What we found, strongest first

A proposed gja5a link to fin length

Found by: our own check (the cell-cell subagent)

An earlier note from this thread offered Daane 2021's `gja5a` convergent substitution as a second link between the flying-fish genomics and the bioelectric fin-size story. The cell-cell subagent checked it and the link does not hold: the fin-length connexin in zebrafish is `gja1b` (cx43, the `short fin` locus), `gja5a` has no ZFIN fin expression or phenotype record, and Daane 2021's own text puts cx43 under *constrained*, not accelerated, evolution along the flying-fish branch. The `gja5a` substitution stays in the map's `fin_size` block as the paper's own data, and it is kept out of any causal chain. The real connexin tie to the potassium-channel story is `gja1b` with Seaver 2025, which is independent of Daane.

genotype/nanopubs/DECISIONS.md · crosscheck-frontier/crosscheck.json (row G4)

"Almost no molecular data exists for flying fish"

Found by: the project lead, reading a primary source the thread had not read

An annotated genome does exist: Xu et al. 2021 assembled 23,611 predicted protein-coding genes for Hirundichthys speculiger (mirrorwing flyingfish), of which 23,492 (99.5%) carry at least one database hit. What flying fish lack is expression and phenotype data, not genes.

And the caveat that keeps it honest: at least one sequence-similarity or domain hit in one of four databases, and nothing more. TrEMBL alone accounts for 23,477 of the hits, and TrEMBL is automatically generated and unreviewed, so 99.5% is a measure of database coverage rather than of understood function.

genotype/map.vetted.slim.json (flying_fish_genome) · genotype/nanopubs/unsigned/correction-flying-fish-genome-exists.trig

"No ray-to-inter-ray signalling arrow could be sourced in any fin of any fish"

Found by: our own nanopublication curation pass, not the thread that wrote it

Ray-to-inter-ray signalling is established in the zebrafish caudal fin by grafting and niche experiments. One query returned three papers that contradict the universal sentence. What survives is the far narrower claim that no fetched source names a ligand in one compartment and its receptor in the other.

And the caveat that keeps it honest: The cause is a process failure, not only a fact error: a title-only read was allowed to support a universal negative, and one of the three papers was already cited by the same thread.

genotype/nanopubs/unsigned/correction-ray-interray-null-narrowed.trig · nanopubs-index/README.md (finding 1)

"The TFF’s wings (aspect ratio 17.9) are more slender than any measured flying fish (3.0-16.9)"

Found by: our own parametric-model fix pass, reading Fish (1990) and Park & Choi (2010) in full

The two numbers used different definitions. 17.9 is span squared over the area of both pectoral fins (Park & Choi S^2/A1); Fish (1990) divides by pectoral plus pelvic fins plus the body between them. On Fish’s definition the TFF is at most 13.9, inside his range. What survives: 17.9 is 1.7 times the one measured specimen (10.68).

And the caveat that keeps it honest: The same pass found the adult flying-fish mass differs 2.2x between the one wind-tunnel specimen (0.060 kg) and the population length-weight fits (0.100-0.132 kg). The 9.9 m/s minimum flight speed holds only for the specimen; at population mass it is 12.7-14.6 m/s, still inside the quoted 10-20 m/s.

anatomy-model/parametric/CHANGELOG.md · physics/flier/tff/SPEC-MISMATCHES.md (item 1) · physics/results.json (overall.sentence)

"Sustained aerobic powered flight is 29 to 38 times short; no teleost the measured record allows has an aerobic ceiling above 0.0458 W" (E3)

