Skip to content

Commit 9418f57

Browse files
committed
P-037 Phase B / B1-F2-F2: R_B retake at the widened instrument (T_B = fdde0a0)
B1-F2-F2 widened scripts/p037_b_extractor_diff_gate.py's MUTABLE_METHODS to include EmitFlowExpr (release-emission, decoupled from ConsumeReleaseArgs's tracking role) after proving the prior 3-method seam was mechanically insufficient for the future B2.1a treatment. That moved instrument_identity (3591d805... -> d93e34e0...), superseding the cfe7e19-based R_B the same way B1-F2 itself superseded a02ccf1's. Four fresh governed takes at fdde0a0 (MOS repo, MOS corpus, verdict/python, verdict/rust), each verified before the next started: 82/192 documents, 105 findings both engines, 0 cross-engine mismatches -- identical to all three prior R_B takes. Diffing the four new artifacts against their cfe7e19-based predecessors confirms every changed line is a commit-identity field or a population-commit path substring; zero facts/verdict content moved. p037_controls.py --engine both stays "all match", p037_fact_shapes.py stays "all 55 shape(s) match", and --post-a1 --engine both stays fully RED (exit 1) -- no semantic treatment occurred. Evidence-only: five files under docs/evidence/, nothing else. named_later.T_B/R_B are not updated here -- that is the naming commit's own separate act, once this is terminal-green. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01AN6xHbpovxrmZ4AMjS7WQA
1 parent fdde0a0 commit 9418f57

5 files changed

Lines changed: 835 additions & 816 deletions

‎docs/evidence/p037-b-baseline-manifest.md‎

Lines changed: 83 additions & 64 deletions
Original file line numberDiff line numberDiff line change
@@ -2,9 +2,9 @@
22

33
epoch: `b`
44

5-
T_B (terminal-green, the measurement-instrument head): `cfe7e19c4f6a9f4d2d7606bcc6f288b52f196bcd`
5+
T_B (terminal-green, the measurement-instrument head): `fdde0a026d264ceff0e44f8c86e7d059207f67e6`
66

7-
population_commit (every take): `cfe7e19c4f6a9f4d2d7606bcc6f288b52f196bcd`
7+
population_commit (every take): `fdde0a026d264ceff0e44f8c86e7d059207f67e6`
88

99
environment: `P037_B_MEASUREMENT_M3`
1010

@@ -17,48 +17,59 @@ retaken, compared, or treated as a control here (`docs/evidence/p037-b-
1717
epoch.json`'s own `predecessor`/`environment` sections state this
1818
precisely).
1919

20-
## This is a retake, not the first or second R_B
20+
## This is the third R_B, not the first or second
2121

2222
The instrument this take measures against is not the one `a02ccf1`/
23-
`df9a433` measured against. That earlier instrument was itself terminal-
24-
green and honestly measured, but B1-F2 found a boundary defect in it:
25-
`frontend/roslyn/OwnSharp.Extractor/` was classified as full measurement
26-
INSTRUMENT with no carve-out at all, while the frozen A1 acceptance matrix
27-
(`docs/notes/p037-formal-kernel.md` #8.1, items 5 and 8) requires an
28-
extractor-side semantic change there -- a treatment surface cannot
29-
simultaneously be frozen measurement instrument. B1-F2
30-
(`docs/notes/p037-formal-kernel.md` #10.8e) corrected the boundary:
31-
`scripts/p037_evidence_b.py`'s `INSTRUMENT_CARVE_OUTS`/`TREATMENT_PATHS`
32-
now carve the whole extractor directory out (the same shape `mos.rs`/
33-
`lower.rs` already had), policed at method granularity by a new, second
34-
item-level gate, `scripts/p037_b_extractor_diff_gate.py`. This is a
35-
governance-boundary correction, not a semantic one: no byte of
36-
`mos.rs`/`lower.rs`/`Program.cs` moved, `p037_controls.py --post-a1`
37-
stayed fully RED throughout (reconfirmed below), and the fix is purely
38-
about what counts as instrument versus treatment in the provenance
39-
closure. `a02ccf1`, `df9a433` and `bfa8647` are kept exactly as
40-
committed -- honest records of the instrument and its baseline at the
41-
boundary B1 originally drew -- superseded here, not rewritten: `git log`
42-
on this file's path still shows their content, and
23+
`df9a433` or `cfe7e19`/`069d5bc` measured against. B1-F2-F2
24+
(`docs/notes/p037-formal-kernel.md` #10.8f) found the B1-F2-corrected
25+
extractor seam -- `ConsumeReleaseArgs`/`ConsumesParam`/`CallReleasesReceiver`
26+
alone -- was itself mechanically too narrow for the future B2.1a semantic
27+
treatment: `ConsumeReleaseArgs`'s answer is read by two different
28+
consumers, `EmitFlowExpr` (to fabricate the unconditional `release`) and
29+
the escape/tracking admission check (`consumedArg`, deciding whether a
30+
local stays tracked at all). A scratch prototype (never committed --
31+
`git diff f8ffd42 -- frontend/roslyn/OwnSharp.Extractor/Program.cs` is
32+
empty at this head) proved that changing `ConsumeReleaseArgs`'s answer
33+
to stop the fabrication also starves the tracking admission check: 7 of
34+
the 8 named acceptance fixtures had their caller orphaned out of
35+
`functions[]` entirely, and the 8th lost tracking of an unrelated,
36+
genuinely-owned local. B1-F2-F2 widened `mutable_methods` to include
37+
`EmitFlowExpr` as a fourth, separately-authorized method -- the seam that
38+
actually owns release-*emission*, decoupled from `ConsumeReleaseArgs`'s
39+
tracking role, which stays frozen and unchanged. This is a governance-
40+
boundary widening, not a semantic one: no byte of `Program.cs` moved
41+
(confirmed against `f8ffd42`), `p037_controls.py --post-a1` stays fully
42+
RED (reconfirmed below), and the fix is purely about which existing
43+
methods a future B2.1a commit is authorized to change. `a02ccf1`,
44+
`df9a433`, `bfa8647`, `cfe7e19`, `069d5bc`, and `5579d80`/`f8ffd42` are
45+
kept exactly as committed -- honest records of the instrument and its
46+
baseline at the boundary each held at the time -- superseded here, not
47+
rewritten: `git log` on this file's path still shows their content, and
4348
`docs/evidence/p037-b-epoch.json`'s own `superseded_by_b1_f2` section
44-
names them explicitly, machine-readable, with the reason.
49+
(plus the B1-F2-F2 treatment note) names them explicitly, machine-
50+
readable, with the reason. Full investigation detail (Finding A/B/C, the
51+
scratch-measured blast radius, the chosen not-yet-implemented direction)
52+
is in `docs/notes/p037-formal-kernel.md` #10.8f and
53+
`docs/evidence/p037-b-epoch.json`'s `treatment.b1_f2_f2_seam_investigation`
54+
-- not duplicated here.
4555

