Reviewer premortem drafted.STRC h09 Reviewer Premortem Placeholder 2026-04-27 now states the hostile-read objections, current answer, residual risk, and minimum next check before external review. Ranking delta: S held; mech 5 / deliv 4 / misha_fit 3 unchanged; reviewer-pack prose does not move tier.
2026-04-27 (15:58 - reviewer-pack placeholders staged for next sessions)
Reviewer-pack queue staged. Created STRC h09 Reviewer Pack Queue 2026-04-27 plus four future-session placeholders: premortem, precedent table, orthogonal WH2 controls, and transfer memo. Ranking delta: S held; mech 5 / deliv 4 / misha_fit 3 unchanged; next_step now points future agents to the reviewer-pack queue before any optional Phase 4l work.
2026-04-27 (15:43 - Phase 4k full AF3 contact audit inconclusive; S held)
Phase 4k full contact audit.scripts/phase4k_af3_contact_parser.py rerun after h09_phase4k_construct_x4_x_actin_x3 completed and downloaded. Result: any geometric pass=true (tandem WH2 proxies show 4 contacts), any confidence-supported pass=false (0 confidence-supported WH2-actin contacts across all 5 jobs; construct_x4 ipTM 0.37 / pTM 0.49 / 0 contacts). Phase 4k does not independently close N>=4, but does not reverse Phase 4m. Ranking delta: S held; mech 5 / deliv 4 / misha_fit 3 unchanged. →STRC h09 Phase 4k Fibril Actin Geometry 2026-04-27
Phase 4k staged. 5 AF3 JSON jobs at ~/STRC/models/af3_jobs_2026-04-27_h09_phase4k_fibril_actin/ (3 multi-copy count=N construct jobs + 2 covalent-tandem-WH2 forced-proximity proxies). Builder: scripts/phase4k_af3_fibril_actin_builder.py. Tests if N ≥ 4 simultaneous WH2 × actin contacts is geometrically achievable. Pass per job: overall ipTM ≥ 0.45 AND ≥4 distinct WH2-actin engagements (assessed from full_data ZIPs). If any of jobs 3, 4, 5 PASS → Phase 4j conservative scenario closes → S promotion. Combined payload phase4k_combined_5jobs.json ready for one-shot Upload-JSON when AFS quota refreshes ~midnight UTC.
Phase 4m STAGED + pre-registered. Comparative ipTM-vs-Kd calibration on Chereau 2005 WH2 × G-actin family (WAVE2 52 nM, WIP 160 nM, MIM 230 nM, WASP 250 nM, Tβ4 760 nM) + h09 tail91 N-term WH2 → predicted Kd_mono soft floor (replaces Husson Tβ4 5–10 mM in Phase 4j). Reference WH2 sequences retrieved from UniProt 2026-04-26: WAVE2 (Q9Y6W5 433-464), WIP (Q14155 29-60), MIM (O43312 724-755), WASP (P42768 430-458), Tβ4 (P62328 2-44). Builder + 6 AFS JSON jobs at ~/STRC/models/af3_jobs_2026-04-27_h09_phase4m_iptm_kd_calibration/; combined payload phase4m_combined_6jobs.json ready. Analysis script phase4m_iptm_kd_analysis.py stub-coded (paste ipTM into REFS_RESULTS dict, runs OLS regression + soft-floor propagation through Tβ4 G→F-actin 10,000× scale + Phase 4j conservative re-run automatically). Pass criteria locked: family R² ≥ 0.5, h09 ipTM in [0.4, 0.75], predicted Kd_mono < 5 mM after G→F-actin scaling. Caveat baked in: AF3 ipTM is NOT a Kd predictor in general (avidity-and-dimers.md §3); within-family comparative ratios only. Doc: STRC h09 Phase 4m Kd Mono Soft Floor Pre-Registration 2026-04-26.
