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: Path C’s verdict: 0/4 windows clear §14 at 8h. Timeframe shift was not directionally helpful per-window. The substance of the document below is unchanged from the internal version; only this framing block and personal/identifying content scrubs are added.

ADR-0018: Phase 5c PATH-C-KILL — timeframe was not the constraint

Context

Phase 5c ran a pure diagnostic test of the timeframe-shift hypothesis raised by ADR-0017: take Phase 2’s locked MacroDonchian default parameters (no hyperopt, no signal-class change) and re-run them at 8h primary timeframe across the same four walk-forward calendar windows that Phase 2 used. The question was narrowly framed: was 4h the binding constraint?

The four-window §14 verdict is FAIL across the board:

Window Trades Sharpe PF Max DD §14?
W1 (2023H1) 25 0.586 1.841 4.10% FAIL (Sharpe)
W2 (2023H2) 22 0.203 1.446 10.36% FAIL (Sharpe)
W3 (2024H1) 25 0.531 2.065 8.85% FAIL (Sharpe)
W4 (2024H2) 19 0.341 1.606 3.51% FAIL (Sharpe)
W5 (2025-01 → 2026-05, supp) 37 0.208 1.546 8.16% FAIL
Full-span (2023-2024, supp) 88 0.389 1.916 8.85% FAIL

Per spec §7 row 2 + ADR-0017 §”Path D escalation gates”, a 0/4 §14 outcome at 8h routes the decision to the four pre-committed gates from ADR-0017 §7.4. This ADR records the kill, the diagnostic interpretation, and the routing forward.

Decision

MacroDonchian8h is killed. PRD §6 stays at 4h primary (spec §8 explicitly forbids amendment except on PASS; the verdict is FAIL, so no amendment fires). Decision routes to ADR-0017 §”Path D escalation gates” — the four §7.4 gates must all be satisfied before any Path D scope is written. If any gate fails, routing escalates to Path E (stop in-universe; ship the SIA framework as the carry-over deliverable).

The tag phase-5c-path-c-killed is applied at this ADR’s commit.

What bound, in plain English

Sharpe was below 1.0 in all four primary walk-forward windows AND in both supplementary windows (the 2025-onward holdout and the full-span aggregate). PF passed comfortably in every window (1.45–2.07 range). The binding constraint is exactly the same as Phase 2 at 4h: returns per unit volatility are too low for this universe at this strategy class, regardless of which primary candle cadence the breakout is drawn against.

Per-window apples-to-apples 4h vs 8h:

Window 4h Sharpe 8h Sharpe Δ Sharpe 4h PF 8h PF Δ PF
W1 0.400 0.586 +47% 1.614 1.841 +14%
W2 0.359 0.203 −44% 2.208 1.446 −35%
W3 0.553 0.531 −4% 2.509 2.065 −18%
W4 0.388 0.341 −12% 1.699 1.606 −5%
Mean 0.425 0.415 −2% 2.008 1.740 −13%

The shift from 4h to 8h does not move Sharpe in a directionally helpful way — the per-window means are essentially flat (4h 0.425, 8h 0.415). PF regresses on three windows out of four. The 8h experiment was operationally clean (Hetzner-deterministic, data manifest verified, no integrity issues — see Task 8 closeout), and the answer is clear: timeframe alone was not the binding constraint.

Why the timeframe-shift hypothesis is decisively refuted

Phase 2’s best per-window walk-forward OOS Sharpe was 0.553 (W3 @ 4h). Phase 5c’s best per-window OOS Sharpe is 0.586 (W1 @ 8h). The two are statistically indistinguishable. No window crossed 1.0; no window came close. On the same calendar windows, 8h MFD does not produce regime-robust §14-clearing performance.

This is the cheapest experiment design that could have falsified the hypothesis: same strategy, same defaults, same universe, same eval pipeline, different primary candle cadence. The hypothesis “timeframe was the binding constraint across Phases 2-4” is now empirically refuted at this strategy + universe combination. The constraint sits elsewhere — universe composition, signal class, or both — and ADR-0017’s Criterion D fired correctly when it routed us to Path C first as the cheap test.

Why not re-tune / re-hyperopt at 8h

Phase 5c spec §3.4 and §3.5 forbid both hyperopt and signal-class introduction within Phase 5c. The diagnostic discipline was binding by design — Path C tests whether the locked Phase 2 strategy clears at 8h, not whether some 8h variant clears after re-tuning. Re-tuning at 8h would be a separate phase entirely (Phase 5d Path D or Phase 6), not a Phase 5c retry. PRD §15.2 forbids bar erosion; the §14 bar (Sharpe > 1.0 AND PF > 1.4) is not revised under any scenario except ADR-0017’s Criterion E pathway with documented evidence, and Criterion E does not fire today — it becomes the routing only if one of the four Path D gates fails downstream.

Lowering Sharpe from 1.0 to ~0.55 to pass the best window would be exactly the bar erosion PRD §15.2 forbids.

Cross-strategy comparison

Extending ADR-0015’s three-phase table with the Phase 5c column:

Dimension MFD (Phase 2e FAIL) BMR (Phase 3 FAIL) FCMFD (Phase 4 FAIL) MacroDonchian8h (Phase 5c FAIL)
Gate of failure §14 (post-hyperopt OOS) §14 (post-hyperopt OOS) SIA (pre-hyperopt) §14 (no hyperopt, diagnostic)
Compute cost to verdict ~Hetzner-hour × 4 windows ~Hetzner-hour × 2 seeds ~3 seconds laptop ~5 minutes laptop (6 backtests)
Best per-window OOS Sharpe 0.553 (W3 @ 4h) -0.083 n/a (gated out) 0.586 (W1 @ 8h)
Best per-window OOS PF 2.509 (W3 @ 4h) < 1.0 all 15 n/a 2.065 (W3 @ 8h)
Binding constraint Sharpe (PF mostly OK) Sharpe AND PF Lift gap σ AND vol-control Sharpe (PF clears in all 4)
Inverted variant tested? No No Yes — also fails n/a (no signal to invert)

Four consecutive strategies under four different operational profiles, three different binding constraints (Phase 2’s and Phase 5c’s are the same shape). The pattern across all four sharpens the universe-revision hypothesis ADR-0015 identified and ADR-0017 formalised: the BTC/USDT + ETH/USDT spot universe at any single-asset primary timeframe may not support strategies that jointly clear Sharpe > 1.0 AND PF > 1.4 OOS friction-adjusted. ADR-0017’s Criterion D fired correctly. Path C’s failure confirms the literature audit’s interpretation: the constraint is structural, not strategy-specific.

The four Path D escalation gates (from ADR-0017 §7.4, verbatim)

These are the binding entry conditions for any future Path D scope document. Re-listed here so this ADR is self-contained:

If any of Gates 1-4 fails, the decision re-routes to Path E.

What we keep (carry-over to Phase 5d Path-D-gate-check or Phase 5e Path-E)

What we don’t keep

What discipline held

References