AUDIT — h12-sonogenetic (Sonogenetic STRC Computational Proof)

  • Tier: C · mech 2 / deliv 2 / misha_fit 2 · next_step: Speculative; no near-term path
  • lit_audit before: deferred (2026-04-23)
  • lit_audit after: partial, 1 blocker
  • Grounding: GROUNDING/hypc-h12-sonogenetic.md
  • Proportionality: C-tier, speculative, no scripts. Short confirmatory pass.

Scope

hypotheses/h12-sonogenetic/scripts/ is empty; the hypothesis is carried by one phase note, STRC Sonogenetics Bifurcation Analysis.md (13.1K). The audit question is therefore not “are the scripts sourced” but “does the bifurcation conclusion depend on any unverified external number.”

Parameter-provenance table

#ParameterValueTreatment in the analysisSourceStatus
1Fold induction, 6×NFAT construct62×point value, used as the anchor”Wu et al. 2023” — no PMID, DOI, journal or title❌ UNVERIFIABLE
2Kd_CaN500 nM nominalswept 100–2000 nM, then 500–5000 nM✅ n/a — swept
3NFAT_threshold0.05 nominalswept 0.01–0.30, then 0.05–0.40✅ n/a — swept
4Hill coefficientswept 2–8✅ n/a — swept
5Saturation / dose-response table (0–70 dB → 16–198% of target)model outputin-house computation
6Ca²⁺ toxicity thresholds (>1000 nM at sustained >100 dB)model output; note states this is a property of all hair cellsin-house
7CV < 1% stability resultmodel outputin-house

Finding 1 — the swept parameters are NOT blockers

Worth stating explicitly because the obvious skim gets it backwards. The note’s line 47 flags NFAT_threshold (0.05) and Kd_CaN (500 nM) as “both too low,” which reads like two unsourced constants. But the analysis never rests on those point values: it sweeps Kd_CaN across 100–2000 nM and again 500–5000 nM, and NFAT_threshold across 0.01–0.30 and again 0.05–0.40, 245 combinations in total. A conclusion robust across a 10–50× swept range does not need a literature value for that parameter, and demanding one would be audit theatre.

This is the correct engineering response to an unmeasured parameter, and h12 did it by construction where h09 had to reach for it retroactively (Phase 4n Monte Carlo). A C-tier speculative hypothesis is, on this narrow point, better disciplined than the S-tier one.

Finding 2 — one unverifiable external anchor

STRC Sonogenetics Bifurcation Analysis.md:87:

“Even if fold induction is 3x higher than Wu et al. 2023 (200x instead of 62x), protein stays under 200% of target.”

and line 156: ### Wu et al. 2023 — used 6xNFAT only.

ls sources/lit/ | grep -i "wu\|2023" returns three Wu papers, none of them this one (a PVDF-TrFE piezoelectric paper belonging to h02, a MYO6 review, an ACBD7 paper). “Wu et al. 2023” carries no PMID, no DOI, no journal, no title.

That is the signature of every phantom this vault has already caught — “Masaki 2009”, “Gueta 2006”, “Ren 2011”, “Sharma 2018”, “Krey 2015”. I am not calling it a phantom: it may be a real paper that was simply never filed. I am recording that h12’s single external quantitative anchor is unverified, which is enough to keep the hub off fixed.

Mitigating, and stated for fairness: the analysis is deliberately built so the conclusion survives a 3× error in this number (“even if fold induction is 3x higher… protein stays under 200% of target”). The saturation cap dominates. So the conclusion is fairly robust to the citation being wrong; the citation is still unverified, and a bare author-year in a research vault is a liability regardless of how much weight currently sits on it.

Verdict

lit_audit: partial · lit_audit_date: 2026-08-21 · 1 blocker

lit_audit: partial
lit_audit_date: 2026-08-21
lit_audit_blockers:
  - "'Wu et al. 2023' (62x fold induction, 6xNFAT construct) — UNVERIFIABLE CITATION. Cited twice in STRC Sonogenetics Bifurcation Analysis.md (lines 87, 156) with no PMID, DOI, journal or title. No matching file in sources/lit/ (checked 2026-08-21: the three Wu entries are a PVDF-TrFE piezoelectric paper, a MYO6 review and an ACBD7 paper). This is h12's only external quantitative anchor and it carries the phantom-citation signature that the 2026-04 audits found in h02 and h05. Mitigating: the bifurcation conclusion is explicitly built to survive a 3x error in this value, so little currently rests on it. Next step: identify the real paper, file it to sources/lit/, or mark the 62x as an in-house assumption. NOT a modelling blocker — Kd_CaN and NFAT_threshold are SWEPT (245 combinations across 10-50x ranges), so they correctly need no literature point value."

Ranking delta

C-tier HELD. No mech/deliv/misha_fit change. See RANKING-DELTA.md.