One-shot 13-job combined payload built.phase4k_4m_4b_v2_combined_builder.py merges Phase 4k (5) + Phase 4m (6) + Phase 4b_v2 deferred (2) into a single ~/STRC/models/af3_jobs_2026-04-27_h09_combined_13jobs.json. All jobs validate: ≤2000 aa each, ≤7 chains each (under AFS limits 5000 aa, 10 entities, 100 jobs per upload). Tomorrow at AFS quota refresh: one Upload-JSON click → 13 drafts → loop Step-1b promotion → 13 running jobs in ≤5 min of UI work.
Phase 4l infrastructure. OpenMM 8.5.1 installed in dedicated venv scripts/venv-md/. M5 Max platforms probed: Reference, CPU, OpenCL (Metal-backed) — no native Metal yet. OpenCL ~5× faster than CPU for production-size systems (~36k atoms RADA16 trimer + TIP3P water). Tool card filed at ~/STRC/tools/OpenMM.md per project tool-card rule. Phase 4l protocol sketched in OpenMM card; 100 ns NPT @ 298 K, 2 fs timestep → C_eff measurement for THIS scaffold. Compute budget: 12–48h on M5 Max OpenCL. Not started — gating decision pending; cheaper path (Phase 4k AF3) runs first.
AFS quota plan for tomorrow’s refresh. 13 jobs total budgeted (under 30 daily quota): 5 Phase 4k + 6 Phase 4m + 2 deferred Phase 4b_v2 (seed777_actin + triple-complex). Order of submission: Phase 4k first (highest leverage), then Phase 4m + Phase 4b_v2 deferred in same batch.
Ranking delta. Tier tentative-S held. mech 5 / deliv 4 / misha_fit 3 unchanged. Three independent gate-tightening paths now staged + pre-registered + tooled, any one closes Phase 4j conservative-FAIL scenario. Risk that reverses tentative-S: if Phase 4k tomorrow shows N < 3 achievable in any tandem-WH2 forced-proximity proxy → conservative degrades to Kd_eff > 1 mM → demote to A. Conditions for full S: (a) Phase 4k passes ANY of {N≥4 multi-copy job, tandem-WH2-x4 covalent} OR (b) Phase 4m predicts Kd_mono < 5 mM OR (c) Phase 4l measures C_eff > 100 mM. → STRC h09 Phase 4k Fibril Actin Geometry (note placeholder; results in 17h), STRC h09 Phase 4m Kd Mono Soft Floor Pre-Registration 2026-04-26.
Goal. Computational substitute for wet-lab Phase 2c gate (WH2 × F-actin Kd ≤ 50 µM) under no-wet-lab project rule. Propagate Husson 2010 Tβ4 × F-actin floor (5–10 mM) through Mammen-Jencks-Karpen polyvalent avidity formalism for assembled RADA16-tail91 fibril × F-actin filament. All inputs are primary-lit-backed (Husson 2010, Mammen 1998, Karpen 1998, Jencks 1981 — all paper notes already in vault).
Method.scripts/phase4j_avidity_kd_model.py (pure analytic; Mammen eqn 1.16: Kd_eff = Kd_mono · (Kd_mono / C_eff)^(N−1)). Grid Kd_mono ∈ {1, 5, 10, 20} mM × C_eff ∈ {13.4, 50, 100, 500} mM × N ∈ {2, 3, 4, 5, 6} = 80 cells. Pre-registered conservative (Kd_mono=10 mM, C_eff=50 mM, N=4) and nominal (Kd_mono=5 mM, C_eff=100 mM, N=5) scenarios.
Headline numbers.Conservative: Kd_eff = 80 µM (FAIL by 1.6× over 50 µM gate). Nominal: Kd_eff = 31 nM (PASS by 1500×). Grid: 47/80 cells PASS = 59% of cooperative-regime parameter space. Fail region is concentrated where (Kd_mono upper-bound 10–20 mM) AND (C_eff at flexible-linker floor 13.4 mM) AND (N ≤ 3).
