# Historical initialization acquisition protocol

**144.51% gate: UNMET.** This protocol becomes effective only after independent review and publication of its verified timestamped commitment. It describes a new initialization-only acquisition, separate from both the existing warm-up and the earlier halted capture; their ledgers and signed code remain untouched. User authorized public OHLCV, quote volume and funding observations from the overlapping dates only for initialization, while leaving all v10 files and results untouched. No historical returns, portfolio decisions or rankings will be computed or displayed.

This attempt is informed by three halted, separately signed capture attempts and a separately registered schema probe. The earlier rejected bodies were withheld and neither the recorded error nor the later probe establishes their exact causes. The probe reported an extra root key whose hash matches `futuresType`, with string type, but did not disclose its value. This version accepts that optional field only if absent or exactly `U_MARGINED`; the value is a conservative fixture choice, not an observed fact. Any other value fails closed.

An intermediate v2 commitment was timestamped but never published or used: its written request schedule incorrectly implied that all funding pages would be requested even after an invalid earlier response. It made no market requests. That unused proof is preserved separately; the later v3 wording fixed the schedule before its source access.

Before making any market request, independently timestamp and publicly publish a manifest committing to this protocol, code/dependency hashes, Python and timestamp-verifier hashes, the fixed eleven-book selection, and the fixed 60-day UTC interval [28 August, 27 October 2026). Require isolated Python execution and bind the complete Python standard-library/native-runtime tree, interpreter binary, timestamp-verifier binary and adjacent native libraries. Verify the timestamp signature and public bytes before collection. This is a fixed input boundary, not a forward start. A later complete execution protocol and independently reviewed paper ledger require their own prior registration.

Prepare the registration and launch the capture using isolated Python, bytecode writing disabled, and separate explicitly selected freshly empty bytecode-cache prefixes. The preparation tool is itself included in the committed file hashes. The launcher verifies its settings and compiles only committed project-source bytes in memory. This reduces stale or substituted local bytecode risk; a compromised interpreter, operating system, administrator or pre-launch runtime cannot be independently ruled out by the program verifying itself.

The signature and the start of this capture must both occur before 28 October 2026 00:00 UTC; an incomplete capture that crosses this deadline may finish against its fixed committed window, but no new capture may start after it. Any HTTP error aborts the entire capture without retries; rate limiting or other service errors must not produce a burst of requests or be silently converted into missing symbols. A subsequent attempt requires a separate prior registration and cannot alter this ledger.

Capture the current Binance USDT perpetual metadata once after registration. Strictly parse bounded JSON, reject duplicate keys, nonfinite or out-of-range numeric tokens, and require a bounded symbol array. First reject the entire snapshot on any duplicate syntactically valid symbol, even if either row would subsequently be excluded. Invalid or absent identifiers prevent proof of source-wide uniqueness and are counted. For each row, exclude it under the first applicable reason in this fixed order: `not_object`, `missing_symbol`, `invalid_symbol`, `missing_required`, `invalid_required`, `unknown_enum`, `invalid_optional`, `identity_conflict`. Required fields are `symbol`, `status`, `contractType` and `quoteAsset`; optional `pair` and `baseAsset` must be valid and consistent when present. Known status/contract enums, identifier syntax, maximum counts and exact precedence are in the committed validator. Record total, accepted and excluded counts, counts per reason and excluded zero-based row ordinals, never the excluded raw values; require an accepted BTCUSDT perpetual or halt. Ignore unneeded metadata fields only for selecting the acquisition list. Store the canonical sanitized list, source and sanitized hashes, optional-field absence counts, request and receipt timestamps and HTTP status; do not retain the original metadata body or unknown values. The snapshot is **INCOMPLETE_UNVERIFIED_FOR_TRADING**: exclusion and current-survivor selection can be market-correlated, and a hash of a discarded original cannot let a reviewer reconstruct the source list or prove authenticity. Current listings are not a historical point-in-time universe. No automatic hand-off to a trading adapter is permitted; the existing draft adapter does not enforce this quality label. Before any paper decision, a separately reviewed trading wrapper must reject this snapshot for trading, compare initialization coverage of every potentially eligible name against a separately certified current trading universe and fail or wait under a separately frozen rule for omissions unless ineligibility is independently established; never select a favorable subset.

Use every currently trading USDT perpetual in that sanitized snapshot as the input-capture list. While the capture remains active, process symbols in lexicographic order: make one bounded 60-day 1d kline request; only if that validates, make up to six sequential 10-day funding requests at limit 1000. Each later funding page is requested only if the preceding page validated. A malformed or incomplete response ends that symbol's attempt, leaving it unavailable; an HTTP error aborts the entire capture. Each response must meet strict request, schema, event-time, 45-second cutoff, trusted exchange/local clock and receipt ordering checks. Log error and any available response digest for incomplete, malformed, saturated or failed data; explicitly mark the digest unavailable if no response bytes were retained. Do not silently backfill, substitute zero or retry selectively. Space funding requests at least 1.1 seconds apart and respect rate limits. Store exact validated raw price and funding responses in a separate append-only hash-chained ledger with source URL, request and receipt time, HTTP status and SHA-256. Retain only hashes of invalid/out-of-scope bodies. Funding history always remains `UNVERIFIED_COMPLETENESS` until a separately reviewed completeness procedure establishes otherwise, including special settlements and source finality. A valid-looking or empty response is not zero funding.

Before the first market request, append and externally timestamp the actual genesis ledger head bound to the manifest. Publish its exact document and proof, independently verify them, and require the collector to verify the public bytes and matching local ledger before it starts. Timestamp additional heads periodically and after completion. Publish the final head and proofs before any subsequent prospective paper decision, and retain an independent copy: the timestamp proves existence by its signed time but does not prove source authenticity or prevent coordinated deletion of local and public copies. If a run fails, never claim that the capture or its funding is complete. Exact input hashes must be fixed before any later paper decision. Follow the source split and explicit gap/EWM rules in the separately reviewed initialization design; any change requires a new prospective registration before looking.

On an interruption, the local records after the last externally timestamped head are an unanchored, incomplete tail. Preserve the tail and any local health record, prohibit automatic recovery, retries or use of those data, and require independent review before any further action. Periodic proofs do not establish that every request or unanchored tail was disclosed.