# Controlling 28 October initialization and startup amendment

144.51% gate: UNMET. Eight launch attempts were registered on 27 October and are preserved. The v9f ledger records an execution at 23:05 UTC in all 22 paper accounts, inside an over-90-second monitoring gap, followed by a failed path at 23:06. That short, interrupted attempt does not establish a complete forward return or promotion. Preserve every prior registration, source record, modeled fill and failure; do not backdate or merge it into a separate new study. This document controls over the earlier drafts, including PROTOCOL_TODAY.md. Its original declaration preceded the first 28 October acquisition; later revisions below were adapted after inspecting initialization evidence and earlier operational failures. This is not an untouched blind initialization. The source snapshot before this amendment is preserved in v9_draft_before_20261028.zip.

The owner authorizes historical public market prices/funding for separate-study indicator initialization and requests the earliest working prospective start. On28October, refresh only the newly completed27October UTC day [2026-10-27T00:00:00Z,2026-10-28T00:00:00Z), after this whole implementation is registered, signed, published and independently verified. This is an incremental initialization acquisition, not a backtest or a forward return. Retain every source response, request/receipt time, error and hash. No protected research_v10_paper path is a source. Do not read or modify that study.

Initial data inputs are the signed v8 panels through26October plus the new27October candle and observed funding-grid sum for each currently supported crypto PERPETUAL/USDT/TRADING exact identity. The request universe is taken from current retained exchangeInfo before any price request. If source onboarding proves a name did not yet exist on27October, retain that fact and no invented bar; it remains ineligible. Request one complete daily kline and one1000-limit funding page per applicable identity, in fixed sorted order. No older historical date is requested by this update, and no saturated, conflicting, missing or invalid response supplies zero data. Unresolved established prices, missing or invalid required funding responses and missing known settlements block startup; structurally valid unresolved grids follow the later controlling eligibility rule. New/relisted identities cannot inherit earlier-contract bars. Grid consistency is observational only; it does not prove funding completeness or finality.

This initialization phase is a one-use acquisition. A durable start marker without a finish marker, any failed required capture, or missing deadline causes START_FAILED, not an automatic retry or a silently delayed first rebalance. All responses remain in the append-only source ledger. They are never posted as position funding costs. Live position funding begins only for actual holdings after the prospective first fill. Initialization progress is reported separately from trading activity.

The first future five-minute UTC boundary is chosen mechanically at least45minutes after registration preparation begins, on28October only, excluding an exact whole hour. Exact first fill, cutoff(fill-8minutes), decision(fill-5minutes), and365day endpoint are in the signed commitment. The longer lead permits the fixed full-universe initialization and fresh observation checks; it is not a guarantee of readiness. No price, signal or outcome chooses this time. If there is no remaining valid same-day slot or any readiness/proof misses it, fail rather than roll to another date. At least three distinct complete live schedule/clock observations must span90seconds before cutoff. Every gap and both coverage endpoints from the registered allowed source-start boundary through fill must be<=90seconds, using the actual premiumIndex and fundingInfo receipts and source-event bindings, not the later note time.

First signals use completed data through27October, with current as-of metadata, source prefix,60-current-contract-valid-bar rule and the separately evidenced observed funding availability. Subsequent daily rebalances remain00:50 cutoff/00:55 decision/01:10 fill. The initial partial live calendar day is28October: funding before activation on that date is not requested or imputed. For that one day only, the separate boolean eligibility override uses27October initialization availability plus fresh complete live schedule coverage; its numerical unknown funding panel stays NaN. No numeric rate or cashflow is invented.29October and later full days use their own complete observed-grid tests. The selected fixed scoring paths consume funding availability, not numerical funding sums; independent invariance tests are required.

Primary evaluation samples are exactly365 non-overlapping24hour intervals anchored to the first fill time. Initial equity is the declared10000USDT seed immediately before first friction; terminal equity is after actual modeled closing spread/fees. Intermediate anchored views use only source marks conservatively received before that exact timestamp and the accounting prefix then available. Missing marks or known pending costs produce null and an incomplete path. Regular01:10 rebalance views are separate, and are not treated as365 equal-duration returns. Full final net reconciliation and family inference remain future separately implemented/reviewed work; no provisional view or annualization meets the owner gate.

The Cloudflare public service retains immutable submission hashes, signed checkpoint documents/proofs, and the latest complete dashboard snapshot. Prior full snapshot bodies, raw market evidence and the complete accounting/trade history are retained on the collector computer. A digest does not recover lost data. This is local durable accounting with an external witness, not a complete off-machine data backup. Loss or missing history prevents a promotion claim. Public snapshot histories are bounded to the latest50 trades per account,367 anchored points,30 completed daily candles and120 observed mark points per held identity; all complete private ledger records remain retained. Public displayed anchored equity is rounded to8decimals; exact arithmetic stays in the ledger. Hourly signed checkpoints supplement decision/intent/execution proofs. The public service/browser verify snapshot-digest binding, not exchange authenticity or timestamp signatures; the operator verifies the latter.

The Windows collector must remain available, connected and signed in. The exact previous input-only FrontierWarmupCollector-f6b006 task is retired only after the new registration/self-check is verified, to avoid duplicate source quotas; its files/evidence stay unchanged. Never change a protected-study task. The new task must not overwrite an existing identity. Disk space and runtime/clock/network failures halt rather than fabricate evidence. Benchmarks are synthetic lower bounds and do not guarantee one-year uptime or capacity. All22 fixed accounts and both cost paths remain visible, including failures. No real exchange orders are submitted.

