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 Path E routing ADR + framework artefacts catalogue. Path E ≠ project failure; documented disconfirmation with mechanism. The substance of the document below is unchanged from the internal version; only this framing block and personal/identifying content scrubs are added.

ADR-0019: Phase 5d Gate 1 FAIL — Path E routed, in-universe attempts terminated

Context

Phase 5d Gate 1 was a pre-committed, full-text verification step on the Fieberg / Liedtke / Metko / Zaremba (2023) Quantitative Finance paper that ADR-0017 designated as the Tier-1 evidence base for Criterion C (cross-sectional factor momentum). ADR-0017 §7.4 fixed the gate language verbatim:

“The Fieberg/Liedtke/Metko/Zaremba 2023 Quantitative Finance paper must be obtained in full text. Profit factor for the long-only winner portfolio (or its 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.”

The Phase 5d Gate 1 report (docs/reports/2026-05-12-phase-5d-gate-1-pf-verification.md, commit 95bb62c) executed this gate. Full text was obtained (CC-BY open-access PDF, channel #8 of the search trail). The findings, verbatim from the report:

  1. PF is not reported in the paper. Anywhere. Not in Table 4 (headline performance), not in Table 13 (long-only winners financed with risk-free rate or BTC), not in Table 12 (liquidity subsamples), not in Table 8 (768-design-choice robustness), not in Table 5 (subperiod). The paper’s risk-adjusted metric is Sharpe only. PF, gain-loss ratio, hit rate, win rate, success rate — none are computed.
  2. No transaction-cost analysis is reported. Zero matches for “transaction cost”, “trading cost”, “fees”, “basis points”, “bps”, “net of”, “round-trip”. All headline figures (Sharpe 1.28 CS winners; Sharpe 1.43 most-liquid-25% long-only winners; Sharpe 1.09 CS winners minus rf, Table 13 Panel A) are gross of friction.

Both findings are independently sufficient to FAIL the gate. The case is strictly weaker than the §7.4 verdict-table row “Full text obtained, PF reported only gross” (which already maps to FAIL): here, PF is not reported at all, and the net-of-fees clause cannot even be examined.

Decision

Path E is routed. Per ADR-0017 §7.4 Gate 1 verbatim, Criterion E fires; Path D is foreclosed. In-universe strategy attempts on the BTC/USDT + ETH/USDT 4h spot long-only universe are terminated.

PRD §6 stays at 4h primary indefinitely. The PRD §6 amendment clause requires a PASS to fire; the verdict is a final-form FAIL, so no amendment is applied.

The §14 bar (Sharpe > 1.0 AND PF > 1.4 OOS friction-adjusted) is not lowered, not contested, not revised. The bar transitions to a new status (see §”§14 bar status” below) — it is documented-not-met-on-this-universe, not eroded.

Two tags are applied at this ADR’s commit:

The two tags mark different decisions at the same commit and are kept distinct so future-readers can cite either transition specifically.

What bound, in plain English

The Fieberg et al. paper, on direct full-text reading, does not actually publish the evidence ADR-0017 required it to publish. The headline Sharpe figures (1.28 cross-sectional winners; 1.43 in the most-liquid-25% subsample) are not contested — they are exactly what the audit said they were. The problem is that Sharpe alone does not clear §14’s AND-clause. §14 requires PF > 1.4 as well as Sharpe > 1.0, both net of fees. The paper:

Gate 1’s premise — “extract PF for the long-only winner portfolio at retail-realistic net fees” — has no source data in the paper to extract from. There is no inferability path either: PF cannot be derived from (Mean, SD, t-stat, Sharpe) alone without distributional assumptions the paper does not make, and the gate language pre-committed in ADR-0017 §7.4 is explicit that PF must be extracted from the paper, not estimated. So the gate’s premise fails by construction. Criterion E fires.

The literature’s “Sharpe 1.28” claim is real; it just does not reach §14’s bar on paper, let alone after retail-realistic friction is subtracted.

Why Path D is foreclosed

ADR-0017 §”Path D escalation gates” is structured as a logical AND across four gates. Any single FAIL is terminal for Path D. Gate 1 has FAILed in the strongest possible form — not “PF < 1.4 net” but “PF undefined in source, gross-only data”. Gates 2–4 (universe match, net-of-fees Sharpe verification, methodological dissent resolution) are not executed because their value as gates was strictly conditional on Gate 1 passing. With Gate 1 failed, Gates 2–4 would only be relevant to a Path D scope that the AND-structure already forecloses.

To be explicit: a Path D scope document cannot be authored without satisfying all four gates. Today’s verdict makes the first gate unsatisfiable from the designated source.

Why we don’t re-evaluate

Cross-strategy comparison

Extending ADR-0018’s four-column table with a Phase 5d Gate 1 column:

Dimension MFD (Phase 2e FAIL) BMR (Phase 3 FAIL) FCMFD (Phase 4 FAIL) MacroDonchian8h (Phase 5c FAIL) Phase 5d Gate 1 (Fieberg verification) FAIL
Gate of failure §14 (post-hyperopt OOS) §14 (post-hyperopt OOS) SIA (pre-hyperopt) §14 (no hyperopt, diagnostic) Literature gate (pre-strategy-build)
Compute cost to verdict ~Hetzner-hour × 4 windows ~Hetzner-hour × 2 seeds ~3 seconds laptop ~5 minutes laptop (6 backtests) ~1 hour reading PDF + search trail
Best per-window OOS Sharpe 0.553 (W3 @ 4h) -0.083 n/a (gated out) 0.586 (W1 @ 8h) n/a — never reached strategy build
Best per-window OOS PF 2.509 (W3 @ 4h) < 1.0 all 15 n/a 2.065 (W3 @ 8h) n/a — PF not reported in source paper
Binding constraint Sharpe (PF mostly OK) Sharpe AND PF Lift gap σ AND vol-control Sharpe (PF clears in all 4) PF not reported AND no net-of-fees data
Inverted variant tested? No No Yes — also fails n/a (no signal to invert) n/a (no signal under test)