4656
## Instrument identity
4757

4858
```
49-
3591d8056a1debabca046ad1faa1715c71516f0ef37d2037da817a3bde2c5d2d
59+
d93e34e0a6e740ad4ffaa6c1894f32c14fbded79b50b5fc2f678da860d2108d7
5060
```
5161

5262
Read from each take's own `instrument_identity` field (computed by
53-
`p037_evidence_b.evidence_fields()`, over Phase B's own, now-corrected
54-
`INSTRUMENT_PATHS`/`INSTRUMENT_CARVE_OUTS` -- the B1-F2 fix). All four
55-
takes below agree on it, byte-for-byte. This differs from the value the
56-
superseded baseline recorded
57-
(`75764c81ac49185adad6a02657b373a7453fc03ba337bb61b86d4694bd8d5065`) --
63+
`p037_evidence_b.evidence_fields()`, over Phase B's own
64+
`INSTRUMENT_PATHS`/`INSTRUMENT_CARVE_OUTS`, unchanged by B1-F2-F2 -- only
65+
the extractor diff-gate's own `mutable_methods` registration moved). All
66+
four takes below agree on it, byte-for-byte. This differs from the value
67+
the superseded baseline recorded
68+
(`3591d8056a1debabca046ad1faa1715c71516f0ef37d2037da817a3bde2c5d2d`) --
5869
expected and confirmed by direct measurement, not assumed: widening
59-
`INSTRUMENT_CARVE_OUTS` to include the whole extractor directory is
60-
exactly the kind of instrument-definition change that must move this
61-
digest, and it does.
70+
`scripts/p037_b_extractor_diff_gate.py`'s `MUTABLE_METHODS` is exactly the
71+
kind of instrument-definition change that must move this digest, and it
72+
does.
6273

6374
## Execution profile (M3, identical across all four takes)
6475

@@ -72,7 +83,7 @@ the fourth take (this retake reuses the same qualified M3 environment
7283
every prior Phase-B take in this session used; only the instrument
7384
definition changed, not the toolchain, so a fresh before/after pair was
7485
not repeated -- the one capture below, byte-for-byte identical to both
75-
prior R_B manifests' own recorded profile, is the evidence for that):
86+
prior `R_B` manifests' own recorded profile, is the evidence for that):
7687

7788
```json
7889
{
@@ -99,7 +110,7 @@ prior R_B manifests' own recorded profile, is the evidence for that):
99110
}
100111
```
101112

102-
Byte-for-byte identical to both prior `R_B` manifests' own recorded
113+
Byte-for-byte identical to all three prior `R_B` manifests' own recorded
103114
profile (and, in turn, to `docs/evidence/p037-a2d-baseline-mos-repo.json`'s
104115
recorded M2 profile) -- the same physical workspace and toolchain,
105116
re-verified rather than assumed unchanged.
@@ -115,16 +126,16 @@ the environment):
115126
- Rust: rustc/cargo 1.94.1.
116127

117128
recipe sha256 (unchanged since qualification; the environment itself was
118-
not touched by B1-F2):
129+
not touched by B1-F2-F2):
119130

120131
```
121132
5fe4ece4c24b720f659a49bbc68b8438080c9e4b8ba554ee40f6eea94c66eb26
122133
```
123134

