Skip to content

ADR-0054 - A07 historical acquisition day-boundary v1

Status: ACCEPTED Date: 2026-10-02

Context

Design gate issue #252 was raised from the Omega consumer path (NeoNix-Lab/omega#15, U5). Omega needs multi-week historical datasets for walk-forward and DSR/PBO validation, while A07 is currently credited only as a narrow Bybit BTCUSDT first vertical over one UTC day. The working question was whether A07 v1 should grow a native multi-day/date-range acquisition path, or whether callers should explicitly orchestrate one day at a time.

The existing accepted authority is deliberately day-shaped:

  • ADR-0023 and the Producer-Consumer Conformity contract prove the first Bybit BTCUSDT vertical on the 2024-01-15 UTC day.
  • A04 declared coverage and S13/S14 publication work on explicit half-open coverage intervals and natural partitions.
  • ADR-0033 repair works over explicit defects and candidates; it does not own acquisition, scheduling, source selection, or a generic self-healing loop.
  • ADR-0044 keeps long gaps explicit unless an attributable repair source/path proves the missing support and integrates through A10 semantics.

Native date-range acquisition would therefore not be a harmless convenience wrapper if it tried to claim one aggregate range as complete without preserving the existing per-partition evidence and repair boundaries.

Decision

A07 v1 remains a one-UTC-day, Bybit BTCUSDT linear historical acquisition profile. Multi-day consumers must build date ranges by an explicit caller-owned day-by-day loop, where each UTC day is acquired, canonicalized, covered, certified, published and later read as its own evidenced partition.

This is the accepted v1 answer, not a claim that A07 can never widen. A future native date-range acquisition path may be proposed only as a separate design gate and implementation slice. That future slice must prove that it preserves the already-frozen guarantees instead of bypassing them:

  • one explicit half-open declared coverage interval per natural partition;
  • S13 certification and S14 publication evidence per admitted partition;
  • no silent aggregation of source failures across days;
  • no weakening of ADR-0033's repair/cutover model;
  • no claim that elapsed time, missing sequence evidence, or an empty provider response proves completeness without attributable source evidence.

The caller-owned loop may still offer a higher-level user experience: it can enumerate days, invoke the A07 path repeatedly, collect per-day outcomes, and surface a range-level summary. That orchestration is not A07's own semantic contract unless and until a future ADR promotes it.

Consequences

  • CAPABILITY_DAG.md and CAPABILITY_MAP.md must describe A07 as the one-UTC-day BTCUSDT-linear v1 profile, with multi-day acquisition performed by explicit caller orchestration.
  • Omega and other consumers needing multi-week datasets should loop day-by-day, then consume the resulting eligible canonical partitions through DataGateway.scan() / DataGateway.read().
  • This issue does not implement native date-range acquisition and does not add a scheduler or job runtime.
  • ADR-0033 remains unchanged: repair stays explicit-defect-driven and does not become the source-selection or multi-day acquisition owner.
  • A future widening of A07 should be tracked as a new issue and must include executable proof for the range orchestration semantics it claims.

Parent tracking: #232. Closes issue #252. Found via NeoNix-Lab/omega#15 (U5).