Close the verification debt: 189 findings fixed, and three boundary breaches my own scan missed
Three documents rewritten against every outstanding adversarial finding. 189 fixed, 15
declined with a stated reason — a finding that could not be verified either way is declined
as such, never silently dropped.
THE BUSINESS CASE NARROWED TWO CLAIMS RATHER THAN DEFENDING THEM, and both are improvements:
Its labelling rule said every first-party figure carries a maturity label. It did not.
Narrowed to the scope that holds — every first-party QUANTITY in the substantive sections
— with operational counts stated by measurement instant instead, because a maturity label
on "this file is 22,712 bytes" says nothing a timestamp does not say better.
Its thesis said a disagreement about a float quantity cannot be closed by re-derivation.
The same vendor documents that establish the seam also bound it: NVIDIA scopes determinism
to one platform-and-version pair, and MathWorks documents bit-wise agreement at zero
tolerance on fixed-point types. So THE SEAM IS PORTABILITY, not the absence of determinism
everywhere. Stated at that size now, which is both true and harder to attack.
THREE PRIVATE IDENTIFIERS CROSSED, AND MY BOUNDARY SCAN SAID CLEAN. I measured the repaired
drafts with the path patterns from the earlier split — cells/, Sources/, .gaiaftcl, /Users/,
file:line — and reported zero. An adversarial lens then found a private fleet-runner name, a
private artifact name and two internal record fields. They are IDENTIFIERS, not paths, and
the scan was scoped to paths.
That is the third time this session a pattern scoped to where I last saw a defect missed it
where it actually was, and the second time on this exact boundary. The harness now carries
check 4c for identifiers, so the next occurrence fails a test instead of passing a grep.
TWO CITED ARTIFACTS DID NOT EXIST: reproduce/reentry-ledger-census.swift (404 — no such
program) and corpus/satcat.tsv (the file is at corpus/study-29/satcat.tsv). A page citing a
program that is not there is a broken reproduction instruction, which is the defect this
repository was restructured to fix. Harness check 4b now verifies every cited reproduce/ and
corpus/ path resolves.
validate.sh: 25 passed, 0 failed, from a fresh public clone.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Front-and-centre: the business case on the wiki, and a sidebar that opens what it says
THE SIDEBAR SAID ONE THING AND OPENED ANOTHER. "Study 29 — Continuous-model shear" pointed
at an address reading Study-29-Affine-Earth-Sovereign-Substrate. I retitled that page when
29 and 30 were split and deliberately kept the old filename "so inbound cross-links
survive". That was the wrong trade: it bought five internal links and cost every reader a
label that disagrees with the page they land on.
Renamed to Study-29-Continuous-Model-Shear, all five inbound references repointed, and a
redirect stub left at the old name so external links still land. The stub says why.
ONE SIDEBAR TARGET WAS DEAD AND ALWAYS HAD BEEN. "QMT manifold repair v5.0" pointed at
QMT-Court-Repair-2026-08-30, which has ZERO entries in the history of this repository — a
page that was never written, advertised at the top of the Paradigm section. Removed rather
than invented: writing a page to satisfy a sidebar entry is backwards. All 63 remaining
sidebar targets resolve.
THE BUSINESS CASE IS NOW A WIKI PAGE, not only a private artifact: The-Exactness-Seam.md,
verified carrying zero private references. Home leads with it, with Study 29 and Study 30
beside it, and with the one instruction that matters — clone the repository and run
reproduce/validate.sh. The same three sit at the top of the sidebar under "Start here".
AND THE HARNESS CAUGHT THE RENAME. Immediately after the move, validate.sh went to
22 passed / 1 failed: "374.4 printed by reentry-alumina-ledger but ABSENT from
Study-29-Affine-Earth-Sovereign-Substrate.md". Its page reference was stale because I had
just made it so. Fixed, back to 23/0. A harness that only ever passes would not have
noticed, which is the argument for having one that can fail.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Boundary split: strip private references, publish the runner programs and corpora
The code repository is private; this wiki is public. Studies published here had been
citing the private tree, and two published yesterday went further than any page before
them.
WHAT CROSSED, measured before this commit:
private paths (cells/xcode/*) 12 files, longstanding
~/.gaiaftcl and home paths 7 files
.swift:NNN line citations 2 files — both added yesterday
verbatim substrate source 1 file — added yesterday, the first ever here
internal constants 1 file — added yesterday
Before yesterday this wiki carried ZERO Swift code blocks and ZERO file:line citations.
WHY THE LINE CITATIONS WERE THE WORST OF IT. "Cite file and line" is an internal
verification standard. Applied to a public page it maps a private tree AND buys the reader
nothing, because they cannot open the file to check it. Worst of both directions.