Interpretation. Avidity rescue from Tβ4 mM-floor down to nM–µM range is real and quantifiable. RADA16 is rigid β-sheet → C_eff in upper Mammen range (~100–500 mM) is more physically apt than the 13.4 mM flexible-linker floor. Conservative-FAIL is a 1.6× margin within input noise; any single concession to the favorable side closes the gate.
Why this matters. The morning lit audit (literature/actin-kinetics.md) confirmed WH2 × F-actin Kd is structurally unmeasurable in any literature paper, with Tβ4 floor 5–10 mM as the only plausible analog. Single-WH2 MD-PMF would just confirm a number we already have; the actual h09 mechanism is avidity-driven, not single-site-driven. Phase 4j puts the avidity argument on quantitative footing.
Open work to close conservative case (no wet-lab):
Phase 4l — MD of free RADA16-WH2 fibril. OpenMM-Metal on M5 Max (apple-silicon-first per memory). 100 ns of fibril self-assembly → directly measure C_eff via WH2 angular distribution and scaffold rigidity. Replaces Mammen-bracketed range with this-scaffold-specific number.
Phase 4m — Comparative AF3 → Kd_mono soft floor. Use h09 WH2 × F-actin ipTM 0.51 (Phase 4b baseline) ratio against WAVE2 × G-actin Kd 52 nM → ipTM 0.6+ to bracket WH2 × F-actin Kd_mono ≥ 10 µM (rough but better than Tβ4 5 mM floor).
Ranking delta. Tier tentative-S held (cannot promote on conservative-FAIL; need 4k, 4l, or 4m to tighten). mech 5 / deliv 4 / misha_fit 3 — unchanged. Avidity framework is textbook-valid (Mammen 1998 §1.16 + Jencks 1981 + Karpen 1998 all in vault). Risk that reverses: Phase 4k shows N < 3 achievable → conservative degrades to Kd_eff 12 mM = catastrophic FAIL → demote tentative-S → A. Conditions for full S now formalized: Phase 4k passes (N ≥ 4 geometrically achievable in fibril × filament patch) OR Phase 4l measures C_eff > 100 mM for this scaffold OR Phase 4m softens Kd_mono floor to <5 mM. Any one closes the conservative scenario. → STRC h09 Phase 4j Avidity Kd Model 2026-04-26. Artifact: artifacts/phase4j_avidity_kd_2026-04-26.json.
Batch delivered. 9 of 11 staged tail91_v2 (Cys11/52→Ser) jobs submitted at 07:21 UTC, all check_circle by 07:46 UTC (~25 min wall-time). Quota burn 9→0. seed777_actin + triple-complex deferred to next AFS quota refresh (~midnight UTC). Submission used Step-1b draft-promotion flow per ~/.claude/skills/alphafold-server/SKILL.md; substring matcher bug (seed7 ⊂ seed777) caught and corrected mid-flow with (?!\d) lookahead.
TMEM145 axis (n=5). ipTM seed7=0.54 / seed13=0.55 / seed42=0.57 / seed99=0.56 / seed777=0.55. Median 0.55, σ 0.010, range 0.54–0.57. Gate cleared (median ≥0.50 ✅, σ <0.05 ✅). Δ vs v0: median −0.05 (0.60 → 0.55), σ tighter (0.037 → 0.010). Cys-scrub is NOT binding-neutral — modest interface cost is real and reproducible across seeds, not a stochastic outlier. Cys11/52 contributed ~10% of predicted ipTM signal.
Actin axis (n=4). ipTM seed7=0.49 / seed13=0.52 / seed42=0.51 / seed99=0.52. Median 0.515, σ 0.012, range 0.49–0.52. Gate cleared. Δ vs v0: median +0.005 (essentially unchanged), σ tighter (0.028 → 0.012). seed7 is the only sub-0.50 read but median over 4 seeds passes; full 5-seed read pending seed777_actin tomorrow.
