docs(experiments): cusum_event measured — feasible, and without a gross edge (#341)

[eaitbrahim]

The measurement #341 asked for: pre-declared, then run. 168 trials, 24 assets, ~5 years of hourly candles, fees 0 / 0.006 / 0.012.

1. Feasibility: yes — and it contradicts the worry filed with the issue

#341 was filed with the concern that "gating cuts n on rules already below the admission floor." It does not.

assets with n ≥ 100	21 of 24
median n	553
range	32 – 884

A 5% cumulative move in an hourly crypto series is a common event, not a rare one. The three below the floor are the two PAXG pairs and TON — gold and the thinnest name in the universe, which is the expected shape. That objection is answered and shouldn't be repeated.

2. Profitability: no, and not marginally

fee	PF median	PF max	above 1.0
0%	0.925	1.238	8 / 24
0.6% (maker, unreachable)	0.553	0.750	0 / 24
1.2% (taker, actual)	0.343	0.479	0 / 24

The intersection of n ≥ 100 and PF > 1.0 is empty across all 168 trials.

3. The diagnostic — this is the rsi_meanrev disease, not the turtle_breakout one

The number that matters is the zero-fee column: only 8 of 24 clear 1.0, median 0.925.

turtle_breakout at zero fee was profitable on 4 of 4 with a max of 2.713 — a real gross edge that cost destroyed, which is why cheaper execution was a coherent thing to want. This rule loses at a median before a single fee is charged. Same symptom as turtle at 1.2%, different disease — and the one execution cannot fix, because zero fee bounds from above everything an execution improvement could buy, and zero fee is already a loss.

4. The knob works exactly as designed, and converges to nothing

mult	threshold	median n	PF median	above 1.0
1	2.5%	1252	0.226	0
2	5.0%	553	0.343	0
3	7.5%	330	0.361	0
4	10.0%	224	0.379	0

Monotonic in the right direction — trading less does lose less per toll, so the mechanism is real. It is also converging to a ceiling well below 1.0 while n falls toward the floor. A fourfold threshold buys 0.15 of profit factor and costs 82% of the sample.

5. The source's own setting is the worst cell

mult=1 is the paper's 2.0–2.5%, which on this venue is exactly one round trip: lowest median PF (0.226) with the highest median n (1252) — the most trading at the least edge. That's #341's friction-scaling argument measured rather than asserted.

Honesty

Arm A is one pre-declared configuration, so it carries no selection bias. Arm B's per-asset best is a maximum of four draws and is stated as such — best cell CRV-USD at mult=4, PF 0.593, still 0.4 short, so the bias changes nothing either way. Screening result only: no walk-forward, no PBO (series_missing). Slippage held at 0.0005 throughout, so the zero column is zero fee, not zero cost.

The ledger row was written twice

Worth recording. Every non-int summary value must be a string: _decode_summary turns anything that isn't an int into a Decimal on read, so a JSON float round-trips to a different value and the row stops hashing to its own row_hash.

verify_chain caught it immediately. The corrupt trailing row was removed — restoring the chain to exactly its prior state, verified clean at 90 rows — and re-appended with the encoding every existing row already uses. The chain verifies at 91 rows.

Changed nothing else

A document, a driver script, a ledger row. No rule row added, nothing promoted, no config or allowlist touched. cusum_event stays registered and untraded.

The measured intersection goes from 0 of 90 to 0 of 114. That is the result, and publishing it is the point.