That exposed a second defect the leak was hiding: the "reproduce every number" sections
told public readers to run programs that lived in a private repository. Unfollowable.
SO THE FIX IS NOT ONLY REDACTION. The ten runner programs are self-contained — zero
imports but Foundation in the two that read a file, zero private references, verified
before publication — and they exist precisely so a stranger re-runs them. They are now
public, with the corpora they read:
reproduce/ 10 programs + README, each naming its control arm
corpus/study-29/ GCAT satcat.tsv, 19,293,257 B, sha256 090ce077…
corpus/flood-lead-time/ four USGS gauge series + the propagation cases
Both corpora verify against their SHA256SUMS after the move, and the GCAT digest matches
the pin taken when it was fetched.
REDACTED EVERYWHERE ON THIS WIKI: cells/* paths, ~/.gaiaftcl and home paths, file:line
citations, the one verbatim source line, and the internal constants. Claims were rewritten
to cite BEHAVIOUR and the live public catalogue rather than source locations — Study 30 now
says outright that the substrate source is not public and that a citation a reader cannot
open is not evidence.
NOT redacted, and flagged rather than decided unilaterally: bare private module filenames
(QCCourtLaw.swift, LatticeDomainIntegration.swift, the Eclipse commands) appear in ten
pre-existing pages. They are none of the five categories in the instruction, and several
name laws that are themselves publicly served.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Split 29/30: the continuous-model shear, and the sovereign edge pod
Founder's call: "Splitting prevents the empirical hardware specification from diluting the
mathematical precision of the atmospheric shear refutation."
STUDY 29 — THE CONTINUOUS-MODEL SHEAR. The satellite-reentry mass ledger against the
continuous chemistry-transport models that consume it. The models disagree on the SIGN of
the ozone response and on the species they attribute it to, and that disagreement is the
shear signal. Five positions now on the page, each attributed: Ferreira et al. 2024
quantify mass only and report no ozone percentage; Maloney et al. 2025 (WACCM6 + CARMA)
find a WEAKER springtime ozone hole; Revell et al. 2025 (SOCOLv4) find at most a +0.27%
ozone INCREASE, from launch alumina, explicitly not simulating reentry; Barker et al. 2026
find chlorine from solid propellant dominant with megaconstellations at 9%, and rocket
black carbon flipping sign between +6.47 and -6.40 mW/m2 on one emissions set under two
forcing definitions; Vliex et al. 2026 (GEOS-Chem) attribute 87.7% to NOx.
The forcing ledger is exactly countable: GCAT dry mass conforms to integer-or-3-decimal
parsing on 1,907 of 1,907 2025 rows, so the lattice unit is the GRAM and no float touches
it. Corpus pinned by sha256.
THE FORTY-NINTH INSTANCE. The page states what it is: 48 domains already served publicly
each name the float form they replaced (geometry drops vol = integral_P 1 dx for
h_P(t)=|tP intersect Z^d|; cs drops IEEE-754 sums for a/b + c/d = (ad+bc)/bd). A
continuous-model shear is an instance of a shipped, catalogued discipline, checkable with
one anonymous GET, not a novel claim.
STUDY 30 — THE SOVEREIGN EDGE POD. The constructive architecture: the open A.E.P-1
specification, the integer HAL, the zero-heap WASM sandbox, the asymmetric NATS leaf, and
a four-tier gate with frozen integers and control arms in both directions. It says on its
own face that it is NOT a shear study, carries one of the recipe's four pieces, and names
the three it lacks.
TWO CORRECTIONS EACH PAGE CARRIES ABOUT ITSELF:
- The cost advantage is WITHDRAWN. Earlier drafts reported 170x and then 256-830x. Both
priced one satellite subscription per sensing node while handing the terrestrial mesh
its gateway backhaul free. Charged symmetrically at 1:100 aggregation, orbital is
cheaper. No cost advantage is claimed anywhere.
- The geometry claim is NARROWED from an eight-axis map onto T^8 to the three-axis
spatial quotient the shipped code performs. wrapped() emits a winding for sx/sy/sz
only; the record's other coordinate-bearing components are emitted unwrapped.
Energy figures corrected to exact integer division: 66,656:1 at the binding 60 tx/hr rate
(the frozen 60 s ingest period binds before the 125 tx/hr regulatory ceiling) and 32,427:1
at that ceiling. Both were wrong in the previous draft, and the program that prints them
was itself printing "66 : 1" until this pass.
Board bumped to thirty studies; white paper v3.4; OPEN set is now 03/05/08/26/27/29/30.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>