Adaptive Acquisition Evidence

Status

current

Published

2026-10-26

Modified

2026-10-26

Category

methodology

Included changes

1. 01Optional shadow observation reuses the baseline acquisition, with an explicit source allowlist, a per-scan budget and up to three candidate recipes. The baseline result is retained.
2. 02Private captures, immutable receipts and frozen-input replay preserve evidence for comparison. Train/holdout and independent-label gates keep unsupported lessons ineligible for operational use.
3. 03The Source Reliability workbench separates scan outcome, saved observations, observer errors, baseline acquisition errors and summary-write failures. Historical, disabled, partial and unavailable summaries remain distinct.
4. 04Explicit locale or regional path changes are rejected as source-variant drift. Acquisition dimensions distinguish current live, partial and historical evidence.

Interpretation boundaries

• 01Shadow observation is disabled by default and requires a private operator configuration. Installing this release does not enable a production pilot.
• 02No automatic recipe promotion, model training or calibrated confidence is claimed. Provisional labels cannot validate their own lessons; broader independent evaluation remains planned.
• 03A recorded comparison is not a source count, extraction-quality verdict or legal conclusion. Private captures and raw errors are not exposed in the panel or public APIs.

Evidence links