STRC Research

01-outcome-chain-disposition

3 min read

h04 dB/audiogram outcome-chain disposition

Verdict

NOT-MODELABLE.

The stale paths are repairable. The dB outcome chain is not repaired by making the scripts execute: zero of eight executable functional-OHC-fraction-to-threshold pairs passes the required local-primary provenance test, and the claimed sources do not identify the fitted scalar transfer function.

Defect 1 — stale-path bug

Class: mechanical; fixable.

  • models/mrna_lnp_audiogram_rescue.py points ROOT to dead pre-split /Users/egorlyfar/STRC/models.
  • models/strategy_b_audiogram_rescue.py points ROOT to dead pre-split /Users/egorlyfar/Brain/research/strc/models.
  • The inputs exist at the canonical post-split paths:
    • /Users/egorlyfar/Missions/STRC/models/mrna_lnp_pkpd_integration.json — 213,873 bytes.
    • /Users/egorlyfar/Missions/STRC/models/strc_mrna_strategy_b_pkpd.json — 233,119 bytes.
  • Baseline reproduction gives exit 1 for both scripts at the stated read_text() calls. This is a path-resolution failure, not a missing-model artifact.
  • The exact live stale-root scan returns 86 Python hits under models/, confirming a widespread migration residue. This route repairs only the two named owners; a vault-wide migration is outside scope.
  • Surgical repair: ROOT = pathlib.Path(__file__).resolve().parent.

Defect 2 — calibration-grounding gap

Class: scientific identification failure; not repaired.

Bredberg 1968

  • No canonical sources/lit/ note.
  • Six bounded attempts found PubMed PMID 4886545 and secondary citations, but no primary full text or passage.
  • No evidence maps 30% functional OHCs to 53 dB.
  • The executable row is internally inconsistent: x=0.30, its inline comment says approximately 50% OHC function, and the narrative table says 50 dB rather than 53 dB.

Anchor disposition: NOT-MODELABLE.

Schuknecht and Gacek 1993

  • Exact primary PDF retrieved and parsed with MinerU under GROUNDING/; no canonical sources/lit/ note exists.
  • The paper studies 21 presbycusis cases using frequency-place-aligned audiograms and cytocochleograms. It explicitly says the sample is too small for statistical analysis and that judgments rest on visual correlations.
  • Exact primary-text search finds no 0.30 -> 53 dB pair, no functional-OHC scalar, and no transduction measure.
  • It describes heterogeneous, localized sensory, neural, strial, conductive, mixed, and indeterminate pathology. That geometry cannot identify one frequency-agnostic curve from whole-cochlea functional fraction to threshold.

Anchor disposition: NOT-MODELABLE.

“Sun 2024 OTOF Cohort 2”

  • The matching Lancet paper is Lv et al. 2024, DOI 10.1016/S0140-6736(23)02874-X; first author is not Sun.
  • The primary abstract reports vector dose and patient ABR outcomes (including 38, 40, 48, and 55 dB at 26 weeks), but it does not measure functional-OHC or transduction fraction.
  • The existing sources/lit/2024-lv-aav1-otof-lancet-trial.md cannot make 0.55 -> 38 dB grep-verifiable because the 0.55 input is absent.

Anchor disposition: NOT-MODELABLE.

Remaining executable anchors

  • 1.00 -> 20, 0.70 -> 30, 0.40 -> 47, 0.15 -> 64, 0.05 -> 76, and 0.00 -> 85 have no locally parsed primary note stating the pair.
  • DB-OTO cohort labels are author-derived interpolations, not measured functional-OHC fractions.
  • “severe,” “severe-profound,” “normal,” and “DFNB16 / Strc-/- baseline” are labels, not parameter provenance.

Per-row disposition: all NOT-MODELABLE.

dB labels are a separate defect

The classification code is ASHA/Clark-like, not “WHO/ASHA”:

  • It merges ASHA/Clark Normal (-10 to 15 dB HL) and Slight (16 to 25 dB HL) into Normal <=25.
  • Its 26-40, 41-55, 56-70, 71-90, and 91+ edges approximately follow ASHA/Clark.
  • WHO 2021 instead uses <20, 20-<35, 35-<50, 50-<65, 65-<80, 80-<95, and >=95 dB.
  • Neither retrieved standard defines <=55 dB as “functional hearing”; Lv 2024 does not define a >=20 dB responder rule in the retrieved primary abstract.

Correcting labels would not identify the missing transfer curve.

Consequence

  • The two ROOT assignments should be repaired so runtime failures report the true state.
  • All three scripts must carry a visible NOT-MODELABLE quarantine: no dB threshold, improvement, category, responder, or “functional hearing” result may be quoted as a literature-grounded prediction.
  • No replacement band, analog calibration, refit, or interpolation is recommended.
  • Reopening requires, at minimum, frequency-specific paired observations with explicit units and locally parsed canonical notes; even then the scalar-identifiability problem requires a new model contract.

Graph View