RVOL × Instrument-Type Cross-Tab

Validation add-on: does RVOL add independent signal once instrument type is controlled? · Built 2026-06-09 · Machine summary: vwap_pull_in_rvol_instrument_type_crosstab.json
Data: Liberator backfill (same as the gate audit). Plain vwap-pull-in n=132 · RVOL populated 108/132 (81.8%) · gap 132/132 · type classified (non-other) 107/132 · no DB query.
promoted-candidate working / context-only rejected insufficient sample
RVOL does NOT survive instrument-type control

Once you split by instrument type, RVOL stops separating winners from losers. Within megacap-tech the <1.0 RVOL bucket is the BEST (+$145.86 / 62.5% / n=16) — the exact opposite of the "calmest is worst" headline. The negative <1.0 story was a symbol confound: IBIT alone lost −$347 over 4 low-RVOL trades (including the −$216 6/3 outlier). RVOL fails even as a floor. Recommendation: D — remove RVOL from the gate; keep it only as non-gating liquidity context.

🔍 The smoking gun

The prior audit flagged "<1.0 RVOL = worst bucket (−$320)." This cross-tab shows that loss lives almost entirely in crypto-proxy / IBIT, not in low RVOL. Control for type and the effect reverses in the good bucket. RVOL was a proxy for "is this IBIT / a crypto-proxy" — and instrument type + symbol rules capture that directly and honestly.

Instrument type — overall (plain Pull-In)

Instrument typenWin %Total P&LAvg
megacap_tech5151.0%+$151.91+$2.98
crypto_proxy_etf3138.7%−$179.02−$5.77
smallcap_highbeta1631.2%−$113.73−$7.11
energy_commodity n=9911.1%−$57.07−$6.34
other_unknown2512.0%−$100.22−$4.01

Megacap-tech is the only net-positive type. Everything else loses — including the 25 unclassified "other" names.

Q1 & Q2 · Floor test: RVOL <1.0 vs ≥1.0 within each type

If RVOL had a real floor effect, <1.0 should be worse than ≥1.0 in every type. It isn't.

Type<1.0 n<1.0 Win%<1.0 P&L≥1.0 n≥1.0 Win%≥1.0 P&L
megacap_tech1662.5%+$145.862744.4%−$8.33
crypto_proxy_etf1241.7%−$292.851338.5%+$7.67
smallcap_highbeta812.5%−$84.872—−$19.95
energy_commodity714.3%−$43.251—−$2.50
other_unknown100.0%−$45.511216.7%−$44.62

Read: in megacap-tech, <1.0 is +$145.86 (the best, not the worst). A <1.0-avoid floor would have killed the strategy's best trades. The negative <1.0 is concentrated in crypto-proxy (−$292.85).

What drives the <1.0 loss (plain Pull-In)

SymbolTypenP&L
IBITcrypto_proxy_etf4−$347.00
RGTIsmallcap_highbeta3−$38.98
CVXenergy_commodity4−$33.31
KOPNsmallcap_highbeta1−$28.80

IBIT (4 low-RVOL trades, incl. the −$216 6/3 outlier) is the entire <1.0 story. The other <1.0 trades net roughly flat-to-positive.

Q3 · Does high RVOL (>2.5) underperform within type?

No consistent penalty — and cells are tiny.

Type>2.5 n>2.5 P&Lread
megacap_tech5−$18.15mild, small n
crypto_proxy_etf2+$15.22positive, tiny
smallcap_highbeta1−$28.30n=1
other_unknown5−$26.97small

Read: no clean high-RVOL penalty. On all trades, the >2.5 × gap>1.5 cell is actually +$223.83 / n=27 — high RVOL is not intrinsically bad; the old story was outlier/symbol-driven. insufficient sample per cell

Q5 · RVOL × gap grid (plain Pull-In) — total P&L (n)

RVOL ↓ / Gap →|gap| ≤ 0.5%0.5–1.5%> 1.5%
< 1.0−$229.31 (17)+$29.96 (17)−$121.27 (19)
1.0–1.5+$41.38 (10)−$4.43 (6)−$60.75 (10)
1.5–2.5+$18.65 (4)+$80.27 (5)−$84.65 (7)
> 2.5+$15.78 (3)−$30.57 (2)−$43.41 (8)
missing+$117.10 (6)+$4.97 (8)−$31.85 (10)

Read: the gap>1.5% column is negative at every RVOL row — gap is the real RED line. RVOL does not separate down any gap column. The one scary cell (<1.0 × ≤0.5 = −$229) is the IBIT/crypto confound again, not a calm-tape effect.

Q6 · RVOL × post-entry state (plain Pull-In)

RVOLFade — Win% / P&L (n)Early expansion — Win% / P&L (n)
< 1.011.1% / −$400.08 (18)80.0% / +$55.64 (5)
1.0–1.50.0% / −$103.52 (8)66.7% / +$53.98 (3)
1.5–2.516.7% / −$56.12 (6)80.0% / +$4.90 (5)
> 2.50.0% / −$36.98 (3)66.7% / +$19.98 (3)

Read: fade loses and expansion wins at every RVOL bucket. Post-entry state fully dominates RVOL; RVOL adds nothing on top of it.

Answers

Q7 · Revised Quality Gate recommendation

Choice: D — Remove RVOL from the Green/Yellow/Red gate until a cleaner, type-controlled RVOL backfill shows it adds something. RVOL may be retained only as a non-gating liquidity/activity context readout (the legitimate kernel of option C) — never as a winner/loser threshold. The <1.0 = avoid floor is rejected (confounded).

OptionVerdict
A. Keep RVOL as a major axisrejected no independent signal
B. Demote to secondary confidence fieldpartial — still implies edge value RVOL lacks
C. RVOL only as a floorrejected as an edge floor; only the pure-liquidity reading survives
D. Remove from gate; keep as contextCHOSEN

Q8 · Exact proposed gate (RVOL removed)

LayerRuleStatus
RED — primaryabs(gap_pct) > 1.5%promoted-candidate
RED — secondaryinstrument type in {crypto-proxy/ETF, small-cap/high-beta, energy/commodity, other} UNLESS post-entry expansion confirms (megacap-tech is the only net-positive type)promoted-candidate
GREENcontrolled gap (|gap| ≤ 1.5%, ideally ≤ 0.5%) AND megacap-tech AND no 3-min fade (no RVOL condition)working
YELLOWcontrolled gap but lower-confidence instrument type OR missing dataworking
HARD EXIT / RED FLAG3-minute fade (wins ~0–17% at every RVOL bucket)promoted-candidate
TREND HOLD ONLY IF10–15 min expansion confirmation + slope acceleration + CBO/range-break evidence if availablepromoted-candidate
RVOLnon-gating liquidity/activity context display onlycontext-only

Priority order, confirmed by this cross-tab: 1) gap   2) instrument type   3) post-entry behavior   4) RVOL only as activity/liquidity context.

Validation status (honest)

Sample-size & method caveats

Report-back summary

Source & related

Read-only validation add-on · computed from on-disk Liberator artifacts, no DB · no edits to /strategies/, /scorecards/, gen_scorecard.py, _scientific_overlay.js, taxonomy, or DB · no Stage B reclassification · thresholds remain working/promoted-candidate pending out-of-sample.