Evidence & provenance.

Returned physical evidence

Each report links the preserved SD image to its independently reconstructed result. Acquisition metadata and hashes identify the bytes used in that assessment.

Separation-based tolerance

The rationale follows eight unresolved decisions that retained-width calibration did not change. The declared rule allocates one third of each unassigned reference gap to each facing band and preserves the rest as unresolved. Preparation evidence identifies the literal instructions, frozen method, validation and installed image. The prospective comparison recognized 864/864 and 863/864 medians across two boots, compared with 854/864 and 859/864 for the baseline. Improvement recurred; boot 2 retained one unresolved decision. Returned evidence preserves the paired outcomes, all raw triples and the remaining rejection gaps.

Reference-width retention

The rationale follows repeated renewal benefit and two regressions from narrowed bands. The declared rule retains reference-derived widths while renewing centres. Boot 1 returned 3,888 observations and identical recognition under both rules: 856 correct and eight unresolved medians. The band comparison retains every exception; the acquisition archive includes matching repeated reads, independent reconstruction and source hashes. The comparison ended after one of three planned boots; repetition was not completed.

Calibration renewal investigation

The rationale follows recurring interval overlap and isolated excursions in returned observations. The declared comparison separates calibration age from observation duration, and the assessment retains failed calibration and every raw value. Preparation evidence identifies the native instructions, candidate image and physical procedure. Boot 1 results compare all four recognition paths; the acquisition record preserves the initial read discrepancy outside the journal and the matching repeated capture. Boot 2 results include all accepted calibrations and both favorable and unfavorable changes after renewal; its two complete capture reads matched. Boot 3 completes the declared series with a repeated net benefit and two median regressions. Its preserved acquisition includes the three-boot assessment.

Held-request investigation

The declared sequence extends consecutive acquisition at unchanged control requests. Its assessment separates diagnostic integrity from coverage under frozen bands. Preparation records identify the exact instructions, validated image and SD readback. The returned acquisition contains all 3,888 observations. Its raw capture and source archive preserve the rejected calibration and all six unresolved readings.

Transition investigation

The transition protocol defines matched direct and omitted-trigger sequences, balanced timing coverage and frozen assessment criteria. Its preparation archive links the source instructions, software validation, installed image and readback. The three returned captures preserve 3,240 observations per boot, two matching read-only acquisitions per capture and assessments reconstructed from each raw journal. Retention/update met its criterion in all three boots. The completed series retains 24 unresolved and 252 unavailable triggered readings; recognition remains unqualified.

Latest returned evidence

The unchanged arrival-boundary repeat completed all 72 conditions. Acquisition counts and resolved first-cycle predictions recurred across two boots; the close early margin remains unresolved. One pair retained a 0.630 microsecond write-relative gap, with overlapping boundary-relative intervals and equal counts. Comparison and next investigation.

Fixed-delay series integrity

All three preparations and returned records are preserved. The completed comparison retains unavailable recognition from run 3's rejected calibration. Its boot-volume assessment records the Windows dirty status separately from native journal integrity.

Verification and access

The source notebook checks artifact hashes, reconstructs returned assessments and validates reported outcomes before this edition is exported. Those checks run against the private evidence archive.

The public file manifest lists the size and SHA-256 hash of each exported file. It can identify changes to this edition; it does not independently authenticate the measurements. See evidence categories and limits.

Data availability

This public edition contains the notebook chapters, methods, reported results and figures. Implementation source files, executable images and complete SD captures remain in the private project archive.

Dotted references identify archived material that is not available for download here. Full independent reproduction of the reported analyses requires access to that material. The public pages retain the experimental limitations and identify the observations behind each conclusion.