AUDIT-h09

AUDIT — h09-hydrogel (Synthetic Peptide Hydrogel HTC)

• Tier: S · mech 5 / deliv 4 / misha_fit 3
• lit_audit before: partial (2026-04-25), 1 blocker listed, 2 items in lit_audit_resolved
• lit_audit after: not-modelable, 4 blockers enumerated
• Grounding: GROUNDING/hypc-h09-hydrogel.md
• Why this hypothesis went first: S-tier + partial is the most dangerous combination in the vault — the highest-ranked hypothesis carrying an admitted hole in its numbers.

Prior work read

• hypotheses/h09-hydrogel/index.md (hub), log.md
• phases/STRC h09 Parameter Provenance Audit 2026-04-23.md (Round 1)
• phases/STRC h09 Parameter Provenance Audit 2026-04-25.md (Rounds 1–3, 19.3K)
• Scripts: phase4j_avidity_kd_model.py, phase4m_iptm_kd_analysis.py, phase4n_avidity_montecarlo.py, phase4k_af3_contact_parser.py, phase4b_af3_robustness_parser.py

The 2026-04-25 audit is genuinely good work — three rounds, real retrieval, two blockers honestly closed. This run does not redo it. It does three things the prior rounds could not: re-checks the one open absence against 2026-08 literature, audits the constants introduced after 2026-04-25 (Phases 4j/4m/4n all landed 04-26/04-27, i.e. after that audit closed), and forces the outcome into the three-state scheme instead of leaving it in an open-ended partial.

Parameter-provenance table

Legend — ✅ sourced: value is grep-verifiable in a sources/lit/ file with canonical type: lit frontmatter. 🔶 analog/derived: honest bridge from a sourced value, derivation stated, NOT a measurement. ❌ not measured: searched and absent from the primary literature. ⚙ internal: a gate or a numerical setting, not a physical constant — needs no literature.

#	Constant	Value	Units	Where used	Claimed source	Local sources/lit/ file	Status
1	WH2(WASP) × G-actin Kd	250 ± 30	nM	4m refs	Chereau 2005 PNAS Fig. 1C (ITC)	2026-04-23-chereau-wh2-actin-pnas.md:61	✅
2	WH2(WAVE2) × G-actin Kd	52 ± 3	nM	4m refs	same, Fig. 1C	…chereau….md:62	✅
3	WIP × G-actin Kd	160	nM	4m refs	same, Fig. 1C	…chereau….md	✅
4	MIM × G-actin Kd	230	nM	4m refs	same, Fig. 1C	…chereau….md	✅
5	Tβ4 × G-actin Kd	760 ± 120	nM	4m refs, G→F ratio	same, Fig. 1C	…chereau….md	✅
6	WH2_KD_GACTIN_M (model)	200	nM	legacy 4d/4i	mid-range WIP(160)–WASP(250)	…chereau….md:71 documents the choice	✅ (choice, bracketed by measured values)
7	Tβ4 × F-actin Kd	5–10	mM	4j KD_MONO floor	Husson 2010	2026-04-23-husson-wh2-multifunctionality.md:49	✅ as Tβ4; 🔶 the moment it is read as WH2
8	WH2 × F-actin side-binding Kd	—	—	h09 single-molecule premise	none exists	—	❌ CONFIRMED NOT MEASURED
9	G_TO_F_ACTIN_SCALE	10 000	×	phase4m…py:44	derived: (5–10 mM)/760 nM = 6 580–13 160×, mid 10 000×	inputs ✅ (#5, #7)	🔶 cross-family analog bridge
10	C_EFF floor	13.4	mM	4j, 4n priors	Kramer & Karpen 1998, 3400 Da PEG, r = 39 Å	1998-kramer-karpen-polymer-ligand-dimers-nature.md:61	✅
11	C_EFF rigid range	50–500	mM	4j, 4n priors	Mammen 1998 rigid-linker bounds	1998-mammen-polyvalent-interactions-angew.md	✅
12	Avidity formalism	Kd_eff = Kd_mono·(Kd_mono/C_eff)^(N−1)	—	4j, 4m, 4n	Mammen 1998 eqn 1.16	1998-mammen….md:38	✅
13	ENTROPIC_HAIRCUT_X	31.6	×	phase4n…py:44	√(10·100), from Mammen’s 10–100× real-vs-ideal shortfall	1998-mammen….md:68	✅ (geometric mean of a sourced range; derivation in-script)
14	RT_KCAL_298K	0.5925	kcal/mol	4j	R = 1.9872 cal/mol/K × 298 K	physical constant	⚙
15	RWM permeability (TMPA)	1.9 × 10⁻⁸	cm/s	PKPD	Salt & Ma 2001	2001-salt-ma-quantification-rwm-permeability.md:44	✅
16	ST clearance t½	60	min	PKPD	Salt & Ma 2001	…salt….md:45	✅ (guinea pig only — species caveat recorded)
17	K_RWM for 14 kDa	0.003	/h	PKPD	Stokes–Einstein extrapolation from #15	…salt….md:61–62 shows the calc	🔶 (paper endorses the method, supplies no table)
18	OHC bundle stiffness, +HTC	5.12 ± 0.46	pN/nm	mechanics	Cartagena-Rivera 2019	dulon-2019-htc-bundle-mechanics.md:39	✅
19	OHC bundle stiffness, −HTC	2.05 ± 0.15	pN/nm	mechanics	same	…dulon….md:40	✅
20	HTC aggregate contribution	3.07	pN/nm	mechanics	5.12 − 2.05, in-note	…dulon….md:49	✅
21	HTC links per bundle	~19	count	per-link derivation	not in the cited paper	—	🔶 architectural assumption (N_stereo − 1)
22	HTC per-link stiffness	0.16	pN/nm/link	4d bundling	3.07/19, external calc	—	🔶 derived from #20 ✅ and #21 🔶
23	STRC_NORMAL_OHC_M	1	µM	competition model	in-house placeholder	—	❌ CONFIRMED NOT QUANTIFIED
24	GATE_KD_M	50	µM	4j/4n pass gate	h09 Phase 2c S-promotion gate (internal)	—	⚙
25	GATE_IPTM_MEDIAN / _SIGMA	0.50 / 0.05	—	4b	pre-registered internal gate	—	⚙
26	CONTACT_CUTOFF_A, CHAIN_PAIR_IPTM_MIN	5.0 Å, 0.20	—	4k parser	method settings	—	⚙
27	ACTIN_LENGTH	375	aa	4k parser	actin sequence length	—	⚙ (sequence fact)
28	N_SAMPLES, RNG_SEED	100 000, 20260426	—	4n	reproducibility settings	—	⚙

