VCET v1.1 / CBO — Mechanism Validation Plan

Date 2026-06-04 · Status PLAN ONLY — no execution, no DB writes, no VCET engine edits, no taxonomy / rename / mechanism_slug · Supersedes the state-at-entry hypothesis of VCET_AT_ENTRY_VALIDATION_PLAN (Phase 4), which Phase 4a falsified · Engine: VCET v1.1 (13 checks + CBO) once a verified package lands — see §9 acceptance gate.

⚠ The question changes

Old question (Phase 4, falsified): “Was the trade in TREND_UP / EXPANSION_CONFIRMED at entry?” — Phase 4a showed it was not: PI→Trend winners clustered in ROTATION at entry, and the single biggest Pull-In loss (IBIT 6/3, −$216) entered in EXPANSION_CONFIRMED. The state label at one instant does not separate winners from losers.

New question (this plan): “Can VCET v1.1 + CBO identify the mechanism of the trade?” The conversion to trend happens after entry, so the signal is in the trajectory and the channel-break geometry, not the entry snapshot. The job of v1.1/CBO is to classify which of four mechanisms a trade actually was — so that management (not the go/no-go) can be matched to mechanism.

Primary filter stays the Quality Gate (RVOL + gap). VCET v1.1/CBO is tested here as a secondary confirmation / mechanism-labelling layer, never as the entry trigger. See §7.

1. Why state-at-entry was not enough

Phase 4a (VCET Pull-In At-Entry Pilot) ran v1.0 at slope_trend_thresh=2.0 over the 55 trades with bars in the kti window. The original four hypotheses (converters live in expansion; leakers live in indecision) did not hold:

Phase 4 expectedPhase 4a actual
≥70% of PI→Trend wins in EARLY_EXPANSION / EXPANSION_CONFIRMED / TREND_* at entry5 of 9 PI→Trend wins entered in ROTATION; the two biggest (DAL +$68, +$22) both ROTATION_FLAT_BANDS
Plain-PI losses concentrate in indecision statesEvery state bucket lost money; the −$216 IBIT loss was the lone EXPANSION_CONFIRMED row — an opposite-side break taken long
State label is the discriminatorState label is a lagging snapshot. The signal that survives is numeric geometry (slope sign, band width, score) and state change after entry

The two structural reasons state-at-entry fails

  1. Timing. “Pull-In → Trend” is defined by a conversion that occurs over the 5–25 minutes following entry. A snapshot at t=0 structurally cannot see it. (Pilot §9.4 explicitly calls for testing state trajectory, not state-at-entry.)
  2. Missing operator. v1.0 has no channel-break confirmation. It cannot tell a fresh break (conversion) from a stale break already in progress (continuation), nor confirm that a break went the opposite way (wrong-side). CBO — absent in v1.0 — supplies exactly that. The pilot named it: “CBO would refine EXPANSION_CONFIRMED and might have caught the IBIT −$216 as a wrong-side trade.”

2. The four mechanisms to classify

The Mechanism Audit (v1.0 + hand-coded rules A/B/C/D) found the vwap-pull-in -> trend bucket is a MIXED bucket: 1 true conversion, 3 rotation scalps, 2 wrong-side escapes, plus a 4th mechanism the rule set could not model (3 trades that entered already TREND_*). These are the four classes v1.1/CBO must separate.

MechanismState at entryPost-entry trajectoryCBO signature (v1.1)Reference tradeManagement
true_pull_in_to_trend ROTATION / COMPRESSION / EARLY_EXPANSION Band width expands ≥1.5× in 5 min; slope accelerates ≥+5° in trade dir; expansion_score crosses ≥50; state → TREND_* within 15 min Break in trade direction, AFTER entry (offset +5..+15) JOBY 5/28 +$12.45 Trend-hold; trail wide; let the runner run
vwap_rotation_scalp ROTATION (often ROTATION_FLAT_BANDS) Stays ROTATION through +25; |slope|<2°; band width flat (≤1.2×); expansion_score<50 throughout No break in any direction through +25 (price stays inside channel) DAL 5/26 +$68.25 Target the opposite band; take the rotation; don’t extend stops
wrong_side_escape Any (often ROTATION or sticky TREND) State / slope move AGAINST trade dir; profit is tiny size, fast scratch, or lucky bounce Break OPPOSITE trade direction within +15 JOBY 5/29 +$0.70 · IBIT 6/3 −$216 “Do not let outcome validate the setup.” Exit / reduce fast
trend_continuation_entry 4th — new Already TREND_* matching trade dir No fresh expansion; trend often decelerating; band width flat/established Break in trade direction occurred BEFORE entry (t≤0); none fresh after LUNR 5/26 (−$11.76, the bucket’s only loser) · BB 5/28 · INTC 6/2 Not a Pull-In at all; continuation rules — don’t expect a fresh leg
insufficient_data No bars in kti window / engine error / session boundary / too few bars to form the CBO channel — No mechanism inferred