Found by: the studio engine, when a real zebrafish broke our own aerobic bound

Run on an unedited adult zebrafish, the studio flagged the fish red for swimming at speeds it was measured to sustain. The Fick ceiling (efficiency × oxycaloric × cardiac output × arterial O2) used a cardiac output borrowed from schoolmaster snapper, 58.48 ml/min/kg. A measured zebrafish at U_crit takes up 2630 mg O2/kg/h (Palstra et al. 2010, PLoS One 5:e14483, PMC3013094), 2.8 times the 950 the ceiling allowed; even its standard metabolic rate is 1.5 times over. The same proxies set the flying-fish numbers. With a zebrafish-derived O2 delivery (1600 mg/kg/h scaled to 60 g) and a skeletal-muscle efficiency (0.23, Gladman & Askew 2024) the envelope becomes 0.0833 W, and the flying-fish shortfall is about 10 to 30 times (central about 20), not 29 to 38. E3 is now version 2.0.0, scoped to its sources; version 1.0.0 is kept. What survives: sustained aerobic flight is still ruled out by an order of magnitude.

And the caveat that keeps it honest: No flying-fish respirometry exists in the project, so the correction is a cross-taxon transfer like the original proxies. No zebrafish cardiac output during exercise was found; which of cardiac output or blood O2 content is low is inferred through Fick. Every case keeps propulsive efficiency at 1 and the best L/D, so the real shortfall is larger than each figure.

physics/FICK-CEILING-CHECK.md · lean/registry/laws/E3.json · lean/registry/law-versions/E3@1.0.0.json · lean/empirical.json (E3) · sim-engine/scenarios/current.mjs

"The papers disagree on sign" about the membrane-potential change caused by gain-of-function kcnk5b

Found by: trying to FORMALISE the dispute in Lean: writing both arms as named hypotheses forced a check of which paper asserted which sign, and the attribution did not survive it

Jiang 2024 proposes a RECONCILIATION of the two readings, not a contradiction of Perathoner 2014: the increased K+-channel activity first hyperpolarizes the membrane potential, and this increase immediately activates Na+ channels to depolarize the cells. Our own stored evidence held that sentence all along — it is in genotype/causal-chains.json with full-text provenance — but the prose summary built from it dropped the third position and recast a reconciliation as a dispute, putting a disagreement in a cited author’s mouth.

And the caveat that keeps it honest: The scientific gap is real and unchanged: no steady-state membrane potential of a wild-type versus alf zebrafish fin-bud cell exists in any source we have fetched, so the direction in the tissue that matters is unresolved. What changed is that it is now a three-arm open question rather than a two-sided dispute, and the two published readings are provably NOT formally contradictory — they conflict only under an added same-preparation premise, carried explicitly as a Lean hypothesis.

genotype/CAUSAL.md (kcnk5b row; unresolved-questions list) · lean/registry/open-questions/questions/oq-kcnk5b-vm-sign.json · lean/registry/open-questions/lean/KcnkSign.lean

The real flying fish also fails

E3 (empirical, Delphi.Empirical.no_aerobic_flight_above_ceiling): level flight at 10 m/s needs 1.34645 W against an aerobic ceiling of 0.0458 W — short by a factor of 29.4. the REAL flying fish fails this row: level flight at 10 m/s needs 29 times its aerobic ceiling. This is the known and expected result (flying fishes glide, they do not fly), and it is exactly why a design that "passes" a row the real animal fails must be treated as suspicious rather than as better.

So failing does not separate the design from its target. the real flying fish ALSO FAILS the gate (E3, by a factor of 29), so "FAILS" does not separate the design from the target

And the gate is weak: 3 of 21 rows are decided

THE GATE IS WEAK. All 3 supplied inputs move a verdict, but only 3 of 21 rows are decided, and the two body inputs move the same single pair of rows (C7b and E2 are one fact). Lean is testing the design on body size, and on whether one amino-acid change is reachable in one nucleotide substitution. It is testing nothing aerodynamic, nothing developmental and nothing about the fins, which is the entire point of the design.

The worked example: one genome says it in one sentence

One species states the annotation-versus-parameter gap better than any aggregate: a genome with essentially complete functional annotation and, in the paper that reports it, none of the eight parameters.

Hirundichthys speculiger, the mirrorwing flyingfish: 23,611 predicted protein-coding genes, 23,492 (99.5%) annotated. Read: full text and both supplementary archives, fetched and parsed by this script.

