INDEPENDENT REVIEW — 24 October 2026

Verdict: Do not deploy the attached rev2 homepage wholesale. It regresses
verified source information and overstates several scientific deductions.
The current concise briefing is a better base. A methods appendix incorporates
the useful additions while preserving the existing corrections and pilot link.

Corrections and dispositions

1. Hutchison 2016 access: resolved. MIT hosts the complete published PDF:
   https://cba.mit.edu/docs/papers/16.04.minimal.pdf
   Title, authors and DOI 10.1126/science.aad6253 match. Relevant sections read;
   do not count retrieval alone as a complete reading. Powner remains unread
   in full in this pass. No institutional subscription is needed for Hutchison.

2. Reading counts: unverified. The supplied ZIP contains a verification report
   but no complete itemized source-reading register. Its 31/27 totals cannot
   be independently certified. Do not reward a paper quota over claim accuracy.

3. Curvature standard errors: 'always finite' is wrong. Invertibility and model
   regularity matter. Practical and structural identifiability differ. A finite
   profile scan without a crossing is not proof of a mathematically unbounded
   confidence interval. Raue DOI 10.1093/bioinformatics/btp358, sections 2.2–2.3,
   read at the publisher. No universal runtime guarantee is warranted.

4. Mechanical analogy: Lahini main text and captions read from arXiv
   1608.02429. Its model and reference 37 support the distributional analogy.
   Physical rates and model-sensitivity eigenvalues are not interchangeable.
   No biological-aging mechanism follows from the analogy. An erasure test is
   useful, not a universal necessary condition for the existence of memory.

5. Functional endpoint: keep tracer permeability primary, resistance supporting.
   PMC6212516 relevant methods explain resistance's ion-transport dependence.
   PMC12832454 relevant results and Figure 4 supply an empirical discordance:
   increased resistance alongside increased macromolecular flux. This is an
   irradiation study, not evidence that Karam's proposed intervention works.

6. Selection and growth: constant endpoint counts and marker abundance do not
   rule out balanced turnover or subpopulation selection. Differentiated ALI
   cultures do not remove proliferation by construction. Rock 2009 abstract,
   DOI 10.1073/pnas.0906850106, describes airway basal progenitors; full article
   not read in this pass. Correlated functional and growth changes leave the
   mechanism unresolved. Ion substitution alone is not a unique causal test.

7. Power: replaced approximate shifted-central-t calculation with noncentral-t
   power. With delta=tau=.15, sigma_w=.12, m=3, alpha=.05: 14 evaluable donors
   meet 80%; 17 yield .8994678 and 18 are needed for a strict 90%. This is an
   illustrative model, not estimated pilot data or a permeability sample-size
   recommendation. Independent quadrature validates the calculation. The
   original author's calculator is preserved in the untouched input archive.

8. Missing outcomes: distinguish enrolled from evaluable donors. Do not keep a
   wholly missing donor in the statistical denominator without a stated model.
   Prespecification must include the primary contrast and time, not endpoint
   name alone. Recovery fractions need denominator and reference rules.

9. Anthrobot regression: rev2 revives 'S5 unavailable' and suggests a preferred
   smaller percentage. The current verified page already names the methods.
   Preserve both conflicting figure/text values and the unresolved sample map.
   No email was sent, and no contact-status date has been changed.

10. Reproduction scope: raw IDATs are not necessary to score an already
    processed methylation matrix. Score audit, clock scoring and raw pipeline
    reproduction are distinct. Do not claim nobody has reproduced anything.
    This pass did not rerun prior ClockQC data analyses or score the new pilot.

11. Other overclaims rejected: donor spread is not automatically biological
    signal; preserved tissues have cellular identity even if personal identity
    is untested; a public protocol without results does not establish that no
    experiment happened. Preserve the current qualified venture descriptions.

12. Integration: the attached homepage is not applied. It would overwrite the
    concise briefing and existing corrections. Its static og:url duplicates
    the PHP-injected tag; its .py download is blocked by current server rules.
    Instead add a routed methods page and explicitly allow its code ZIP. Keep
    the current navigation, contact invitation, pilot and existing analyses.

Verification boundary
The new calculator is executed and tested. Local links, fragments, metadata,
route registration, sitemap and archive contents are checked. No IDAT work,
new clock scoring, complete independent reading of the prior register, email
delivery or production deployment is claimed. Browser rendering is not newly
certified by this review; the new page reuses existing responsive site styles.

Deployment
Use the full public_html archive OR the patch ZIP, not the supplied old rev2
homepage. Extract at the existing public_html root. Back up first. The patch
contains complete replacement metadata files based on the current supplied
deployment tree; reconcile if production has changed independently.
After upload, open /research/living-systems/research-notes/ and its code ZIP,
check mobile layout, and confirm one og:url in the PHP-served HTML. The new
pilot remains blocked from scoring until its own publication gate is satisfied.