In the audit, trend_continuation_entry fell into insufficient_data because v1.0 saw only “already TREND_UP at entry” with no way to date the break. Giving it a real label is the single clearest demonstration of CBO’s added value (§6, V-Q2).

3. The CBO insight (the crux)

CBO adds two facts v1.0 never had: break DIRECTION and break TIMING-relative-to-entry

CBO (Channel-Breakout Operator, Pardo Ch. 28) tracks whether price has broken its recent N-bar high/low channel, which way, and when. Per the v1.1 config it runs as a confirmation layer, not a hard gate: enable_cbo=true, require_cbo_for_expansion_confirmed=false. Those two facts turn the four mechanisms — which overlap badly on a single state label — into a clean decision surface.

CBO break directionCBO break timing vs entry→ Mechanism
Same as trade dirAFTER entry (+5..+15)true_pull_in_to_trend — the break IS the conversion; geometry confirms it
Same as trade dirBEFORE entry (t≤0)trend_continuation_entry — entered into an already-confirmed break
Opposite trade dirAny (within +15)wrong_side_escape — the channel broke against you (IBIT)
NoneNo break through +25vwap_rotation_scalp — price never left the channel

This is why the reframe matters. Phase 4a tried to separate four mechanisms with one categorical field (state_at_entry) and they collapsed onto each other in ROTATION. Break-direction × break-timing is a two-axis surface; the four mechanisms occupy four distinct cells. State trajectory and band/slope geometry then confirm the cell rather than carry the whole decision.

4. Mechanism decision logic (CBO-augmented)

Apply in priority order. The order is the audit’s C > A > B > D, with trend_continuation_entry inserted as a CBO-enabled class above the conversion test (a pre-entry break must be ruled out before a post-entry break can be called a conversion).

given trade with direction D in {LONG, SHORT}, entry t0,
      snaps at offsets {-5, 0, +5, +10, +15, +25} min,
      CBO fields per bar: cbo_break_dir in {NONE, UP, DOWN}, cbo_channel_pos

# 1. wrong_side_escape  (safety first)
if cbo_break_dir opposite D within +15
   OR state -> TREND_* opposite D within +15
   OR slope accelerates >=15 deg against D within +5:
       -> wrong_side_escape
       (HIGH if CBO opposite-break confirmed; MED if only slope/state)

# 2. trend_continuation_entry  (NEW - requires CBO timing)
elif cbo_break_dir == D at some offset t <= 0   # break predates entry
     AND state already TREND_D at entry
     AND no fresh expansion after entry (bw not >=1.5x; expansion_score stays < 50):
       -> trend_continuation_entry
       (HIGH if pre-entry break + already TREND + decelerating slope)

# 3. true_pull_in_to_trend  (conversion)
elif entry state in {ROTATION, COMPRESSION, EARLY_EXPANSION}
     AND cbo_break_dir == D first confirmed in +5..+15   # fresh break after entry
     AND band_width(+5) >= band_width(0) * 1.5
     AND slope(+5) - slope(0) >= +5 deg (D-signed)
     AND expansion_score crosses >= 50
     AND state transitions toward EXPANSION_CONFIRMED / TREND_D within 15 min:
       -> true_pull_in_to_trend
       (HIGH if CBO break + >=3 of 4 geometry sub-checks align)

# 4. vwap_rotation_scalp
elif cbo_break_dir == NONE through +25
     AND state stays ROTATION at +15 and +25
     AND |slope| < 2 deg at 0,+5,+10,+15
     AND band_width(+25) <= band_width(0) * 1.2
     AND expansion_score < 50 at all snaps:
       -> vwap_rotation_scalp
       (HIGH if all hold cleanly)