And in that same paper, counted term by term: all 25 parameter-bearing terms occur 0 times in the full text — rna-seq, half-life, kcat, diffusion, hill coefficient, molecules per cell and the rest. The verification is a count, not an inference.

'Functionally annotated' here means at least one hit in at least one of four databases. TrEMBL alone supplies 23,477 of the 23,492, and a TrEMBL hit is a sequence-similarity match to an unreviewed entry, not a measured function. Annotation by homology is the weakest sense of the word.

simulation-readiness/readiness.json (worked_example)

The precedent, to size the task

The closest thing that exists to what is being asked: a model that predicts phenotype from genotype for every annotated gene of an organism.

• 525 genes, in 28 sub-models — the life cycle of a SINGLE CELL; no anatomy, no tissues, no development
• built from over 900 primary sources and more than 1,900 experimentally observed parameters
• that is 3.62 parameters and 1.71 publications per gene

Not a prediction, just the same parameter density applied to a larger gene set, to size the task. Development would need more, not fewer, parameters per gene: a zygote-to-adult model adds cell states, positions, divisions and movements that a single bacterial cell does not have.

simulation-readiness/readiness.json (precedent_m_genitalium)

The strongest corroboration

Our most load-bearing caveat holds from outside. Transfer of the lat4a x kcnh2a result to flying fishes is HYPOTHESIS ONLY: no flying-fish gene has ever been manipulated, the two conceivable routes (reverting lat4a F326L; reverting the kcnh2a regulatory state) are untested in any organism, and the one double mutant that changes fin proportion redistributes size between fins rather than scaling one fin up.

This is the strongest independent corroboration in the whole cross-check. Four separate records, from two different issuers and three different retrieval routes (Europe PMC abstract, NOAA full-text PDF, a local repository source register), all keep the transfer at hypothesis level; one of them types it that way in the schema itself (payload.origin = 'hypothesis'), and the Daane documentary registration calls it in terms 'an unestablished transfer hypothesis'. We searched their graph for any record asserting the transfer as established and found none. Their Problem also names the specific untried next step we do not have: search the 53 exocoetid SRA records for paired-fin lat4a/kcnh2a expression.

crosscheck-frontier/crosscheck.json (row G3) · crosscheck-frontier/REPORT.md (section 1.1)

The one disagreement

internally inconsistent: map.json says settled, HANDOVER says unproven. A genuine conflict on the same question from overlapping sources, and they are probably right. Both sides consulted ZFIN and NCBI Gene; they report 'lat4a' as neither a ZFIN symbol nor an alias, with 0 hits in NCBI Gene and UniProt, and they name slc43a2b as a second candidate our map never considers. Our 'settled' rests on an indirect chain (a human alias plus a ZFIN curator's attachment of the paper's phenotypes to slc43a2a), which is an inference, not an identity. Our own HANDOVER.md already flags the overstatement, so the real disagreement is between two of our files, with the collaborator on the side of our HANDOVER.

Our files: genotype/map.json; genotype/README.md; genotype/HANDOVER.md

crosscheck-frontier/crosscheck.json (row G5) · crosscheck-frontier/REPORT.md (section 2.1)

Their Lean check shares our Mathlib pin

Our lean/lake-manifest.json pins mathlib at 56a005855ea17e0b…. Their receipt declares 56a005855ea17e0b…. These are the same revision, compared in this build. So two apparently independent Lean checks share a trusted base.

