Honest framing up front: the direction is solid — calm tape wins, hot tape loses, and how price behaves in the first few minutes after entry separates winners from losers. The exact numbers (RVOL 1.5 vs 2.5, etc.) are working rules pulled from group averages on a small winner sample (15 Pull-In→Trend trades). Use them as a discipline, not as gospel.
The 60-second script. If you can’t answer all five fast, the trade isn’t ready — pass.
*MTF/MA stack is a macro context filter, not a per-trade discriminator — in the data it separates the two slugs but not winners from losers within plain pull-ins. Use it to read regime, not to greenlight a trade. RVOL is the primary filter, but it’s missing on many historical rows; if RVOL is unavailable, lean on gap% + your visual tape read.
09:00–10:00 ET held 91 of 127 plain pull-ins and −$401 of P&L proven. The hour isn’t the enemy — selection inside it is. In that window, demand RVOL < 2 AND gap < 1% or pass.
| Condition | 🟢 Green | 🟡 Yellow | 🔴 Red |
|---|---|---|---|
| RVOL | 1.0–1.5 | 1.5–2.5 | > 2.5 |
| Gap % | < 0.5% | 0.5–1.5% | > 1.5% |
| Time of day | After 10:00, or calm open | Open hour but controlled | 09–10 ET hot / emotional |
| VWAP slope | Flat-to-rising (with dir) | Rising, needs to hold | Against you, or vertical |
| Band width | Normal / contained | Starting to expand | Already blown out (> +2 SD) |
| Price location | At / near VWAP pullback | +1 SD, still structured | Extended far from VWAP |
| MTF context | Supportive | Mixed | Conflicted / weak |
| Action | Take, normal size | Half size + confirm | Pass / reclassify |
Can’t answer all five quickly? Pass. The setup may not exist on this name today.
“Pull-In → Trend” is the strongest outcome bucket in the data (+$367, 93% win) proven — but when the mechanism was actually reviewed, only 1 of 6 classifiable trades was a true trend conversion; most were rotation scalps, and 64% of the bucket’s dollars are too old to verify proven. Translation: don’t hold a plain pull-in like a trend trade just because it might earn the label. Make it prove the trend after entry.
| Gate | Size | Grade | Daily-Stop units* |
|---|---|---|---|
| 🟢 GREEN | Full planned size | A / A+ | up to 1 full unit |
| 🟡 YELLOW | Half size, confirmation-only | B | ≤ ½ unit |
| 🔴 RED | No trade (or tiny review-only) | C / no-grade | 0 (or ¼ unit if studying tape) |
*1 unit = your pre-defined per-trade max risk (a fraction of your Daily Stop at /daily-stop/). The grade→size mapping is a working convention — confirm your own unit size. The point is simple: full size only where the edge is (green), reduced where it’s muddy (yellow), nothing where it leaks (red).
Once you’re in, the tape tells you which trade you actually have. Manage to behavior, not to the label.
Illustrative — how the framework reads at the click. (MSFT and IBIT mirror real history in the data; AAPL/NVDA are hypothetical.)
This playbook is not meant to sit alone in /system/. It is the bridge between the two halves of the system: the Strategy Library says what you’re allowed to trade; the Scorecards grade how you executed it. This card supplies the gate, the classification, and the grade logic that connect them.
| Layer | Its job | Where it lives | What this playbook feeds it |
|---|---|---|---|
| Strategy Library | The WHAT — source of truth for the setups you trade | /strategies/ (data: /system/strategies.json; display mirrors _scientific_overlay.js) | Becomes the official entries for the 4 slugs below |
| Execution Scorecards | The HOW — grades execution quality per trade | /scorecards/ (classifier _scientific_overlay.js#classifyEntry) | gate_color, setup_classification, and the grade logic |
| VCET / live classifier | The WHEN — real-time gate + label at the ticket | VCET engine (v1.1 + CBO) | reads the JSON gate; supplies post-entry state/band fields |
| AMS system hub | The umbrella architecture | smb.a1prospect.com/ams/system/ | registers this as a reusable component |
| Draft slug | Live DB slug | What it actually is |
|---|---|---|
vwap_pull_in | vwap-pull-in | Tradeable setup — calm-tape pull-in |
vwap_pull_in_to_trend | vwap-pull-in -> trend | Tradeable, but outcome-based — earns the hold by confirmation |
chase_extended | chase-extended | Warning overlay, not a setup (off-book sentinel, not an edge) |
pass_no_trade | — | Decision state — a logged no-trade, not a strategy |
Draft object: vwap_pull_in_strategy_library_draft.json — full library metadata for all 4 (setup_type, market_context, required_conditions, disqualifiers, triggers, confirmation, risk, management, invalidation, scorecard_fields, reason_codes, validation_status), aligned to the real strategies.json schema. It is a draft: the underscore-vs-hyphen slug form needs reconciling, and chase_extended / pass_no_trade belong on the warning/decision axis, not the setup axis, before anything is promoted.
The goal is to turn this into structured strategy metadata that the Library, classifier, scorecards, and VCET alerts all read from one source — so the gate on this page is the same gate that fires live and the same gate the scorecard grades against. Promotion is gated on the backtest below.