# 5. insufficient_data
else: -> insufficient_data   # no bars / engine error / session boundary / channel not formed
Classv1.0 rule (audit)What CBO adds in v1.1
wrong_side_escapeSlope accel ≥15° opposite, or opposite-dir TREND within 15 minA crisp binary — the channel broke the other way — catches IBIT-type losses earlier and cleaner than waiting for a 15° slope swing
trend_continuation_entryNone — dumped into insufficient_dataPre-entry break timing makes it classifiable for the first time; pulls LUNR/BB/INTC out of the unknown bucket
true_pull_in_to_trend≥3 of 5 fuzzy geometry sub-checksThe CBO break-in-direction becomes the spine; geometry confirms rather than carries. Pins the conversion instant
vwap_rotation_scalpInfer rotation from flat slope + flat bands“CBO never fired” is a positive signal for rotation, not an inference from absence

5. Validation design

ElementSpecification
PopulationAll canonical_strategy_slug = 'vwap-pull-in -> trend' (n=15) as the target; plain vwap-pull-in (n=127) and chase-extended (n=56) as contrast / control. Read-only on main.trade_context. Same population as Phase 4a.
Barskti five_min, 14-day window. NO_BAR risk is the headline limitation — in the audit 6 of 15 PI→Trend trades (64% of the bucket’s $) predate the window and cannot be classified. The bar-source decision (extend window / external feed / accept pilot scope) stays a James decision, carried over from Phase 4 §2.2. Default: classify what has bars; grow the sample over time.
EngineVCET v1.1, verified (13 acceptance checks pass + CBO present — §9 gate). enable_cbo=true, require_cbo_for_expansion_confirmed=false. In-memory only — do NOT build main.vcet_bars until v1.1 proves discriminating power (Phase 4a recommendation).
ThresholdsPer-liquidity-tier slope_trend_thresh rather than a single global value (the 2.0 confound named in Phase 4a): large-cap individuals ~2.5, broad ETFs ~1.5–2.0, small-caps ~4–5 (06/01 calibration). CBO channel length per the shipped v1.1 config (Pardo default, confirmed from the package — not assumed), tuned alongside per tier. Overrides build a non-default VCETConfig at the call site; engine code untouched, Pine parity preserved for default runs.
Features per tradeTrajectory snap at {−5, 0, +5, +10, +15, +25}: state, vwap_slope_deg, vwap_band_width_pct, expansion_score, range_compression_score, vcet_score, price_distance_from_vwap_sd, PLUS the new CBO fields cbo_break_dir, cbo_break_offset, cbo_pre_entry_break (bool), cbo_channel_pos.
ClassifierThe §4 CBO-augmented decision logic, priority C > continuation > A > B > D. Emits mechanism + confidence in memory only.

6. Validation questions & success criteria

A validation plan must say what would count as success and what would kill the idea. These six questions test whether v1.1/CBO genuinely classifies mechanism — not merely whether it emits labels.

V-Q1 — Separation

Do the four mechanism classes occupy distinct, non-contradictory CBO + geometry cells?

PASS: no trade lands in a contradictory cell — e.g. no “rotation” trade that actually threw a CBO break, no “conversion” whose break was pre-entry. Each class’s defining features hold at the trade level.

V-Q2 — CBO beats v1.0 (the headline test)

Does CBO reclassify trades that v1.0 could not? Specifically, does pre-entry-break detection pull the three known trend_continuation_entry trades (LUNR 5/26, BB 5/28, INTC 6/2) out of insufficient_data into a real class?

PASS: ≥ the 3 known continuation trades correctly relabelled via pre-entry break; the audit’s 6 classifiable trades keep their labels under the CBO logic (no regressions).

V-Q3 — Wrong-side catch

Does an opposite-direction CBO break flag the known wrong-side trades (JOBY 5/29, HL 6/3) and the IBIT 6/3 −$216 plain-PI loss — earlier or cleaner than v1.0’s 15° slope-accel rule?

PASS: CBO direction at the break bar is opposite trade direction for these trades, confirmed at an offset ≤ where v1.0’s slope rule would have fired.

