# Response to the supplied Scientific Validation Review (2)

20 October 2026. Baseline reviewed here: local NetPhysics 0.6.0. New build: 0.7.0.
The supplied document contains a large older assessment and an added 0.4.1 review.
Its ratings are subjective review judgments, not independently validated scores.

| Finding | Resolution |
| --- | --- |
| Matched-history null rejection is not sufficient scientific progress | Accepted. E01-P puts frozen predictive performance on new independent specimens first. E01-S contrast remains a secondary diagnostic. Conventional history softening is implemented as a separate illustrative reference. |
| Candidate needs credible material rivals | Accepted. Protocol requires calibrated Mullins and rate-dependent alternatives, with a combined rival when justified. Reference curve is synthetic; no empirical rival fit or calibrated candidate is claimed. |
| Prediction metric, uncertainty and meaningful margin need definition | Implemented a frozen-prediction scorer with independent-unit weighting, paired cluster resampling, joint calibration-draw support, absolute and relative criteria, and leakage/error checks. Proposed 10% improvement and 5% normalized-error thresholds still require pilot justification. |
| Homepage still features the older memory intake | Updated the flagship to predictive testing and linked the current measured result and pulse audit. The older 1,330-row intake remains available and retrospective. |
| AICc weights absent | Partly stale: synthetic P03 already displays weights. Added the four-model measured-data weights to the metabolic page, explicitly training-only and conditional on that model set. No measured result changed. |
| SI units, citations and duplicate Planck source need correction | Already addressed in 0.6.0. Preserve the SI mapping, corrected source slugs/redirects and original downloads. Calibration remains missing. |
| Python/Node agreement described as independent verification | Clarified as a same-author port/implementation cross-check. It is not independent reproduction or human peer review. |
| Need ORCID, repository, DOI, registrations and external reviews | Still requires verifiable external records. Prepared archives, JSON plans and author manifests are not completed publication or registration. Do not infer someone's ORCID from a common name. |
| Evening-status paragraph reports 626 species, confirmed curvature, repository and three registrations | No supporting dataset, exact repository URL, registration links or report was supplied or resolved for those claims. Retain the verified 573-species result and label the other claims unverified, rather than inventing completion or declaring global nonexistence. |
| Report predicts the elastomer outcome from known softening | Narrowed: conventional models provide a rival prediction under explicit assumptions; an actual specimen needs calibration. An experiment is not predetermined by a review assertion. |

## Keep the verified metabolic result

The archived WR05 analysis has 573 eligible species from 5,416 original rows,
372 training species and 201 held-out species, with 99 families overall. The
primary 10 kg hinge RMSE is 0.200464 dex versus 0.198095 for the single line.
Its primary descriptive screen fails. Better training fit or a positive fitted
slope change does not establish held-out superiority, absence of all curvature,
a universal breakpoint, or a new biological law. Both alternative mass-column
eligibility counts checked for this review yield 573, not an explanation for 626.
A different dataset may exist, but its provenance and exclusions must be supplied.

The new table shows weights already present in the archived results JSON:
M0 0.000000154442; 10 kg hinge 0.400578427; quadratic 0.046817523;
20 kg hinge sensitivity 0.552603896. They normalize over these four models,
including the sensitivity model, and rely on the stated independent-error training
likelihood approximation. They are not probabilities that a mechanism is true,
phylogenetically corrected evidence, or scores on unseen observations.

## Remaining evidence gates

- Material selection, calibrated candidate/rivals, uncertainty and independent
  pilot data; then justified sample size and a frozen confirmatory plan.
- Exact repository URL plus immutable commit/release and an actual published DOI.
  Zenodo reserves a DOI before publication, but registers it when a record is
  published: https://help.zenodo.org/docs/deposit/describe-records/reserve-doi/
- Real registration links, version/date and a clear distinction between future
  experiments and retrospective analysis. OSF guidance:
  https://help.osf.io/article/330-welcome-to-registrations
- Actual reviewer identities, scope, conflicts and reports. Same-author ports,
  AI critiques and author-prepared CODECHECK manifests are not independent review.

The source-author pulse calculation remains conditional, with its Figure 3
time/parameter discrepancy open. The original 30-source audit queue is not
upgraded to “all read.” A newly checked documentation page does not amount to
full-text verification of every older cited paper.

No new account, submission, registration, DOI, outreach, paid affiliation or live
deployment was performed while preparing 0.7.0. This is an evidence-status report,
not an assurance of legal protection or a claim that external records cannot exist.