Interpretation. tail91_v2 is the production-viable candidate (zero free Cys → no oxidation/disulfide-shuffling liability; SAP shelf-life and reconstitution chemistry safe). The 0.05 ipTM hit is the developability tax; predicted affinity decrease is sub-σ when normalized to scenario-band uncertainty. Actin axis insensitive to scrub — consistent with Phase 4b finding that tail (not WH2) is the actin-engaging surface, and Cys11/52 are not at the actin interface.
Triple-complex (cross-bind 0.32 problem from v0) untested for v2 in this batch — deferred. Hypothesis: scrub does not change cross-bind topology since both Cys are off-interface; expect v2 triple ≈ 0.32 ± noise.
Ranking delta. Tier tentative-S held. mech 5 / deliv 4 / misha_fit 3 / lit_audit partial — all unchanged. Promotion path to full S still gated on Phase 2c WH2 × F-actin Kd; per project rules wet-lab is off-limits, so the open computational path is MD-PMF (umbrella sampling on M5 Max via MPS-accelerated GROMACS or OpenMM-Metal). Risk that reverses tentative-S: if v2 triple-complex (next batch) drops below 0.30 or actin seed777 is < 0.49, would tighten σ floor and could trigger demotion review. Conditions for full S unchanged: WH2 × F-actin Kd-equivalent ≤ 50 µM established (computational MD-PMF), Phase 4 PKPD already PASS, Phase 4b 5-seed σ <0.05 ✅. → STRC h09 Phase 4b tail91_v2 Developability Scrub 2026-04-26 (note to be written). Artifact: ~/STRC/hypotheses/h09-hydrogel/artifacts/phase4b_tail91_v2_2026-04-26_results.json.
2026-04-26 (11:1x — Phase 4b AF3 robustness gate PASS on both axes; A→tentative-S promotion triggered)
Robustness verdict — both axes PASS. tail91 × TMEM145 median ipTM = 0.60 ± 0.037 across seeds {7, 13, 42, 99, 777}, range 0.51–0.61. tail91 × actin trimer = 0.51 ± 0.028, range 0.51–0.58. Gate (median ≥0.50 AND σ <0.05) cleared comfortably on TMEM145, exactly at threshold on actin (single high-outlier seed7 at 0.58 dominates σ; 4/5 seeds at 0.51 = AF3 attractor). Phase 3b’s 0.57 was the lower-tail of a 0.51–0.61 distribution, not a lucky seed.
Ablation reads. noWH2 × TMEM145 = 0.51 → tail is the real driver, WH2 contact-mass in single-partner Phase 4a was opportunistic hydrophobic artifact. noWH2 × actin = 0.51 → Phase 4a “tail also binds actin” was real, not single-partner artifact. WH2 is not the load-bearing actin handle in this construct — important for Phase 2c experimental design.
Triple complex (cross-binding test) = 0.32. Partners cross-bind under competition. Phase 4d bundling math still works at 1–10 µM peptide because RADA16 multivalency dominates, but Phase 2c bundling assay must control for off-target tail × actin engagement (e.g., titration with G-actin trap). Not catastrophic but a real complication.
Ranking delta: A → tentative-S triggered. mech 5 / deliv 4 / misha_fit 3 held. Promotion is tier-only, on robustness clearance. Conditions for full S: Phase 2c WH2 × F-actin Kd ≤ 50 µM (Gate 1, P0 wet-lab); Phase 2c bundling assay ≥ 100 crosslinks/µm at 1 µM peptide (Gate 2, P0-light); Phase 4 PKPD already PASS. Risks that reverse: WH2 × F-actin Kd > 100 µM → demote to B; or bundling fails under physiological [G-actin] competition → demote. A_promote_pending_2c_kd. → STRC h09 Phase 4b AF3 Robustness Gate 2026-04-26
tail91_v2 batch staged. Cys11→Ser, Cys52→Ser developability scrub (zero Cys, removes oxidation liability). 11 jobs at ~/STRC/models/af3_jobs_2026-04-27_h09_phase4b_tail91_v2/, combined upload at ~/.playwright-mcp/h09_phase4b_tail91_v2_batch.json. Submit when AFS quota refreshes (~midnight UTC). Expected: ipTM matches v0 within σ < 0.05 if scrub is binding-neutral; if it drops, scaffold is Cys-dependent (unlikely; Cys11/52 are not at the tail × TMEM145 interface).