Worse, their receipt bytes cannot be retrieved: their catalogue reports 0 artifacts declared downloadable, 0 hosted and byte-checked, and 0 formal receipt contents read. If their check is cited anywhere, it must be cited as a declaration we could not verify.

lean/lake-manifest.json · crosscheck-frontier/crosscheck.json (row L2) · crosscheck-frontier/REPORT.md (section 2.3)

Much of their store is a bulk import of our registry

A large part of namespace node-a is a bulk import of DELPHI's OWN nanopublication registry: claim ids thompson1917-*, darwin1859-*, wolfram2002-*, levin-*, hamilton-*, and Evidence ids delphi-claim-test--*. They carry our claim IRIs under w3id.org/delphi/resource/claim/, our nanopub paths, our extractor agent id, our 'verification: lean' flags, and sources at github.com/fractastical/delphisci read 'through the keyless delphisci MCP'. A match on any of those ids is our own work reflected back and is never independent corroboration.

Of 1,287 claim versions in their store, 1,248 are in the node-a namespace and 39 in delphi-frontier; 1,214 declare no origin. Nothing in their store is accepted: every record examined carries imported_unreviewed or proposed.

crosscheck-frontier/crosscheck.json (reading_rules, their_store_as_measured)

Drafts by thread

nanopubs-index/index.json

Validation result

Checker: nanopub-java 1.95.0, one run per file. The project's recorded trap is real and was handled: np check exits 0 on an RDF parse error, so the verdict is parsed out of the transcript rather than taken from the exit code.

Each of those is expected and by recorded decision: the records are unsigned drafts using purl.org/nanopub/temp/ placeholders, which are replaced by trusty URIs at signing.

nanopubs-index/index.json (drafts[].validation) · nanopubs-index/README.md (Validation)

What the cross-check says should change in what we claim

• Fix `genotype/map.json:1604`.
• State the shared Lean toolchain if their check is ever cited.
• Never count the `thompson1917-*` / `darwin1859-*` / `wolfram2002-*` / `levin-*` / `delphi-claim-test--*` records as external agreement.
• Nothing in the cross-check weakens the lat4a x kcnh2a caveat, the gja5a retraction, the aerobic-power elimination or any of the six refutations.
• Two optional strengthenings, if the lead follows them up:
• Add the Saro-Cortes stability-lift trade-off to the flier constraints

crosscheck-frontier/REPORT.md (section 5)

Findings that exist only as prose, with no record

List A: dashboard claims no nanopublication draft records. List B: drafts citing no external source.

• A1. Two whole threads have zero records
• A2. The sim-engine pre-registered test that FAILED
• A3. Genotype headline numbers with no record
• A4. Lean results with no record
• A5. Two of sixteen claim tests have no evidence record
• A6. Counting mismatches a reader will trip on
• B. Drafts with no external source — 15 of 157 drafts

nanopubs-index/gaps.md

Limits of the simulation-readiness measurement

• It measures databases, not the literature.
• SIGNOR was unreachable, not empty.
• PaxDb was sampled, not exhausted.
• BRENDA's licensed bulk data was not obtained.
• The 4-per-gene convention is a choice.
• None of this is flying-fish data

simulation-readiness/README.md (Honest limits of this measurement)

Four mismatches found by building the TFF spec as a mesh

Found by building the spec as a mesh, and logged rather than fixed. Among them: only 29.5% of the required flight muscle fits inside the body. That is a property of the specification, not of an animal.

• Two aspect ratios share one name (the substantive one)
• Rounding
• Two incompatible readings; no mesh satisfies both
• The flight muscle does not fit

physics/flier/tff/SPEC-MISMATCHES.md

Deliberately not done in the design validation

• Nothing was digitised from a figure.
• No zebrafish pectoral fin ratio was invented
• No bird data.
• No `measured` value in any source file was altered
• No personal identifier

design-validation/RUNS.md (Things that were deliberately not done)

Ideas not yet checked against prior art

Every idea or choice adopted in this project, with its origin. researched = survived a logged prior-art search; reach-synthesis = synthesised from other sources, not yet checked against prior art; serendipitous = arose by accident. Seeded 2026-10-31 by the hub; most entries are reach-synthesis because the research rule was adopted after they were made. See AGENTS.md, 'Standing order: every new idea is researched'.

36 of 41 adopted ideas have not survived a logged prior-art search.

decisions/ideas.json