Touch-and-Turn tier-D RDDT follow-up

Status at registration: NOT RUN. This document was written before the tier-D RDDT evaluator was executed. The locked admission rule is at least 360 of 390 RTH minutes on every regular session (no more than 30 missing minutes). The proportionate early-close ceiling is 16 missing minutes from a 210-minute session. These numbers are fixed before any tier-D outcome statistic is read.

Why this tier exists

Tier C admitted at most 19 missing minutes from a 390-minute regular session. RDDT was then observed to have 370 of 390 minutes on 2024-07-31: 20 missing, one minute beyond that locked budget. Tier D is an ad hoc tier created after seeing that one-minute miss. It is not an independent replication, a new family, a repair to tier C, or evidence that the admission threshold was chosen without looking at RDDT's tape.

The wider tier has a distinct sealed tape identity. Tier C remains unchanged at its original ceiling. Tier D permits holes inside the opening 15 minutes; any such session is disposed as incomplete_tape, exactly as in tier C, rather than having an opening range reconstructed or padded.

Locked packet

• Symbol and window: RDDT only, the same 512 NYSE sessions from 2024-07-24 through 2026-08-07 used by the roster-A/B/C date vector.
• Tape admission: every regular session must contain at least 360/390 observed RTH minutes. An early close may miss no more than 16 of 210 minutes. Nothing is imputed. A session with a hole in its opening 15 is retained only as incomplete_tape and enters no rate denominator.
• Evaluator: the existing touch-and-turn-evaluator/v1, unchanged, including every execution correction in SPEC-AMENDMENTS.md.
• Signal and exits: the unchanged primary genome: 15-minute opening range; causal ATR14_RTH through the prior RTH session; range at least 25% of that ATR; fade the opening candle; 90-minute resting limit; 1-tick penetration; 0.382-range target; 0.191-range stop; five-minute-before-close time exit.
• Sizing: fixed notional 100,000,000,000 minor units against 1,000,000,000,000 initial capital, whole shares, with entry cost reserved outside notional.
• Costs: the same charged 6 bp adverse entry plus 6 bp adverse exit, zero fee. The daily model's ceil(5 bp) + ceil(1 bp) per fill remains the uncharged comparison line.
• Statistics: the existing report implementation and deterministic bootstrap settings used by tiers A-C. Report the one-symbol tier-D pool's excess return against the same zero benchmark, win rate with its date-clustered 95% interval, gross and charged-net basis points per trade with 95% intervals, profit factor, and tape contamination beside the outcome. No grid, parameter sweep, symbol selection, or direction control is added.
• Sealing: run once through the existing Touch-and-Turn build/spawn runner, with a fresh-process determinism check. Persist the content-addressed run JSON, a manifest, and SHA-256 digests under the same Touch-and-Turn artifact-directory convention as the sealed primary run.

Hypotheses and decision rule

The locked descriptive hypotheses are the same claims already refuted by the family: the video's win rate exceeds 70%; its loss share is below 30%; and the strategy's mean charged excess return over the zero benchmark is positive after the same 12 bp round trip. Gross expectancy is reported to show whether costs explain the sign; it is not a separate promotion test.

Tier D has no path to a new alpha claim or promotion. The Touch-and-Turn family is already refuted, and this ad hoc tier cannot change that evidence state regardless of sign, confidence interval, or profit factor. The decision rule is therefore:

1. Report every RDDT result and the date-clustered uncertainty without selection.
2. Report pooled excess/net bps, win rate, gross bps, profit factor, and both the wide ordered-session and realised-trade contamination rates.
3. Label the tier completeness-only and ad hoc. Do not run follow-up variants, tune the 360-minute floor, add the result to tiers A-C, or use it to claim alpha. A positive result would be an observation requiring a separately pre-registered family on unseen data; a null or negative result merely extends the already-refuted family record.

Results

Execution-custody disclosure: the first execution was voided because it altered an executor-bundled module; the pre-registered admission rule was unchanged for this replacement execution.

Completeness result: same refutation, no new alpha claim. The one permitted RDDT run produced 299 trades on 299 dates, a 31.10% win rate (date-clustered 95%: 26.09%–36.45%), −1.78 gross bps per trade (bootstrap 95%: −11.89 to +8.25), −13.78 charged-net/excess bps per trade (date-clustered 95%: −24.33 to −2.88), and a 0.726 profit factor. There were 93 wins and 206 net losers, so the loss share was 68.90%. RDDT does not rescue the already-refuted family.

• Tape/accounting: 512 sessions, 10 ATR warm-up, 53 incomplete_tape sessions with no intact opening range, 449 decidable sessions, 409 submitted orders, and 299 fills. The corpus admitted 398 missing minutes across 175 sessions. No missing bar was padded or imputed.
• Contamination, wide: 108 of 409 ordered sessions (26.41%) had a hole somewhere in the full order-or-potential-holding window. Three ordered sessions had an order-window hole; 108 had a potential holding-window hole.
• Contamination, realised: 3 of 299 trades (1.00%) had an order-window hole and 0 of 299 (0.00%) had a hole inside the realised entry-to-exit holding window. The negative result is therefore not carried by a realised holding-window gap.
• Classification: ad_hoc_completeness_only_refuted_family. Tier D is counted as the family's fourth attempt, is ineligible to become a survivor, has no Holm family or evidence state of its own, and cannot change the Touch-and-Turn verdict.

Seal and integrity receipts

• Run id / canonical-preimage SHA-256: 33fb4629e37f467ca7414a9c2a926dd4a8f68203e59f400149b5a155fed05fb5. The whole pipeline matched this id in two fresh processes.
• Persisted run JSON SHA-256: 94da216903cea765a405793627f8e3d49ebadc439085da9968ae0dd5a8e2f9db.
• Manifest SHA-256: ee38a7c967813fee24d4f53768223babc50a6be9b54ed0e1aa59c15c27ae0509.
• SHA-256 sidecar SHA-256: 080cd341282775519a974692ccd730a0c83e476d4da454c66862e47fdeabeec3.
• RDDT corpus revision: 3aac8592f9219b7f31c2e5bee5a2ab70cead9080977e4dcccbcde1ba88bd43e7; corpus-file SHA-256 0a9d56822275c3814325afb890eff75cb479f93771b88a811622602a6620d4ae.
• Import-attestation SHA-256: 9829aed5eae34bd5b4537c125e4ecbffeed3c50a7f28d88be6faa52104ad8eb4.
• Pre-run registration snapshot SHA-256: 4638ab74e3e88f89fc4bb1978fc25ca9eb2f479bcb511ade569557ed3ca53adc.
• Artifact directory: [local source path] It contains the run JSON, manifest, .sha256 sidecar, and immutable pre-run registration snapshot; the sealed corpus and import attestation are under the sibling [local source path] convention.