C04 fixture synthesis — storage evidence in four main results

This is a procedure trial on synthetic test fixtures, not a finding about external systems. The output remains in the workshop because its inputs are a local test corpus. No production review or public synthesis was replaced.

Evidence boundary

Selection: all four generated main reviews in the isolated fixture repository. Three declare code-grounded evidence; one declares doc-grounded evidence. All fixture analysis cutoffs are 2026-09-04. The bundle contains the full retained results, compact review identities, comparison contracts, method code, and instructions. The fixture repository has no legacy collection, local run state, or source checkout.

  • Bundle: /tmp/commonplace-c04-ao19i4vg/evidence
  • Manifest SHA-256: f3a118a842b37f2b62123c8962bba29cb3d5cecdbdf2671a07143d640e4fdb46
  • Matrix SHA-256: f778407205bcd3c6db209eff6757b175d3a474a14b1dea6e3bd1cded2ee6755f
  • Verification mode: local; no independent semantic checker.

The selection is a fixture population and supports no claim about the size, coverage, or prevalence of the production corpus. Its full result paths below are relative to the isolated bundle, not this checkout's published collection.

Code-grounded fixtures

One of one fixtures with an assessed storage set at wired-or-stronger evidence has file storage. That same fixture also has SQLite storage. It contributes one system to the denominator despite having two storage values.

The other two code-grounded fixtures are excluded from that denominator: claimed-fixture supplies only a claimed vector-storage value, and unknown-fixture leaves storage uninspected. The trial therefore supports no whole-population claim that all three use files or that the excluded fixtures lack file storage.

In wired-fixture, canonical record OBJ-1 describes session notes in Markdown files and a SQLite lookup index rebuilt from those notes. This supports the two-store classification and a primary-record/derived-index distinction in this fixture. Its Runtime account says no dynamic check was planned. The record therefore does not establish an observed rebuild, successful recovery, or a performance advantage.

Intended evidence citation: kb/reports/retained/agentic-system-analysis/AAS-2026-09-04-wired-fixture-01/result.md, Shared records → Operative objects, OBJ-1; full-result SHA-256 f06af7058957d128f9dd587b42d1383cc5eca0392889641a6e2ad6df96d98cdf. Its compact-review SHA-256 is 3cc4ea235b42720379bf61a0f56c0367258730f05a09cc95e5ca06d934a3225f.

Doc-grounded fixture

doc-fixture declares a claimed SQLite value. It appears here as documentation coverage and contributes nothing to the implementation/operation denominator. No inspected or observed SQLite behavior is inferred from that value.

Query ledger

  • Fields: source_tier, storage_substrate, storage_substrate_assessment, storage_substrate_basis.
  • Eligible: code-grounded, known storage set, basis wired, observed, or causally supported. Numerator: membership of files in that set, once per row.
  • Numerator / denominator: 1 / 1.
  • Included and matched run: AAS-2026-09-04-wired-fixture-01.
  • Excluded runs: AAS-2026-09-04-claimed-fixture-01 (claimed basis), AAS-2026-09-04-unknown-fixture-01 (uninspected), and AAS-2026-09-04-doc-fixture-01 (doc-grounded).
  • Withheld: whole-population storage prevalence, negative storage claims for excluded rows, observed operation, recovery reliability, and causal benefit.

The executable query, final verification, and broader rejection checks are recorded in C04 acceptance.