You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 20b09c9
Browse filesBrowse the repository at this point in the historyBrowse files
AR1/AR2 (a scratch, never-committed two-crate PoC replaying CH3-3's
per-result SARIF boundary against the held R1 treatment) falsified that
boundary's completeness before any T_B was retaken against it: AR2 (a
static selection correctly removing a fabricated double-dispose) measures
a correct, empty AFTER SARIF result set, so a per-result properties bag
has no result to attach a witness to. CH3-3 was a correctly preregistered
hypothesis given the evidence available at the time, not a mistake -- this
amendment supersedes and narrows it, it does not extend it.
Final boundary: run.properties.p037_call_site_witnesses[] (run-level,
plural, populated even when results==[]), schema p037-verdict-snapshot/5.
verdict.rs's struct Finding/fn transfer_note and render.rs's struct
Properties/fn sarif_result -- CH3-3's own item set -- are RE-FROZEN, with
hostile tests proving each rejected; impl Finding stays mutable as the
additive evidence entry point; verdict.rs's fn check_facts and render.rs's
struct Run/fn build_sarif join as the new item-level surface. lower.rs
widens by exactly one item, struct Lowering -- not fn guarded_use_lowered/
fn apply_guarded_uses, which were found (mechanically, via git merge-base
and rust_items()) not to exist anywhere on this branch's own lineage; that
mechanism lives only on R1's separate lineage, built after the branches
diverged. registered_new_items stays {} by design: nothing is
preregistered ahead of the treatment commit that actually defines it.
New scripts/p037_b_cli_diff_gate.py governs rust/crates/own-cli/src/
ownir.rs at item level (fn check/fn display only), the seam a real
run-level producer needs on the own-cli side; scripts/p037_evidence_b.py's
instrument closure is extended to a three-gate check (rust/extractor/cli)
and its own selftest fixtures updated to the CH3-8 shape.
classify_call_site() ports classify()'s own, already-tested
APPLICATION_REFINEMENT branch to the call-site witness (a static pos/neg
selection against an existing split summary, now carrying its own
selection_license), pinned against the two real measured shapes: AR1
(before=nothing, after=a real OWN001 at the resource's acquire site,
never naming the callee) and AR2 (before=a real OWN003 at the acquire
site, after=nothing). CONDITIONALITY_HONESTY's own logic is unchanged.
docs/evidence/p037-b-epoch.json: production_diff_gate.rust/cli record the
final boundary; capture_transport and the CH3-3 instrument-consequence
narrative gain a resolution field each, pointing at this commit, with the
historical text kept verbatim rather than rewritten; a new, explicitly
non-governed note cross-checks this work against PR #371's own
record-absence ruling (PASS -- consistent, not depended upon, not merged).
Zero production semantics: rust/, frontend/, and ownlang/ are byte-
identical to this commit's own parent. Touches only scripts/p037_b_*.py,
the new scripts/p037_b_cli_diff_gate.py, and docs/evidence/p037-b-epoch.json.
instrument_identity moves again, as expected and already recorded --
measured after this commit, never predicted. ruff and strict mypy are
clean; every affected selftest passes, including the new
p037_b_cli_diff_gate.py's own 17 cases; test_p037_a2d_epoch.py and
test_p037_cumulative_evidence.py are unaffected and green.
No T_B named, no R_B taken, no R1 commit/push/replay. Held for
independent review before any retake.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AN6xHbpovxrmZ4AMjS7WQA
0 commit comments