# Prospective warm-up collection protocol

Registration is effective only when the matching manifest and pre-start activation are independently timestamped and published.

**144.51% gate: UNMET.** This is a data-collection phase only. There are no paper trades, positions, equity, returns, CAGR or promotion decision. The separately selected cohort is the ten highest historical full-period candidates among the latest 150 v15/v16 combinations plus the TSI(25,13) control, frozen by the `TOP10_SELECTION.json` SHA-256 recorded in the eventual commitment. The user explicitly permitted collecting newly arriving data in calendar overlap with the ongoing v10 study; the collector must not read or alter its files or use any prior protected observations.

Before any new market observation, publish the final protocol, the immutable cohort hash and an independently timestamped manifest of the collector and environment, with an absolute future UTC start. Apply only newly arriving data after that start. Begin with a point-in-time snapshot of Binance USDT perpetual metadata during the 23:30 and 23:45 UTC operating slots; no historical metadata is backfilled. For any daily bar, use the latest valid metadata snapshot received between 29 minutes 15 seconds and 45 seconds before the UTC day began. A listing appearing after that snapshot is omitted for that day; if no timely snapshot exists, record a gap and do not fill it later. This sampled universe is an approximation, not a complete instantaneous listing record.

Retrieve only the immediately preceding full UTC day, no sooner than 45 seconds after its close and no later than 5 hours 59 minutes 15 seconds after its close. Request and receipt time must both be within the window, checked against the exchange clock and a locally monotonic wall clock. A symbol can be read only if it was in the qualifying earlier snapshot. Validate every retained candle and funding field against a strict schema and timestamp; retain exact validated response bytes and their hash. Do not retain unverified or late market bodies; record error type, response hash and availability gaps. Funding is labeled **UNVERIFIED**, including rate type and mark; it is never assumed complete or zero and cannot be used as a paper return. Retain exact metadata response bytes only when the entire documented, bounded metadata schema validates; unknown schema fields cause a gap and a hash-only diagnostic. Independent review must confirm the schema and its source-time limitations before activation. No missing day is silently backfilled. A 60-consecutive-valid-day eligibility rule applies once trading logic is separately approved.

The collector uses a physically separate database and OS lock. Every event is hash chained; this alone is not tamper-proof. Publish periodic signed external timestamp proofs of the ledger head, including event count, before calling a series externally anchored. Persist any untrusted-clock outage as an explicitly untrusted operational health record. If the user's computer is offline, asleep or logged out and scheduled collection misses the valid window, show the gap; do not invent data. Before activation independently review the actual collector, timestamps, scheduler and the public witness. The initial activation must create and publicly witness the actual first ledger event before the start. Subsequent runs reconcile every public signed checkpoint against the complete local ledger; a snapshot may be used only if an independently verified checkpoint containing it was publicly received before the relevant day.

A separate fully specified, independently reviewed paper-execution and accounting protocol must be timestamped before the later trading evaluation. The candidate start rule is the first Monday at 01:00 UTC at least 168 elapsed hours after the complete execution protocol and implementation pass independent review and receive their own timestamped commitment, and only if, at a fixed 00:55 UTC readiness cutoff, each of the ten books and the TSI control has at least eight currently tradable names with 60 consecutive valid fresh bars and its declared signals and funding inputs available. Log each Monday’s readiness count and missing-data reasons. If the condition fails, apply the same rule on the next Monday; do not defer an otherwise qualifying start. Availability and outages can correlate with market conditions, so this trigger is not statistically independent of the market. Complete review and commitment timing must not use warm-up prices, signals or returns. Never use warm-up returns, price patterns or interim outcomes to select the cohort, adjust the rules or choose the start. The outcome period is a new 365-day calendar window beginning after that second commitment, not the warm-up period. Preserve all ten hypotheses and the TSI control; no interim outcome-driven selection. No forward performance or promotion claim can be made from warm-up data.