V-Q4 — Conversion precision (single-case sanity)

For the one known true conversion (JOBY 5/28), does CBO fire an in-direction break at ~+10/+15 coincident with the band expansion and TREND_UP transition?

PASS: CBO break offset aligns within ±5 min of the state→TREND transition. (n=1 today — a sanity check, not a base rate.)

V-Q5 — Secondary, not primary

Used as a layer on top of the RVOL+gap Quality Gate, does the mechanism label add management-relevant discrimination without contradicting the gate?

PASS: mechanism classes correlate with management outcome (conversion trades hold to +2 SD; rotation trades mean-revert at the band), AND the gate’s GREEN/RED zones remain the first filter unchanged. VCET v1.1/CBO informs how to manage, not whether to enter.

V-Q6 — Falsification (CMT Ch. 36 discipline)

State the null and what would kill the idea: if CBO break direction/timing does not separate the mechanisms — rotation trades also throw breaks, or continuation trades show no pre-entry break, or conversions show no aligned break — then v1.1/CBO does not add mechanism-classification power.

KILL CRITERION: if ≥2 classes are indistinguishable on the CBO surface across the classifiable sample, stop — retain RVOL+gap + the v1.0 trajectory heuristics, do not propose mechanism_slug, and reconsider only if a later v1.x changes the operator.

7. The Quality Gate stays primary

This plan does not move VCET ahead of the gate. The VWAP Pull-In Quality Gate — RVOL + gap — remains the go/no-go first filter:

ZoneConditionAction
GREENRVOL 1.0–1.5 AND gap < 0.5% AND MTF grand > 25Take; hold for conversion if price closes above +2 SD
YELLOWRVOL 1.5–2.5 OR gap 0.5–1.5%Reduced size only; must articulate thesis before entry
REDRVOL > 2.5 OR gap > 1.5% OR symbol in {IBIT, AMD, ETHU, RGTI}Stop / reclassify
Order of operations: RVOL + gap decides whether to enter. THEN the first 5 minutes of CBO + state trajectory tell you which of the four mechanisms you are in — and therefore how to manage. v1.1/CBO is a confirmation/management layer, never the trigger.

8. Sample-size honesty & the commit gate

9. Safety & the v1.1 acceptance gate

Verify the package is actually v1.1 BEFORE running anything

On 2026-05-25 the on-disk package was v1.0 and blocked the whole plan. v1.0 has no CBO and only 7 automated checks — running this validation on it would write meaningless “CBO” columns. Before step 1, confirm the package is genuinely v1.1:

Hard constraints (VCET project rules): no writes to any prod schema · do NOT modify vcet_engine.py, vcet_indicator_v1.pine, or anything under docs/ (versioned + parity-audited) · new helpers are NEW files under python/ · dry-run / read-only first; idempotent inserts only if/when a write step is ever approved · show every script for review before running · the Phase 2 (build main.vcet_bars) decision is joint, not autonomous.

10. Deliverable of the run (when approved)

One read-only CSV + one static HTML evidence table — same shape as the Mechanism Audit, one row per classifiable trade:

Column groupFields
Tradesymbol, date/time, side, net P&L
Quality GateRVOL, gap%, zone (GREEN/YELLOW/RED)
CBO (v1.1)cbo_break_dir, cbo_break_offset, cbo_pre_entry_break
Trajectorystate & slope & band_width & expansion_score at each of the 6 offsets
Verdictmechanism, confidence (HIGH/MED/LOW), and a “CBO vs v1.0 Δ” column showing what CBO changed vs the audit’s v1.0 call

No DB writes. No engine edits. No Scorecards refresh. New file at vcet/python/… + one HTML at /system/VCET_V1_1_MECHANISM_VALIDATION_REPORT.html.

11. What this plan does NOT change

12. The one-line question this plan answers

“Can CBO break-direction and break-timing — combined with state trajectory — tell us which of the four mechanisms a Pull-In actually was, well enough to manage by mechanism rather than by an outcome-based slug name?”

If yes → VCET v1.1/CBO becomes the secondary management layer behind the RVOL+gap gate, and (after the §8 commit gate) mechanism_slug earns its schema. If no (V-Q6 kill) → RVOL+gap + v1.0 trajectory heuristics stand, and nothing in the taxonomy moves.

Links