Revision 2026-10-28T07:00Z (disclosed before re-registration; amendment 11): an established identity whose 27October observed funding grid is permanently inconsistent (for example an intraday funding-interval switch, observed on ARKUSDT) is recorded with UNRESOLVED evidence, contributes NaN funding input and availability=false, and is excluded from funding-dependent eligibility for the affected day. It does not block startup of the separate study. This preserves the no-zero, no-imputation, fail-visible rules while keeping the program startable; candle, streak, schedule-coverage and source-binding requirements are unchanged.

## Revision 2026-10-28T07:40Z (clarification to amendment 11)
The unresolved-grid treatment (UNRESOLVED evidence, NaN funding, eligibility exclusion, never zero-imputed) applies identically to the bootstrap day and to every later full day: an established contract whose observed funding grid can never reconcile is excluded from funding-dependent selection on those days rather than blocking the study. A follow-up adversarial review (same day) confirmed no zero-imputation or silent-skip path exists under this rule.

## Revision 2026-10-28, after failed attempt m, before attempt n
All thirteen attempts a through m remain preserved and their commitments are bound into the next manifest. Attempts i through m include additional 28 October operational and initialization observations. Attempt m captured 523 initialization symbols and failed on a genuine monitoring gap before any account, target or execution record. The fixed eleven books and both cost paths remain unchanged; the new prospective study is an adapted operational restart, not an independent discovery or untouched preregistered backtest. No past observations count as new forward performance.

Uniform funding eligibility rule: for initialization or a completed full day, any structurally valid retained funding page whose observed grid does not reconcile is labeled UNRESOLVED with a null numerical sum and excluded from funding-dependent selection for that day. This is not evidence that the grid defect is permanent or that an exchange omission did not occur. Empty, malformed, conflicting, saturated or missing required source responses, missing declared expected live settlements and missing established price inputs remain failures under their existing rules; they may not be relabeled as grid exclusions. Retain each unresolved identity, raw funding records and status in the evidence chain. Held-position known funding obligations and cashflow reconciliation are unchanged: unresolved due costs prevent valid net NAV and promotion, never become zero.

The monitoring scheduler uses actual premium/funding receipt deadlines to limit sleep and refreshes immediately after bootstrap and execution, and checks whether a refresh is due after decision work and other blocking operations. A real receipt gap over 90 seconds remains fatal. The first future five-minute slot is still selected mechanically with at least 45 minutes of lead; no prior fill is relabeled or resumed. The replacement Windows scheduler has a new attempt-specific identity and a no-market environment check precedes collection.

Unresolved funding availability excludes positions only at the archived eligibility stage. The archived volume ranks still use valid price/volume inputs across their original universe; unresolved names can occupy a rank slot without being replaced by lower-ranked eligible names. No ranking formula is changed by this restart.

Controlling quote rule: source quote age at local observation is at most 10 seconds; with the +/-2-second venue clock bound its conservative age is at most 12 seconds. Venue timestamps may lead actual local receipt by at most 2 seconds; bytes must actually be received before construction. Venue time minus 2 seconds must be strictly later than decision plus 45 seconds. Conservative quote age at the modeled fill is at most 30 seconds. Accounting receipt may occur up to 5 seconds later, implying conservative quote age at that actual receipt at most 35 seconds. The fixed spread and fee remain a paper model, not proof of executable liquidity. This supersedes conflicting earlier DETAILS clauses.

Controlling clock probe rule: each scheduled clock acquisition may make at most three complete probes when the preceding complete response fails the unchanged +/-2-second full request/receipt interval bound. All probes are retained; status/format/network/latency or rate-limit failures remain fatal. There is no unrecorded retry or relaxation of the accepted clock uncertainty.

Allocation adaptation disclosure: the current portfolio implementation, introduced during earlier operational attempts, excludes new allocations whose modeled changes exceed observed top-of-book side quantity, preserves held-position exit depth requirements and renormalizes the remaining positive/negative sides to equal gross totals. If a side empties, the allocation becomes cash. This is an execution adaptation to the archived targets, not an exact implementation of their historical backtest. Exact adapted actions, exclusions, quotes and costs are retained in the intent. The new prospective result cannot be presented as validation of the unchanged archived execution model.

Partial first-live-day exception: a valid empty historical funding response for the post-activation part of 28 October can remain explicitly PARTIAL_FIRST_DAY with null numerical sum only if no retained expected settlement is missing. This does not prove no event occurred. Full completed-day empty funding responses remain missing-source failures. The prior-day availability override remains boolean only, and all held-position cashflow obligations remain independently due.

Final startup-lead amendment, before attempt n signing: the mechanical lead is 30 minutes after preparation begins, superseding 45 minutes wherever earlier text conflicts. This operational adaptation uses the elapsed acquisition duration of failed m, not current prices, targets or returns. The first five-minute UTC boundary at least 30 minutes after preparation, excluding a whole hour, is frozen in the manifest. Cutoff remains fill minus 8 minutes, leaving approximately 22 minutes from preparation for acquisition and witnessing; one previous approximately 11-minute capture does not guarantee readiness. Activation/publication margins, all continuity guards and failed-start preservation are unchanged. A miss fails this attempt; no silent shift or historical fill is permitted.

Attempt o superseding failed n: n never acquired market data. Its commitment was signed at 14:55:34 UTC, but activation at 14:56:15 was only 41 seconds later and failed the required >=45-second check. Both artifacts remain retained and n is now included with a-m in the exact fourteen-prior-attempt manifest. The replacement activation function waits at least 46 monotonic seconds after verified commitment retrieval before creating/signing its artifact, then still checks the independent TSA timestamps and existing cutoff margin. No strategy, initialization interval, fees, execution formulas, outcome window or readiness rule changes. The next future start remains mechanically chosen with the already declared 30-minute lead.