For external readers: This document is one of 10 curated case-study artefacts from Project Aṣe, a quantitative crypto-trading research project that produced a documented negative result with mechanism across four strategy attempts plus a literature audit, all under pre-committed criteria (Sharpe > 1.0 AND profit-factor > 1.4 OOS, called “§14” internally) that held without bar erosion across 19 ADRs. The full project repo is private (scope B of the Phase 5e shipping spec). Start with ./00-readme.md for the reading guide. Read order context: The decision ADR that locked Path C first + pre-committed the four Path D escalation gates. The substance of the document below is unchanged from the internal version; only this framing block and personal/identifying content scrubs are added.

ADR-0017: Phase 5b Decision — Path C first, Path D gated, Path E in reserve

Context

Phase 5a executed the pre-Phase-5 literature audit specified in docs/phases/phase-5-scope.md §4. The audit is committed in two parts: the v1 report at commit 8fdb5e4 and the open-source verification update at commit 389b756. Together they constitute the evidence base on which Phase 5 scope §5.1’s pre-committed criteria fire.

The binding question from scope §4.1 was:

“What Sharpe ratios and profit factors do published / open-source crypto trading strategies actually achieve on retail-accessible surfaces, particularly BTC/ETH spot long-only at various timeframes?”

After two waves of search across 12 queries, ~25 candidate sources reviewed and 14 cited (8 Tier-1 peer-reviewed, 2 Tier-2 working papers, 3 Tier-3 institutional, 1 absence-as-finding), the audit’s central findings are:

The audit’s §9 summary one-liner: “Criterion D fires. The §5.1 decision routes to Path C first. Path D escalation requires satisfying the four post-verification gates in §7.4 — failing which the decision re-routes to Path E.”

Decision

Phase 5c begins on Path C. Concretely: PRD §6 will be amended to move the primary timeframe from 4h to 8h, and MFD will be re-run on 8h BTC/USDT + ETH/USDT spot data as the cheap test of the timeframe-shift hypothesis. Estimated duration per scope §6 Path C: 1–2 weeks. The Phase 5c scope document is the next deliverable after this ADR.

Path D is authorised conditionally, not unconditionally. If Path C fails its own SIA or §14, Path D may only proceed after all four gates in audit §7.4 are satisfied (re-listed in §”Path D escalation gates” below). If any gate cannot be satisfied, the decision re-routes to Path E (stop in-universe, ship the SIA framework).

Path B is rejected. The literature does not support “a fifth signal class within the same 4h BTC/ETH long-only universe” as the highest-leverage move. Three signal classes have already failed §14 in distinct ways under different binding constraints.

Path E is in reserve, not triggered today. Cross-sectional positive Sharpe evidence (caveated and specification-dependent) is present in the literature, so strict Criterion E does not fire at this audit. Bar revision is therefore NOT a legitimate option at this point per scope §7.5.

Mechanical application of §5.1 criteria

Pre-committed criteria from docs/phases/phase-5-scope.md §5.1, applied to the audit findings:

Criterion Requirement Audit evidence Verdict
A ≥ 2 Tier 1/2 studies, Sharpe > 1.0 AND PF > 1.4 OOS friction-adj, BTC/ETH 4h spot long-only 0 qualifying studies. Matrix row A1 empty. Hudson & Urquhart finds no BTC OOS predictability. Does not fire
B Multiple studies showing Sharpe improvement on 8h+ timeframes for majors long-only Directional improvement only (Quantpedia: 0.33 → 0.80). No Tier-1/2 study clears §14 on majors long-only at any non-4h. Does not fire
C Long-only cross-sectional perp evidence achieving §14, mechanism not long-short asymmetry Sharpe 0.45–1.42 across Tier-1 specifications. PF unverified everywhere. Sources self-flag “substantial trading costs.” Does not fire
D Mixed/inconclusive: cells with partial evidence, no clean winner A empty, B partial, C partial-positive with material caveats (range 0.45–1.42, PF unverified, methodological dissent). FIRES
E No published evidence §14 achievable anywhere on retail-accessible long-only crypto spot Cross-sectional has positive Sharpe evidence even if caveated. Strict E does not fire today. Does not fire

Criterion D’s decision branch (verbatim from scope §5.1): “Path C first (smaller change, cheaper test). If C fails its own SIA or §14, escalate to Path D.” The audit’s verification update tightens that escalation: Path D is gate-conditional, not automatic.

Why this routing is honest given the evidence

Path D escalation gates (pre-committed, binding)

Lifted verbatim from audit §7.4. These are pre-committed in this ADR as binding constraints on any future Path D scope document. Path D is not authorised until all four gates are satisfied. If any gate fails, the decision re-routes to Path E.

Gate 1 — Full-text PF verification. The Fieberg/Liedtke/Metko/Zaremba 2023 Quantitative Finance paper must be obtained in full text. Profit factor for the long-only winner portfolio (or closest reported analogue) must be extracted. If PF < 1.4 net of fees, Criterion C strictly fails on the AND-clause of §14, and the decision re-routes to Criterion E (Path E), not Path D.