Totals: 14 ✅ sourced · 6 🔶 derived/analog · 2 ❌ not measured · 8 ⚙ internal.

Retrieval performed this run

target	tool	exact query	outcome
WH2 × F-actin Kd	rag search	"WH2 domain binds the side of actin filament dissociation constant"	top-3 all Chereau 2005 structural passages; no Kd
WH2 × F-actin Kd	WebSearch	WH2 domain F-actin filament side binding dissociation constant Kd measured cosedimentation	7 results; only G-actin Kds (~50 nM VgrG-1, 1.5 µM ciboulot); no F-actin Kd

Both persisted verbatim in GROUNDING/hypc-h09-hydrogel.md §G1. Combined with the 7 distinct searches logged in the 2026-04-25 audit, cumulative attempts = 9, well past the 5-attempt stop rule. The finding is the absence.

Substantive correction to the vault’s reasoning

The 2026-04-25 audit justified the absence structurally: the WH2 groove is “partially occupied by the next monomer in the long-pitch helix,” so no coherent side-binding surface exists. Chereau 2005 says the opposite. Verbatim (§G1a):

“the actin-binding interface of these WH2 domains does not interfere with intersubunit contacts in F-actin and can thereby coexist with the actin filament”

So the missing Kd is not explained by steric impossibility. It is missing because nobody ran the titration. This is a weaker basis for the absence — and it cuts both ways for h09: the mechanism is not structurally excluded (mildly good news), but the vault can no longer lean on “physics forbids it, so the gap is harmless” (bad news for the confidence attached to the gap). Recorded rather than smoothed over.