Five consecutive FAILs across five different binding constraints. Phase 2 and Phase 5c failed on the same shape (Sharpe-only). Phase 3 failed on both Sharpe and PF. Phase 4 failed on lift-gap σ (SIA framework). Phase 5d failed at the literature-evidence layer before a strategy was ever scoped. The convergent pattern: across in-universe strategy classes (breakout, mean-reversion, funding-conditioned, timeframe-shifted) and across the literature base ADR-0017 designated as the foundation for the next class to try (cross-sectional momentum), no path to §14 is supported by evidence on this universe.

§14 bar status

The §14 bar (Sharpe > 1.0 AND PF > 1.4 OOS friction-adjusted) enters documented-not-met-on-this-universe status as of this ADR.

This is not bar erosion. The bar:

The bar transitions from a contested-theoretical state (“§14 might be achievable on this universe with the right strategy class — let’s see”) to a documented-empirical state (“across four strategy attempts plus a literature audit plus full-text verification of the designated Tier-1 source, §14 has not been met on this universe”).

This ADR is the receipt. If any future operator — the original author, a collaborator, a fork — considers deploying capital on BTC/USDT + ETH/USDT 4h spot long-only, this ADR is the binding record that says: this surface has been tested under documented discipline and does not support §14-clearing strategies per current evidence. The receipt does not say “never possible”. It says “not demonstrated, on documented evidence, as of this date”.

Phase 5a audit §5.6 corrigendum (primary finding)

The Phase 5a literature audit (docs/reports/2026-05-12-pre-phase-5-literature-audit.md §5.6) attributed the phrases “incurs substantial trading costs” and “extracts alphas largely from short positions” to Fieberg et al. (2023). Direct full-text reading (Gate 1 report §4.4, commit 95bb62c) confirms:

This corrigendum is recorded as a primary finding of this ADR, not a footnote, because it documents a specific failure mode of AI-assisted literature research that is independently valuable to future-readers:

Failure mode: Paywall reliance + abstract-only summarisation = attribution errors that look authoritative.

The Phase 5a audit was executed under realistic constraints (no full-text access at the time; multiple Zaremba papers in the citation graph; abstract-level summarisation as the available evidence base). The conflation that resulted was an honest one, but it was load-bearing for ADR-0017’s framing of Path C / Path D, and it would have remained load-bearing for any Path D scope document built on top of it. The pre-committed Gate 1 step is what exposed it, before — not after — 4–8 weeks of strategy work were sunk into a foundation that misread the source.

The corrigendum strengthens the Gate 1 FAIL, not weakens it. If the audit’s verification-pass framing had been correct (i.e., if the paper itself had acknowledged “substantial” costs), Gate 1 would still FAIL because PF was never reported. The corrected reading is even more pristine: the paper does not comment on the cost question at all, leaving the gross-only evidence base unqualified.

Teaching artefact. Future literature audits under this framework should treat unverified-from-full-text claims as provisional and tag them in the audit text. Any pre-committed downstream gate that is load-bearing on a specific paper claim should include a full-text retrieval step before a scope document is authored on top of it. The Path D gate structure already had this property by accident-of-design (Gate 1 fired before Path D scope was written); future audits should make it explicit-by-design.

What we keep — the framework artefacts catalogue

These are the framework artefacts that survive the in-universe kill and constitute the genuine project output. They are listed with their reusability scope so future-readers can extract any subset independently.

Publishable

Reusable infrastructure

Workflow discipline (the meta-artefact)

What we don’t keep

What discipline held

Bigger picture

Five phases of attempted-or-prevented in-universe work. Four strategy classes empirically tested (MFD, BMR, FCMFD, MacroDonchian8h). One literature foundation empirically verified and found wanting (Fieberg et al. 2023). One corrigendum exposed before-the-fact that would have biased any Path D scope on top of it.

At each decision point, discipline held:

The project converged on a documented negative result with mechanism, which is more valuable than producing a deployable strategy that wouldn’t have worked. Two specific reasons:

  1. The negative result has a receipt (this ADR + the four predecessor kill ADRs + the Phase 5a audit + Phase 5d Gate 1 report). The receipt names the universe, the strategy classes attempted, the bar, the literature evaluated, the corrigendum, and the routing. A future operator reading this can decide whether to revisit the surface with new evidence rather than re-running the same experiments blind.
  2. The framework artefacts are genuinely rare for retail quant work. SIA-style pre-hyperopt signal-independence testing is published-ish in academic settings but operationally rare in retail tooling. Cross-machine deterministic Freqtrade pipelines with sha256-pinned data manifests are not free out of the box. Pre-committed audit methodology with mechanical verdict application is essentially absent from the retail-quant literature. The infrastructure has value independent of any specific strategy succeeding on top of it.

What Phase 5e is

Phase 5e is the packaging and publication phase. It is a fundamentally different kind of work than Phases 2–5d, which were strategy attempts. Phase 5e is about shipping the framework artefacts catalogued above as standalone reusable infrastructure — most obviously the SIA harness (publishable per §”What we keep”), and secondarily the deterministic-backtest + look-ahead-safe-alignment + data-manifest tooling as a coherent quant research substrate.

The Phase 5e scope document (boundaries, deliverables, success criteria for the packaging work) is the next operator conversation, not part of this ADR. The phase-5e-path-e-routed tag at this commit marks the entry; the scope is authored separately.

References