.097.242011.78

Verdict: LONG into a TREND_DOWN crash. Price fell 11.92 → 11.73 in first 5 min (−1.6%). The +$0.70 P&L is noise — likely instant scratch or tiny share count. VCET was reading the down-trend correctly.

6. HL 2026-06-03 10:38 LONG +$12.25 → wrong_side_escape HIGH — but see note

RVOL 0.99 · gap −4.69% · MTF −57 (weak) · no catalyst

tstateslope°bw%exp
−5TREND_DOWN−4.943.590
0TREND_DOWN−3.163.500
+5TREND_DOWN−2.043.3720
+10TREND_DOWN−1.223.310
+15TREND_DOWN+0.693.290
+25TREND_DOWN+2.493.3015

Important nuance: Slope was DECELERATING through the snap window (−4.94° → +2.49°). Slope CROSSED ZERO around +15min. Mechanically this is closer to a reversal-from-down entry — the down-trend was rolling over while VCET's state lagged (state-sticky behavior). Rule C called wrong_side_escape (state-based and technically correct); trader's intent looks like early reversal entry. MTF at −57 (weak) consistent with a fading trend.

Verdict: Rule says wrong_side_escape HIGH; mechanically defensible as reversal-attempt. Profited because slope crossover happened within the management window.

4. Per-trade summary table

TradeProposed mechanismConfidenceEvidenceP&L
AAOI 4/10 LONGinsufficient_data—NO_BAR (pre-kti)+$136.29
WULF 4/13 LONGinsufficient_data—NO_BAR+$18.48
CRWV 4/13 LONGinsufficient_data—NO_BAR+$9.21
RGTI 4/15 LONGinsufficient_data—NO_BAR+$8.92
CAR 4/16 SHORTinsufficient_data—NO_BAR+$53.76
SPY 5/1 LONGinsufficient_data—NO_BAR+$8.58
LUNR 5/26 LONGinsufficient_data—Entry TREND_UP (already trending); no rule fits−$11.76
DAL 5/26 13:56 LONGvwap_rotation_scalpHIGHROTATION 25min, slope ≈ 0, bw 1.49 flat+$68.25
DAL 5/27 13:56 LONGvwap_rotation_scalpHIGHROTATION 25min, slope −1.34 → −1.40, bw 2.86 → 3.05+$22.50
JOBY 5/28 09:31 LONGtrue_pull_in_to_trendHIGHbw 2.98 → 6.42, slope +1.6° → +49°, state → TREND_UP @ +15+$12.45
BB 5/28 11:56 LONGinsufficient_data—Entry already TREND_UP; trend continuation entry+$18.40
DAL 5/28 13:56 LONGvwap_rotation_scalpHIGHROTATION + compression (comp=30), bw 2.84 → 2.75+$5.59
JOBY 5/29 09:31 LONGwrong_side_escapeHIGHLONG into TREND_DOWN; slope −3.9° → −52.7°; barely escaped+$0.70
INTC 6/2 LONGinsufficient_data—Entry already TREND_UP; trend continuation entry+$4.05
HL 6/3 LONGwrong_side_escapeHIGHLONG into TREND_DOWN, slope decelerating (reversal interpretation also defensible)+$12.25

5. Answers to the four questions

Q1. Is the profitable vwap-pull-in -> trend bucket mostly...

C — MIXED bucket.

Of 15 trades: 1 true conversion (6.7%, 3.4% of $) · 3 rotation scalps (20%, 26.2% of $) · 2 wrong-side escapes (13.3%, 3.5% of $) · 3 trend-continuation entries (20%, 2.9% of $; includes the bucket's only loser) · 6 unclassifiable due to bar history (40%, 64% of $).

The bucket's profitability is real, but the dollar edge inside the classifiable subset is driven by rotation scalps, NOT by trend conversion. The 1 confirmed conversion made only $12.45 — less than 4% of total dollar edge.

Q2. Should vwap-pull-in -> trend remain a canonical slug for now?

YES — with a documented mechanism caveat (already added 2026-06-04 to Quality Gate, Cheatsheet, MGP block).

Keep it because: it IS a profitable pattern (+$367, 93% win rate); the slug is wired into Phase 2 backfill (297 rows), workflow docs, performance report; removing/renaming mid-stride would break dashboards.

Needs the caveat because: the name overstates the trend-conversion mechanism by ~5× in the classifiable subset.

Q3. Should we add mechanism_slug later?

YES — this audit provides the empirical basis.

The 6 classifiable trades show distinct mechanism signatures detectable from VCET trajectory at 5-min granularity, with different P&L profiles (rotation +$32 avg, true conversion +$12 avg, wrong-side +$6 avg), implying different management styles.

Sequencing: wait for ~15-25 more classifiable trades (3-6 months) before committing schema. Today's n=6 is too small.

Q4. Should the Quality Gate be updated to warn that "Pull-In → Trend" is outcome-based until mechanism is confirmed?

YES — done. Caveat added 2026-06-04 to Quality Gate, Cheatsheet, and MGP block. Mechanism-aware management rule section also added to all three.

6. Practical trading takeaway for tomorrow

Watch the FIRST 5 MINUTES after a Pull-In entry — the mechanism reveals itself fast.

Stop-rule suggestion: if 5 min post-entry the band hasn't expanded and slope hasn't accelerated, treat as rotation scalp regardless of slug. Manage to band. The breakout isn't coming.

7. Honest caveats

8. What did NOT change

Phase board unchanged.

Links