Pull-In / Strategy Bucket Research Hub
Last updated 2026-06-04 · Lineage of VWAP Pull-In and canonical-strategy-bucket research artifacts ·
All pages read-only on production (no DB / classifier / taxonomy / VCET changes).
This hub indexes the research arc that has been auditing the canonical strategy buckets — what each slug actually contains, where the edges come from, and where the labels overstate the mechanism. Read the workflow docs (top section) before any trade; read the audits (middle) to understand what the slug names actually mean today; read the plan (bottom) for what comes next.
Workflow documents (in-use, pre-trade)
Use these before clicking buy. The Quality Gate is the primary filter; the cheatsheet is print-ready; the MGP block is copy/paste for the daily morning email.
VWAP Pull-In Quality Gate
workflow
Full pre-trade gate (GREEN / YELLOW / RED) on RVOL + gap + MTF, with §3 reclassification rules for hot tape, §5 mechanism-aware post-entry management, and a research-grade hypothesis section.
Includes the 2026-06-04 mechanism caveat: vwap-pull-in -> trend is outcome-based, not mechanism-confirmed.
Quality Gate Cheatsheet
workflow
Print-ready one-page summary of the gate, the 7-question pre-trade checklist, and the post-entry management rule.
Source for the Intraday Checklist chart-side card.
MGP Block (Morning Game Plan)
workflow
Copy/paste block for the daily Morning Game Plan email — gate, decision rule, post-entry management, one-sentence rule.
Drop into the Morning edition of the MGP each day.
Bucket audits & trajectory reviews
What each canonical-slug bucket actually contains. Each audit follows the same pattern: scope, raw evidence (CSV), provisional reclassification with source/proxy/heuristic-tagged rules, and an answer to "is this bucket profitable for the reason the slug name implies?"
chase-extended Bucket Audit
audit
58 trades, +$198.28. Headline: bucket is a mixed residual label hiding better mechanisms — true chases (n=2) are net negative; profit comes from 9 mislabeled trend_trade + event_driven candidates and one outsized NVDA outlier.
Answer: B + C, supported by D. Without 9 mislabeled trades, bucket nets −$135. Without NVDA 6/2 +$199, nets −$1.
Pull-In → Trend Mechanism Audit
audit
15 trades, +$367.67. Headline: bucket is MIXED — within the 6 classifiable trades, the dollar edge comes from rotation scalps (3 of 6), NOT trend conversions (1 of 6).
Answer: outcome-based slug; only 1 of 6 trades is mechanically what the name claims.
PI → Trend ROTATION Winners Review
trajectory review
Trigger document for the mechanism audit. The 5 ROTATION-at-entry Pull-In → Trend winners trajectory-snapped at −5, 0, +5, +10, +15, +25 min. Only 1 (JOBY 5/28) was a true ROTATION → TREND_UP conversion.
Established the "first 5 minutes reveal the mechanism" finding.
VCET Pull-In At-Entry Pilot (Phase 4a)
pilot
kti-only in-memory VCET pilot across the Pull-In family. Original hypothesis (PI→Trend winners cluster in EARLY_EXPANSION/EXPANSION_CONFIRMED/TREND_UP at entry) was NOT supported — they entered in ROTATION. VCET numerics show signal even though categorical label doesn't.
Source dataset for the trajectory + mechanism audits.
Bucket performance reports
Cross-bucket P&L summaries — the canonical taxonomy view of the dataset.
Canonical Strategy Performance Report
cross-bucket
~310 trades summarized by canonical slug, with the VWAP Pull-In split (plain −$364 / Pull-In → Trend +$367) headline. Read with mechanism-caveat lens.
Bucket-level P&L is real; bucket-level mechanism is the open question the audits address.
VWAP Pull-In Split Diagnostic
diagnostic
RVOL + gap discriminator analysis that produced the Quality Gate rules. Plain PI losers cluster at RVOL ~4 / gap +2%; PI → Trend winners cluster at RVOL ~1.1 / gap −0.9%.
Source data for the GREEN / YELLOW / RED gate definitions.
Refinement plans
What changes if and when we tighten the classifier — additive, non-breaking. Plan-only artifacts; nothing implemented until each plan has explicit approval.
Classifier Refinement Plan
plan
8-section plan: current problem (outcome-based Priority 3/5 rules); mechanism distinction (A/B/C); proposed additive mechanism_slug field + 7 companions; rule candidates; impact on existing reports; recommendation A.
Final recommendation: keep canonical slugs, add additive mechanism_slug; wait for 15-25 more classifiable trades before committing schema.
VCET v1.1 Mechanism Acceptance Test Plan
plan
Test design that decides whether v1.1 + CBO earns a workflow slot as a mechanism classifier. 8 hypotheses, 7 mechanism classes (true_pull_in_to_trend / vwap_rotation_scalp / trend_continuation_entry / wrong_side_escape / true_chase_extended / range_break_hitchhiker / rubberband_backside), 6 acceptance criteria (A1-A6), adversarial v1.0 baseline, PASS/FAIL/CONDITIONAL verdicts.
Blocker: v1.1 engine not on disk. Plan stays inert until source arrives; bar archive blocker noted for rare-class statistical power.
What ties this together
Every audit so far has reached the same shape of conclusion: the canonical taxonomy is correct, but the classifier rules that populate it are outcome-based, not mechanism-based. The Pull-In → Trend bucket mixes true conversions with rotation scalps; the chase-extended bucket mixes true chases with mislabeled trend trades and event-driven setups. The fix is not to rename the slugs (which would break dashboards) but to add an additive mechanism_slug field once enough bar-history data has accumulated to populate it reliably. Until then: use RVOL + gap as the primary pre-trade filter, read the first 5 minutes post-entry to identify the actual mechanism, and don't let the slug name validate management decisions after the fact.