EXPANDED ORIGINAL MANUSCRIPT â€” EDITORIAL UPDATE
Scope: PHASE-TTC: From Motion to Time, original full manuscript.

Owned files changed:
- sections/introduction.tex
- sections/evttc_competition.tex
- sections/discussion.tex
- sections/protocol.tex
- sections/competition_details.tex
- tables/evaluation_populations.tex
- scripts/make_population_table.py

Preservation checks
All original includegraphics paths, input tables, labels, and citation commands
in the six manuscript files remain present. No original figure or table was
removed or replaced by the seven-page manuscript. The corresponding source
word counts increased from 2,971 to 4,284 (~1,313 words added). Paragraphs describing
the original inference, baselines, paired disagreement, signed traces, latency,
grouped fitting, metrics, and transfer limitations remain. Only statements
rendered stale by the new target audit were rewritten, preserving their dated
historical interpretation.

Corrected claims
1. Lack of mapped 434 targets now describes the 28 October snapshot. The
   corrected 1 November local audit recovers all 434 targets, superseding earlier
   423-target summaries. It does not turn output disagreement into error.
2. Original-vs-adapted local error is separate from the archived Garl benchmark,
   the 468-query Extended-new transfer, published three-sequence RTE, the 2,608
   cadence cohort, and native 5,805-query output-sensitivity replay.
3. Negative-prediction interpolation belongs only to the documented local
   evaluator; the signed original prediction files and figures remain intact.
4. Slider-750 local 283.931058% versus organizer 132% remains unresolved. The
   adapted local 76.071442% is not ranked directly against the organizer value.
5. The quality-adapter improvement is historically exposed. Exclusion from
   the final fit does not erase prior model-selection exposure.
6. Canonical timing and the 48-fit router experiment retain failed promotion
   controls despite some favorable full-population aggregates.
7. Native replay measures sensitivity, availability, sign and numerical drift,
   not Test12 candidate accuracy. Template token count is labeled as such,
   rather than implying that native candidate tokens received official scores.
8. Conclusion now carries new positive and negative results while retaining
   original tail-error and deadline limitations. Official ranking remains dated.

New section references: sec:quality-adaptation, sec:canonical-cadence.
New evidence source IDs use N01--N06, avoiding original E-series collisions.

Generator verification
make_population_table.py reads original public ledgers plus the copied adapted,
canonical and native JSON. Population counts are derived, not handwritten.
The first regenerated version matched the reviewed table exactly; the final
native row changes only 'scored tokens' to 'template tokens' for clarity.
Ruff check and format check pass after formatting. A second generation produces
the same table bytes as the first; no training or sensor inference was run.
