D4 pre-read (draft v1) — for the 2026-10-22 quarterly review

Drafted 2026-07-25, three months early and deliberately so: the evidence is fresh, and drafting the review before the intervening quarter's results exist is itself a pre-commitment — October's session should diff reality against this document, not compose a narrative after the fact. Per Research Program §6 D4: reviews spend against §3 budgets, D1–D3 status, and whether the program document still describes reality.

1. Budget status (as of this draft)

2. What the quarter established (headline evidence, citations in RESEARCH-FINDINGS)

1. The machinery is done and trustworthy. Edge Lab (bit-exact acceptance), level engine (causality-proven), paper loop (streaming==batch), MC harness (block-only), prereg template with ex-ante cost kill rule, 199-test suite, public evidence site. The project's marginal cost of a new rigorous experiment is now ~hours.
2. The mentor's mechanizable claims failed comprehensively (0-for-5 on direction: entries, direction logic, stop-structure fit, level reactions, H4 ignition). What remains of Track M value: his event concepts seeded two real research objects (displacement, streaks), his session observation matched measured structure, and his risk discipline informed the executor design. D1 (October 31) should be assessed against this base rate: even verified mentor data is unlikely to yield a mechanized edge directly; its realistic value is calibration (original stop placements for the B4 question) and cost truth (his fills).
3. One effect family is asset-class structure (displacement continuation: 4/5 instruments, XRP cleanest) — but its BTC execution died on the Validation holdout (regime inversion post-2022). The event class is real physics; harvesting its direction is regime-fragile.
4. One candidate is alive but weakened (up-streak persistence: BTC execution BAR MET; instrument map patchy 2/5 decisive + XRP inversion; regime conditioning ambiguous/underpowered). Look-2 decision rests with Richard.
5. Cost structure is destiny. Vantage-class BTC ≈ 3.1bp is the only venue/instrument combination where any tested effect clears its own drag; ETH (22.6bp) killed a registration by arithmetic alone. Any instrument-expansion ambition runs through sourced cost models first.

3. Questions D4 must actually answer in October

• Q1 — Is the technique class viable in the current regime? Determined largely by what happened between now and then: the look-2 outcome (if spent), the mentor-data calibration (if delivered), and any new research-split candidates. If both spent looks have failed by October and no new candidate has cleared the bar, D4 should seriously rehearse D2's exit taxonomy (cheaper instrument per unit volatility / wider stops / cheaper execution) rather than default to "keep scanning."
• Q2 — Does the two-track reading still hold? (Research on public data proceeding while Phase 0 blocks capital and mentor-lineage claims.) Assess against D1: if gates 0.1/0.2 are still open in October, Track M is struck per D1's automatic downgrade and the project is de facto a pure systematic-research program — the brief's framing should be updated to say so.
• Q3 — Is the hypothesis spend buying knowledge or motion? Test: of the ~223+ spent, how many produced findings that changed subsequent decisions? (This quarter's honest answer: the cost law, the artifact catalog, the universality maps, and both candidates' verdicts all did; the H4/level/session nulls retired real design space. If October's marginal scans are not retiring space or changing decisions, slow down.)
• Q4 — Capital readiness, honestly. Even on the optimistic path (look 2 passes), the remaining gates are: Vault forward-test, Gate B paper (≥100 trades in ~2-3 months of live paper at run-21 event rates), measured-cost hardening (Vantage demo), MC-sized capital allocation, and the executor fork decision (#3). Zero capital deployment before those is a Q4-2026/Q1-2027 earliest reality — D4 should say this number out loud to keep expectations honest.

4. Standing risks to name at the review

• Selection pressure on the research split grows with every scan — the FDR ledger controls the formal rate, but the informal risk (designing "new" families that are cousins of survivors) needs active honesty. Mitigation in force: pre-declared pattern rules, refusal records, this document.
• Regime non-stationarity is the dominant unmodeled factor (two demonstrations: displacement across time, streaks across instruments). Gate-state v2's regime-health fields are the operational response; they need implementation before any paper run.
• Single-instrument, single-venue concentration: everything harvestable lives on BTC-at-Vantage-class-costs. That is a fragile foundation and should be stated as such whenever "the edge" is discussed.

5. Pre-committed for October

This draft is the baseline. The October session diffs: budgets vs §1, D1–D3 status vs §2–§3, and answers Q1–Q4 with the intervening quarter's registered evidence only. If this document's framing turns out wrong, the review says so and amends the program — it does not quietly rewrite this pre-read.