Inventory updated. Parser phase4b_af3_robustness_parser.py already in scripts/, ready for ZIP downloads to ~/STRC/models/af3_results/h09_phase4b/. Outputs aggregate.{json,tsv}. Used JSON path today; ZIP path remains for full per-residue confidence + ranking_score downstream of paper figure prep.
2026-04-25 (late night — Phase 4 AF3 results parsed; jobs from 2026-04-24 finally read)
Phase 4 AF3 results for 6 hydrogel constructs (submitted 2026-04-24, completed same day, parsed today via Playwright MCP). Snapshot at ~/STRC/hypotheses/h09-hydrogel/artifacts/phase4_af3_2026-04-24_results.json. Top model ipTM/pTM:
⚠️ identical to tail84 — WH2 specificity NOT validated by AF3
tail91_noWH2 × TMEM145
0.51
0.49
TMEM145 binding partly WH2-independent
tail91 × TMEM145 × actin (full triplex)
0.32
0.45
🔴 triplex doesn’t form in one binding mode
tail71 × TMEM145 × actin
0.33
0.45
🔴 short tail weaker
Findings:
TMEM145 GOLD-pocket binding axis is robust; tail84 best length.
Actin binding marginal at AF3 level; WH2 motif specificity unconfirmed (noWH2 control = WH2 same ipTM).
Triplex hypothesis under question — single AF3 prediction cannot bridge tail-TMEM145-actin simultaneously. Two-stage model (tail-on-TMEM145 first, then WH2-on-actin) more consistent.
AF3 caveat on WH2 does not falsify Chereau 2005 ITC measurements (WASP 250 nM, WAVE2 52 nM, etc.) — those are at ITC level; AF3 trains on resolved structures and may not encode short-motif specificity for actin side-binding.
Ranking delta: tier A held | mech 4 → 4− (TMEM145 axis robust; WH2-actin axis caveated by AF3 noWH2-control symmetry; lit basis Chereau ITC unchanged) | deliv 3 held | misha_fit 3 held | next_step held (Phase 2c WH2 wet-lab Kd remains the load-bearing blocker; no PK/PD compute until measurement). Mech ”−” qualifier: Phase 4 AF3 shows a methodology limitation (AF3 can’t distinguish WH2 from noWH2 control), not a biology refutation. → STRC h09 Phase 4 AF3 Hydrogel-TMEM145-Actin Results 2026-04-25 (note to be written).