Gate 2 — Universe match. Path D scope as currently drafted (top-N liquid perps, BTC/ETH/SOL/BNB/XRP minimum) does not match the literature’s evidentiary universe (3,900+ coins). Either (a) Path D scope is expanded to top-20 or top-50 perp universe with explicit liquidity gates, OR (b) full-text retrieval of Fieberg et al / Liu-Tsyvinski-Wu confirms the cross-sectional momentum effect persists when restricted to top-10 by volume. Without one of these, Path D’s empirical foundation is inadequate.

Gate 3 — Net-of-fees Sharpe verification. The 2025 Risk-Managed Momentum paper’s “robust to transaction costs” claim must be quantified. What friction magnitude was tested (1 bp? 10 bp? 50 bp?), and what is actual net Sharpe at realistic crypto retail fees (typically 10–20 bp round-trip on Binance spot)? If the paper tested friction at < 10 bp, the result does not generalise to retail Binance and Path D’s foundation weakens further.

Gate 4 — Methodological dissent resolution. Grobys & Shahzad’s Sharpe-undefinability finding must be addressed. Either Path D plan acknowledges the strategy is built on a metric the literature itself contests, OR substitutes a metric not requiring finite second moments (Calmar, Sterling, max-drawdown-relative). The first is acceptable if explicit; the second is preferable.

What invalidates this decision

This decision is invalidated, in whole or in part, by any of the following:

  1. Path C MFD-on-8h outcome: This is the obvious downstream branch.
    • Path C passes §14: Decision validated — universe was the constraint, project has a §14-clearing strategy. ADR-0017 closes successfully.
    • Path C fails §14 (most likely): Decision routes to the Path D gate-check above. If all four gates pass → Path D scope authorised. If any gate fails → Path E.
    • Path C fails SIA pre-§14: Same routing as §14 fail, but with stronger Path E pull (constraint is structural, not strategy-specific).

  2. New Tier-1 evidence surfaces that PF-verifies the literature. If a peer-reviewed paper appears that reports full-text profit factor > 1.4 net of realistic retail fees for long-only crypto momentum (especially on a top-N-by-volume universe), Gate 1 is satisfied pre-emptively. This could route Phase 5c around Path C directly into gated Path D — but only with a new ADR amending this one.

  3. Path C surprise pass that invalidates the universe-constraint read. If Path C passes §14 but inspection reveals the pass is driven by an artefact (e.g., a single regime window, look-ahead in 8h data alignment, lookahead in macro-1d filter at 8h cadence), the validation does not transfer to a sound §14 clearance. Re-open the audit’s interpretation of Phases 2/3/4 failures and the cross-phase universe-constraint hypothesis.

  4. Friction reality at Binance retail diverges materially from Path C’s backtest assumptions. If during Path C execution we discover that realistic Binance spot retail fees (incl. partial-fill slippage, withdrawal-cost amortisation, taker-fee tier reality) are materially higher than the 10–20 bp the audit assumed, the §14 bar’s net interpretation shifts. A Path C net-Sharpe pass at 20 bp may not survive at 35 bp, in which case the routing changes.

  5. Bar revision (§14 itself) is NOT a valid invalidation route from this ADR. Per scope §7.5 and §8.2, bar revision is only legitimate under Criterion E with documented evidence. Today Criterion E does not fire. If Path C fails AND any Path D gate fails, Criterion E becomes the routing — and only at that point does bar revision become a documented option for separate ADR-led discussion.

Triggers 2 and 3 require a new ADR amending or superseding ADR-0017. Trigger 1 fires per the routing already specified above. Trigger 4 is handled inside the Path C scope document (the audit constraint becomes the friction-cost spec).

What we keep, what we don’t

Carry-over from Phase 4 to Phase 5c (Path C)

Discarded for Phase 5c

Preserved as historical record (do not delete, do not cite as live evidence)

Cross-phase context

Three consecutive in-universe attempts failed §14 under three different binding constraints (ADR-0011, ADR-0013, ADR-0015):

Phase Strategy Binding gate Compute cost What it ruled out
2 MFD §14 OOS Sharpe (best 0.31) Hetzner-hours × 4 windows “Price-only trend-following on 4h BTC/ETH”
3 BMR §14 OOS PF (all 15 < 1.0) Hetzner-hours × 2 seeds “Mean-reversion on 4h BTC/ETH”
4 FCMFD SIA pre-hyperopt (lift gap σ) ~3 seconds laptop “Funding-rate as an orthogonal entry gate to MFD”

The Phase 5a literature audit then asked the meta-question Phases 2-4 could not answer in isolation: is the §14 bar empirically achievable on this surface at all? The audit’s evidentiary answer:

The mechanical-criterion application above resolves this distribution-of-evidence as Criterion D, not Criterion C (the literature is genuinely caveated, not unambiguous) and not Criterion E (some positive evidence does exist).

What discipline held

References