pin LF so the sidecars verify on a Windows checkout
acer+1: Claude Opus 5 (1M context)
3970fb6
The index already held LF, but with no .gitattributes a Windows checkout
materialised AGENT-DISCIPLINE.md with 728 CR bytes, hashing d0e18b4b23818b33
while its committed sidecar records 18432cac9676d195. A byte-literal sidecar
reader failed here and was right to.
Same pin path2-two-shadow-recovery already carries. Nothing about the stored
objects changes; only what lands on disk.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
write the missing sidecar for this receipt
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
330a62e
This receipt carried no .sha256 companion, so nothing could verify it on a
fresh clone. The sidecar records the hash of the bytes as they are stored now.
It does not assert the receipt was correct when written - only that these are
the bytes, so any future change becomes visible instead of silent. Written only
where no sidecar existed; nothing already recorded was replaced.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
f890ecf
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
4abcdb6
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
f4d77d8
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
243d4e6
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
e0149b4
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
0ebda22
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
f8c038b
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
d4d8b8c
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
de9929d
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
cc24454
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
d78791a
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: sections 13, 14 and 15
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
ef92551
13 - an infinite run does not end; sample it, then close it. Exit 124 is
`timeout` saying I stopped it, not evidence about the runner. Early
exit is the bug and no exit is health.
14 - pin the toolchain AND say why. Measured: 27 Rust repos, 0 pins, 32
distinct version strings. An agent that chose 1.97 was not disobeying;
there was no gate to pass.
15 - MEASURED_IS: what the thing IS, which is not a verdict. Count, obtaining,
falsifier, scope - all four mandatory. A row without a falsifier is an
unfiled claim.
Plus a section 2 addendum: float also fails reproducibility across hosts and
range above 2^53, both found in live corpus code.
Measured on the falcon seat, carried verbatim, landed from the acer seat.
Renumbered deliberately - this file already carried a section 12.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
CI: three-valued verdicts, so the gate stops lying in both directions
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
6e8f599
Two faults, both found by this gate behaving the way AGENT-DISCIPLINE.md
section 12 says a gate must not:
1. clippy reported failure on a repo with no crate at the root. A repo
that cannot be measured is NOT_MEASURED, not broken.
2. the float grep printed "OK: zero float types" when src/ did not
exist -- a pass declared after scanning zero files. It could not
fail, so it proved nothing.
Now every check emits MEASURED_PASS / MEASURED_FAIL / NOT_MEASURED with
a mandatory reason, the float step reports how many files it scanned,
and the final gate goes red only on MEASURED_FAIL.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
CI: three-valued verdicts, so the gate stops lying in both directions
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
9847ab4
Two faults, both found by this gate behaving the way AGENT-DISCIPLINE.md
section 12 says a gate must not:
1. clippy reported failure on a repo with no crate at the root. A repo
that cannot be measured is NOT_MEASURED, not broken.
2. the float grep printed "OK: zero float types" when src/ did not
exist -- a pass declared after scanning zero files. It could not
fail, so it proved nothing.
Now every check emits MEASURED_PASS / MEASURED_FAIL / NOT_MEASURED with
a mandatory reason, the float step reports how many files it scanned,
and the final gate goes red only on MEASURED_FAIL.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
CI: three-valued verdicts, so the gate stops lying in both directions
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
cad4e2e
Two faults, both found by this gate behaving the way AGENT-DISCIPLINE.md
section 12 says a gate must not:
1. clippy reported failure on a repo with no crate at the root. A repo
that cannot be measured is NOT_MEASURED, not broken.
2. the float grep printed "OK: zero float types" when src/ did not
exist -- a pass declared after scanning zero files. It could not
fail, so it proved nothing.
Now every check emits MEASURED_PASS / MEASURED_FAIL / NOT_MEASURED with
a mandatory reason, the float step reports how many files it scanned,
and the final gate goes red only on MEASURED_FAIL.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: land on the default branch, add section 12
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
8376f3b
The doc has only ever lived on `agent-discipline-preload`, so no agent has
ever loaded it -- a preload on a non-default branch never runs.
Section 12, A VERDICT IS THREE-VALUED: MEASURED_PASS / MEASURED_FAIL /
NOT_MEASURED with a mandatory reason. A clock may measure, it may never
judge. Every verdict carries its vantage. And the mirror failure, a step
that cannot fail -- if the thing under test were completely broken right
now, would this step turn red?