2026-04-25 (night — Zhu 2019 eLife re-ingest from MinerU)
literature: Zhu 2019 eLife (krey2019.pdf, single-cell proteomics of E15 chick utricle hair cells) re-ingested from MinerU PDF parse. Existing paper note 2026-04-23-krey-2019-stereocilia-proteomics-elife author list FIXED (was hallucinated as “Krey JF, Chatterjee P, Halford J…”; correct: Zhu Y, Scheibinger M, Ellwanger DC, Krey JF, Choi D, Kelly RT, Heller S, Barr-Gillespie PG). Added verbatim TMSB4X vs ACTG1 riBAQ table (Fig 2C) and Table 1 95% CI confidence intervals. Three new atomic notes filed: P0 Hair Cell Biophysical Reference Table, P0 TMSB4X Actin Buffering Stoichiometry Hair vs Supporting Cells, P1 Recipe — TMSB4X-Buffered Free G-actin Pool Estimation, P2 TMSB4X Drop During Hair Cell Differentiation. Headline finding: hair-cell TMSB4X riBAQ = 0.006 ± 0.002 vs supporting-cell 0.056 ± 0.012 (9.3× drop, p<0.001) — endogenous β-thymosin-class buffer concentration in the h09 target compartment is ~3.5 µM, sets the floor for synthetic WH2-peptide effective concentration. STRC absence in dataset re-confirmed (chick utricle ≠ mammalian OHC; Table 3 + MaxQuant output checked). No tier/score change — Phase 2c WH2 × F-actin Kd wet-lab gate unchanged. → 2026-04-23-krey-2019-stereocilia-proteomics-elife
2026-04-25 (night — Pollard 2016 actin review re-ingest from MinerU)
Ranking delta A → B. Post-batch-2 review: WH2 × F-actin side-binding Kd confirmed NOT MEASURED in literature; structural/assembly pieces solid (Chereau 2005 ITC resolved, STRC copy# absence documented). A-tier requires a measured Kd for the side-binding gate; demoted until Phase 2c wet-lab delivers. AF3 triple-complex and ipTM work continues. Trigger: vault-wide audit review post 2026-04-25 batch — Lit Audit Status 2026-04-25.
2026-04-25
literature — Erratum filed for foundational SAP review: Gelain, Luo, Zhang (Chem. Rev. 2021, 121, 5093 — corrects misspelled author name in 2020 review, DOI 10.1021/acs.chemrev.0c00690). The 2020 review itself is not yet parsed into the vault despite being the most comprehensive published EAK16/RADA16 hydrogel reference. Action: retrieve + MinerU the corrected DOI to populate papers/2020-gelain-eak16-rada16-chemrev.md with force-field / gelation / mechanical numbers. → 2021-gelain-luo-zhang-eak16-rada16-correction
Lit-audit SI follow-up round 3: Chereau 2005 SI Table 2 RESOLVED via Fig. 1C image in main text — exact ITC Kd values: WASP (430-458) 250 nM, WAVE2 (433-464) 52 nM, WIP (29-60) 160 nM, MIM (724-755) 230 nM; model 200 nM is WASP-proximal and defensible; 5 Anna’s Archive + PMC + PNAS + Sci-Hub queries tried, all blocked (SI inaccessible from non-browser). Krey 2019 eLife STRC absence CONFIRMED — all eLife supplementary materials checked (MaxQuant output, heatmaps, scRNA-seq gene list 186 genes): STRC absent from chick utricle dataset; STRC_NORMAL_OHC_M = 1 µM documented as in-house estimate. lit_audit_blockers 3 → 1 (WH2 × F-actin remains only hard blocker). A held. → STRC h09 Parameter Provenance Audit 2026-04-25
Lit-audit gap closure round 2: Salt 2001 full text retrieved via Anna’s Archive (SciDB mirror) + MinerU parsed — confirmed no MW-scaling table in paper (Stokes-Einstein K_RWM = 0.003/h is paper-endorsed approach); Chereau 2005 main text retrieved via Anna’s Archive — main-text ratios yield WAVE ~100-200 nM / WASP ~500 nM-1 µM; SI Table 2 not separately indexed. 2 cochlear-pkpd ⚠ → ✅. Blocker count 4 → 3. lit_audit: partial, 3 real blockers remain. A held. → STRC h09 Parameter Provenance Audit 2026-04-25
Lit-audit gap closure: WH2 × F-actin Kd confirmed NOT MEASURED (exhaustive PubMed + Dominguez 2016 + Chereau 2005 + Husson 2010 search); Chereau 2005 SI Table 2 still inaccessible (PNAS 403); Salt 2001 still paywalled (abstract confirms values); Cartagena-Rivera 2019 author error in paper note corrected; per-link stiffness derivation documented as external calc. lit_audit: partial, 4 blockers propagated to hub frontmatter. A held. → STRC h09 Parameter Provenance Audit 2026-04-25