Why not-modelable and not another eternal partial

partial carries the implication “keep searching, it will close.” After 9 distinct retrieval attempts across two independent audit campaigns four months apart, that implication is false and is now actively misleading — it invites a future agent to spend another cycle re-searching a settled absence. Every one of h09’s remaining blockers is of the not-measured-anywhere kind, not the not-yet-retrieved kind:

• #8 WH2 × F-actin Kd — no primary measurement exists (9 searches).
• #23 STRC per-OHC concentration — no proteomics study quantifies it (Krey 2019 eLife SI exhaustively checked; STRC absent from every table).
• #9 G→F scaling — cannot be sourced because #8 does not exist; it is a cross-family transposition from Tβ4, and Chereau’s text (above) actively undercuts the assumption that the G→F penalty transfers between domain families.
• 22 HTC link count and per-link stiffness — the source paper reports no link enumeration; the number is architectural, and no paper supplies it.

not-modelable states the true situation: these values are unobtainable from the primary literature and only wet-lab work can produce them. That is a directive, not a defeat. It points at Phase 2c instead of at another search.

Note also that the prior round parked #23 in lit_audit_resolved as “blocker closed as unresolvable.” Under the three-state scheme that is precisely a not-modelable blocker, not a resolved one. Calling an unobtainable value “resolved” is the exact bookkeeping that lets a hole quietly stop counting. It is moved back into the blocker list.

What is NOT wrong with h09

Stated plainly so the verdict is not misread as a downgrade of the science:

• 14 constants are ✅ sourced and grep-verified this run (evidence in verify.txt).
• The avidity chain (Mammen/Jencks/Karpen) is fully sourced and correctly cited.
• The model does not assert a fabricated WH2 × F-actin Kd. Phase 4j uses the Tβ4 analog explicitly labelled "closest measured analog", and Phase 4m replaced the scenario value with a calibrated prediction (Kd_F = 1.48 mM, family fit R² = 0.690) rather than an invented number. That is the right way to handle an unmeasurable parameter.
• lit_audit describes literature provenance, not hypothesis quality. A not-modelable verdict on an S-tier hypothesis is coherent: the science routes around a gap that the literature will never fill.

The honest residual risk is #9: the gate-closing Phase 4m number inherits a cross-family analog bridge. A hostile reviewer will attack there first.

Verdict

lit_audit: not-modelable · lit_audit_date: 2026-08-21 · 4 blockers

lit_audit: not-modelable
lit_audit_date: 2026-08-21
lit_audit_blockers:
  - "WH2 x F-actin side-binding Kd — NOT MEASURED anywhere in the primary literature. Re-confirmed 2026-08-21 (rag search + WebSearch); 9 distinct attempts across 2026-04-24 and 2026-08-21. Only Phase 2c wet-lab cosedimentation/bundling can produce it. Note: the prior structural justification for the absence is WRONG — Chereau 2005 states the short WH2 interface does NOT interfere with F-actin intersubunit contacts and can coexist with the filament, so the value is unmeasured, not sterically impossible."
  - "STRC_NORMAL_OHC_M = 1 uM — NOT QUANTIFIED in any proteomics study. Krey 2019 eLife (PMC6855842) all SI checked: MaxQuant output, expression heatmaps, 186-gene scRNA-seq list, antibody table — STRC absent from all; chick utricle dataset, wrong tissue for a mammalian-cochlear HTC protein. In-house placeholder. Moved out of lit_audit_resolved 2026-08-21: an unobtainable value is a not-modelable blocker, not a closed one."
  - "G_TO_F_ACTIN_SCALE = 10000x (phase4m_iptm_kd_analysis.py:44) — cross-family analog bridge, not a sourced constant. Derived from Tb4 G-actin 760 nM (Chereau Fig. 1C) over Tb4 F-actin 5-10 mM (Husson 2010) = 6580-13160x, mid 10000x, then transposed onto WH2. Assumes the G-to-F affinity penalty is a property of the actin surface rather than of the domain; Chereau 2005 undercuts that by stating Tb4 and WH2 engage the filament differently. Load-bearing for the Phase 4m Kd_F = 1.48 mM gate-closing result."
  - "HTC links per bundle (~19) and per-link stiffness (0.16 pN/nm/link) — external architectural derivation, not measured. Cartagena-Rivera 2019 (PMC6382404; previously misattributed in-vault as 'Dulon 2019') reports the aggregate 3.07 pN/nm HTC contribution but no link count and no per-link value. N_links = N_stereo - 1 for a 20-stereocilium OHC is an assumption; no paper supplies the enumeration."

lit_audit_resolved is retained on the hub for the two genuinely-closed items (Chereau SI Table 2 via Fig. 1C; Salt 2001 full text), minus the STRC_NORMAL_OHC_M entry, which moves to the blocker list.

Ranking delta

S-tier HELD. mech / deliv / misha_fit — see RANKING-DELTA.md for the recommendation and reasoning; the chief makes the final call.