For every VWAP Pull-In attempt (passes included), the scorecard should capture:
| Field | Values |
|---|---|
setup_classification | plain_vwap_pull_in / pull_in_to_trend / chase_extended / pass_no_trade |
gate_color | green / yellow / red |
rvol_at_entry | numeric |
gap_pct | numeric |
vwap_slope_state | rising / flat / falling |
band_width_state | contracting / flat / expanding / blown_out |
price_location_vs_vwap | signed SD offset at entry |
first_3min_result | holding / fading / expanding |
first_5min_state | rotation / early_expansion / wrong_side |
first_10_15min_confirmation | expansion_persisted / expansion_decayed / never_expanded / failed |
management_mode | rotation_scalp / trend_hold / failed_pull_in_exit / no_trade |
rule_followed | true / false |
mistake_type | chased_hot_tape / oversized_yellow / traded_red_gate / held_rotation_like_trend / ignored_3min_failure / no_clear_invalidation / good_execution |
improvement_note_plain_english | free text |
Additive scorecard columns — design only. No gen_scorecard.py or _scientific_overlay.js change until approved.
Profitable-but-poorly-graded trades are the priority review target — they reward bad process and hide the leak. Each maps to a mistake_type so the scorecard can tally which error costs the most. (Same logic in vwap_pull_in_actionable_gate.json → scorecard_integration.grade_logic.)
VCET should give the one-glance answer at the ticket, not make you read this page live. The eventual live output is simple:
| Field | Example value |
|---|---|
| Setup Type | Plain Pull-In / Pull-In→Trend / Chase-Extended / Pass |
| Reason Code | CALM_PULLIN / HOT_NEEDS_CONFIRM / HOT_PASS / DAMAGE_SYMBOL / FIRST_HOUR_HOT |
| Suggested Risk | Full / Half / None (grade A·B·C) |
| Management Mode | Rotation-scalp / Trend-trail-if-persists / Cut-on-fade |
| Invalidation | VWAP loss / lower band loss / 3-min fade |
Codes match vwap_pull_in_actionable_gate.json so VCET can read the gate directly. VCET (v1.1 + CBO) is also the eventual supplier of the post-entry state/slope/band fields that confirm Pull-In→Trend in real time — that’s the bridge between this card and the mechanism work.
These are working rules. Here’s exactly what to test next so they become proven rules (or get fixed):
Run read-only against the Liberator backfill. Verdict: PARTIALLY SUPPORTED — revise before promotion. Data-proven: |gap|>1.5% (the strongest filter), the 3-min fade-exit, 10–15 min expansion confirmation, and instrument type (megacap-tech works; crypto-proxy/small-cap fail). Not supported: the RVOL thresholds — bucketed, the calmest <1.0 RVOL bucket was the worst (−$320), so "hot tape = the leak" doesn't hold. Promote the proven parts; rework the RVOL axis. Still needs an out-of-sample pass.
The next step isn’t another backtest — it’s collecting 20–50 live attempts under the frozen rules. Use the tracker to log each pull-in (gate, classification, 3-min / 5-min / 10–15-min checkpoints, mistake tag). Promotion stays HOLD until the sample fills.
Before this becomes an official Strategy Library rule (promoted into strategies.json / the classifier), validate:
Until these pass, every threshold here stays working and the library draft stays validation_status: working_rule_needs_backtest. (Section K is the same list framed as the immediate research queue; this is the promotion gate into the live system.)
So this card stays honest. Source: main.trade_context through 2026-06-03 (n=310 classified). Everything below is from the linked research pages; nothing is invented.
| Bucket | n | Net P&L | Avg | Win% |
|---|---|---|---|---|
Plain vwap-pull-in (leak) | 127 | −$364.47 | −$2.87 | 35.4% |
vwap-pull-in → trend (edge) | 15 | +$367.67 | +$24.51 | 93.3% |
| Family net (both) | 142 | +$3.20 | — | 41.5% |
| Group | RVOL | Gap % |
|---|---|---|
| Plain PI losers (n=80) | 4.01 | +2.11% |
| Plain PI winners (n=45) | 1.38 | +0.35% |
| PI→Trend winners (n=14) | 1.14 | −0.88% |
The exact gate cut-points are working — these are averages, not validated per-trade boundaries.
Of 15 PI→Trend trades, only 6 were mechanically classifiable: 1 true conversion (+$12), 3 rotation scalps (+$96, the bulk of the dollars), 2 wrong-side escapes. Six more (+$235, 64% of the bucket) are older than the bar window and unverifiable. The bucket is real and profitable, but the name overstates “trend conversion” by roughly 5×. Hence: confirm after entry, don’t assume.
| First-window behavior | Win % | Avg P&L |
|---|---|---|
| Early expansion | 71% | +$485 |
| Fade | 26% | −$700 |
| Follow-through > +1σ | 79% | — |
| Follow-through < −1σ | 19% | — |
From the deeper Liberator full-window backfill (vwap-band-entry-surface research). This is why the management section leads with “cut fades early.”