STRC Research

STRC h04 Outcome Chain Disposition 2026-08-21

3 min read

STRC h04 Outcome Chain Disposition 2026-08-21

h04 dB/audiogram outputs are NOT-MODELABLE: both stale paths are repairable, but zero of eight transfer anchors is grounded as a fraction-to-dB pair.

The two crashing audiogram scripts have mechanical post-split path defects. Repairing those paths makes code execution possible but does not make the resulting dB values scientifically quotable; the independent calibration chain remains unidentified.

Problem

Three scripts own the functional-OHC-fraction to ABR/audiogram layer. Two fail before loading existing JSON inputs, while the shared transfer script executes and emits dB results from eight hard-coded calibration points. This proof separates runtime repair from literature grounding.

Method

  • Reproduced all three scripts with complete stdout, stderr, and exit codes.
  • Confirmed both JSON inputs at the post-split models/ directory and both hard-coded pre-split roots as absent.
  • Enumerated all eight executable transfer pairs, every classification boundary, responder/functional thresholds, output clamp, and baseline dB values.
  • Queried the STRC QMD index, parsed BookLibrary RAG corpus, public primary records, Anna’s Archive, and Sci-Hub.
  • Retrieved Schuknecht and Gacek 1993 by exact DOI and parsed its 16 pages with MinerU vlm-engine; PDF SHA-256: 50c8843e7cec2d3062115a3cd660bfad1659aa70834ba140c22528f25ad835e9.
  • Required a canonical type: lit note plus a grep-verifiable paired functional-fraction and dB value for each anchor. No analog substitution was permitted.

Results

LayerFindingDisposition
mrna_lnp_audiogram_rescue.py runtimeInput exists (213,873 bytes); dead root is /Users/egorlyfar/STRC/modelsMechanical repair: use script directory
strategy_b_audiogram_rescue.py runtimeInput exists (233,119 bytes); dead root is /Users/egorlyfar/Brain/research/strc/modelsMechanical repair: use script directory
Bredberg 1968No canonical note; six bounded attempts yielded no primary text or stated fraction-to-dB pairMissing anchor
Schuknecht and Gacek 1993Primary text uses visual, tonotopic case correlations and says 21 cases are too few for statistics; no executable pair appearsDoes not state the calibration
Alleged Sun 2024Matching paper is Lv et al. 2024; it reports dose and ABR outcomes but no functional-OHC/transduction fractionMisattributed and non-identifying
Other six transfer pairsNo local primary note states both values in compatible unitsMissing anchors
dB categoriesCode is incomplete ASHA/Clark-like classification and conflicts with WHO 2021Labels also need separate correction

End-to-end grounded executable calibration pairs: 0/8.

The deeper model geometry also remains unresolved. A whole-cochlea scalar fraction cannot determine a frequency-specific audiogram when pathology and delivery vary by cochlear place; applying the same unsupported scalar curve independently to three zones does not identify it.

Verdict

  • Outcome: NOT-MODELABLE
  • Mechanical status: both stale ROOT assignments are fixable.
  • Scientific status: dB threshold, improvement, category, responder, and functional-hearing outputs remain quarantined and must not be quoted.
  • Repair condition: reopen only with locally parsed, canonical, frequency-specific paired observations and a declared model that addresses tonotopy and compatible dB units.
  • Forbidden shortcut: do not replace an anchor with a closest analog, refit a new band, or infer functional-OHC fraction from vector dose.

Ranking delta

  • Hypothesis h04: tier B → B | mech 3 → 3 | deliv 2 → 2 | misha_fit 2 → 2.
  • No tier/score move: invalidating the endpoint transform does not falsify upstream biology or change the existing LNP delivery bottleneck.
  • Next step unchanged. Literature-audit status changes from fixed to partial; the dB outcome chain remains quarantined.

Connections

Graph View

Backlinks