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: land on the default branch, add section 12
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
8fbecdf
The doc has only ever lived on `agent-discipline-preload`, so no agent has
ever loaded it -- a preload on a non-default branch never runs.
Section 12, A VERDICT IS THREE-VALUED: MEASURED_PASS / MEASURED_FAIL /
NOT_MEASURED with a mandatory reason. A clock may measure, it may never
judge. Every verdict carries its vantage. And the mirror failure, a step
that cannot fail -- if the thing under test were completely broken right
now, would this step turn red?
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: land on the default branch, add section 12
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
0a46a53
The doc has only ever lived on `agent-discipline-preload`, so no agent has
ever loaded it -- a preload on a non-default branch never runs.
Section 12, A VERDICT IS THREE-VALUED: MEASURED_PASS / MEASURED_FAIL /
NOT_MEASURED with a mandatory reason. A clock may measure, it may never
judge. Every verdict carries its vantage. And the mirror failure, a step
that cannot fail -- if the thing under test were completely broken right
now, would this step turn red?
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
agent discipline: land on the default branch, add section 12
Jesse Daniel Brown+1: Claude Opus 5 (1M context)
bd9896d
The doc has only ever lived on `agent-discipline-preload`, so no agent has
ever loaded it -- a preload on a non-default branch never runs.
Section 12, A VERDICT IS THREE-VALUED: MEASURED_PASS / MEASURED_FAIL /
NOT_MEASURED with a mandatory reason. A clock may measure, it may never
judge. Every verdict carries its vantage. And the mirror failure, a step
that cannot fail -- if the thing under test were completely broken right
now, would this step turn red?
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Pin channel 1.81.0 with clippy+rustfmt; rust-version = "1.81"; replace
rust-1.97-independent-verification.yml with rust-1.81-clippy-verification.yml
(clippy --all-targets -D warnings, hard fail on any f32/f64, tests, receipt).
Re-run on real rustc 1.81.0 (eeb90cda1) before landing: clippy -D warnings
clean, zero float types, all tests pass. The run records in docs/ now read
1.81 because the runs were actually repeated on 1.81 - not because the lines
were edited.
Lint fix needed to reach clippy-clean on 1.81 (behaviour-preserving):
op_ref: drop the needless & on the left operand of a str/String comparison (src/lib.rs:304). 19/19 tests.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Document Path 1/Path 2 and verify Rust 1.97
Jesse Daniel Brown
Unverified signature
48d66ac
Merge the audited Path-1/Path-2 framing, DBBH→DBWH analysis, storage-backed applicability, verification receipts, and Rust 1.97 CI workflow.
A single lossy shadow is NON-INVERTIBLE (H(X|s)>0, Fano bounds every decoder). Bit-exact recovery
relocates entropy, never beats Shannon: Path 1 = content-addressed retained store (IMPLEMENTED here;
NOT Slepian-Wolf at H(X|s) — the store already holds X, the ~64-bit AGT is a dictionary selector);
Path 2 = a second jointly-injective shadow (CANON theorem, not yet built). The two papers hit a real
final wall (0.65 ceiling = I(X;S)<H(X)); DBBH-CQP works the COMPLEMENTARY regime = lossless RECALL of
a retained object, not shadow-inversion. Adversarial verify RETRACTED the SW-binning over-claim.
Posted regardless of verdict for bilateral attack/compare/verify/PR (acer<->liris).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DAi1nuxgvuMBwsSsvc4eeT
Pure-Rust zero-dep HBP hot-path (json=0). Folds in the full research thread: the two papers
(BrainJanus + exoplanet), the actual Asolaria formulas (prism/quant/frozen-brain/GAC), and
acer's PRISM-COMB-0LOSS canon. BEHCS 64/256/1024/HyperBEHCS(60D) bijective ladder + PID-specific
60D Q-PRISM cube + IX-737 double-black-hole capsule (arm/collapse/revoke) + AI/HW coms +
address-only lossless crossing (shadow not payload). Full test ladder 19/19 (8 unit + 4
integration + 3 suite + 4 system) on WSL rustc 1.96, zero deps. MEASURED: bijections, capsule,
lossless-by-address, held-cases. CANON: IX-737, PRISM-COMB-0LOSS. DESIGN: the coms framing.
UNVERIFIED: live Hilbra tunnel, hardware fire, millions-faster.
Independent second implementation of Liris host8/dbbh_coms_quant_prism.rs (Q-PRISM PR#1, 6/6),
attack-verified re-run on the acer seat.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DAi1nuxgvuMBwsSsvc4eeT