124135
Recipe: `provision.sh`, retained outside this repository at
125136
`/root/p037-a2d-m2-recipe/provision.sh` on the qualified machine. The
126-
environment/toolchain was not changed by B1-F2 -- only the instrument's
127-
own `INSTRUMENT_PATHS`/`INSTRUMENT_CARVE_OUTS` definition moved.
137+
environment/toolchain was not changed by B1-F2-F2 -- only the extractor
138+
diff-gate's own `mutable_methods` definition moved.
128139

129140
## Workspace
130141

@@ -146,23 +157,23 @@ new T_B before the next was started. None ran concurrently.
146157
4. verdict/rust snapshot (`--epoch b --engine rust`)
147158

148159
All four takes and verifies exited 0. The checkout was clean at
149-
`cfe7e19c4f6a9f4d2d7606bcc6f288b52f196bcd` before the first take and
160+
`fdde0a026d264ceff0e44f8c86e7d059207f67e6` before the first take and
150161
remained at that exact commit, clean, after the fourth verify.
151162

152-
| p037-b-baseline-mos-repo.json | `bf8c6a9bb9595b2cfe53eeba368365a583dde561ceb174e843e92585e64e5e09` |
153-
| p037-b-baseline-mos-corpus.json | `d4fd0f7e553c33b144656be8b4df3c5e82f8d4b28b86ed1324e6cd3a371e98db` |
154-
| p037-b-baseline-verdict-python.json | `970f80ceb23215127ecd1838ff640ffd0b26294425523ba90e4b88a8688edc85` |
155-
| p037-b-baseline-verdict-rust.json | `cbad406da85ffa71bc6028e2b9fc066c570ea62c5d5bfe86f40d7382bb72a311` |
163+
| p037-b-baseline-mos-repo.json | `9d06afd0471485c93979a4c1f9355a6fcd7f91408d1dd6293ded2bc6f8934c1c` |
164+
| p037-b-baseline-mos-corpus.json | `c0421181f37c41b685f14cdf0c37f1251b3fbf7f34056ccd0eeaa87634b62423` |
165+
| p037-b-baseline-verdict-python.json | `96623cf01d289ead9df0ee67aa0df1b8b77b0cd740d3a0e62630827d0cbfcaab` |
166+
| p037-b-baseline-verdict-rust.json | `e98ff9b81a1c3aab7059a86495cf48a5169888f8512966450182af78ac4649d6` |
156167

157168
Every record above carries, and all four agree on:
158169

159-
- `source_commit`: `cfe7e19c4f6a9f4d2d7606bcc6f288b52f196bcd`
160-
- `population_commit`: `cfe7e19c4f6a9f4d2d7606bcc6f288b52f196bcd`
170+
- `source_commit`: `fdde0a026d264ceff0e44f8c86e7d059207f67e6`
171+
- `population_commit`: `fdde0a026d264ceff0e44f8c86e7d059207f67e6`
161172
- `epoch`: `b`
162173
- `environment_id`: `P037_B_MEASUREMENT_M3`
163174
- `dirty`: `false`
164175
- `is_evidence`: `true`
165-
- `instrument_identity`: `3591d8056a1debabca046ad1faa1715c71516f0ef37d2037da817a3bde2c5d2d`
176+
- `instrument_identity`: `d93e34e0a6e740ad4ffaa6c1894f32c14fbded79b50b5fc2f678da860d2108d7`
166177

167178
## Cross-engine parity at baseline
168179

@@ -176,17 +187,24 @@ snapshots do not have a built-in cross-engine comparator by design
176187
pair: "measures the engine, not the change"), so parity was checked
177188
directly: all 192 corpus files carry identical `(line, code, level)`
178189
finding sets and identical exit codes on both engines -- 105 findings
179-
each side, 0 mismatches, exactly matching both prior `R_B` takes' own
180-
numbers. This is the empirical form of the frozen `facts_expectation`
181-
claim (`docs/evidence/p037-b-epoch.json`'s `measurement_policy`): nothing
182-
about the pre-treatment facts or verdict surface moved -- not on the
183-
first instrument, not on the B1-F1-corrected instrument, and not on the
184-
B1-F2-corrected instrument boundary either, because B1-F2 changed no
185-
production semantics.
190+
each side, 0 mismatches, exactly matching all three prior `R_B` takes' own
191+
numbers. Diffing the four new JSON artifacts against their immediately
192+
preceding (`cfe7e19`-based) counterparts confirms the same thing from the
193+
other direction: every changed line is a commit-identity field
194+
(`source_commit`, `population_commit`, `materialization_root`,
195+
`instrument_identity`, `blob`, an ephemeral `sealed_path` tmpdir) or a
196+
population-commit substring embedded in a sampled file path -- zero
197+
findings, facts, or verdict content differs. This is the empirical form
198+
of the frozen `first_semantic_hypothesis.facts_expectation` claim
199+
(`docs/evidence/p037-b-epoch.json`): nothing about the pre-treatment facts
200+
or verdict surface moved -- not on the first instrument, not on the
201+
B1-F1-corrected instrument, not on the B1-F2-corrected instrument
202+
boundary, and not on the B1-F2-F2-widened extractor seam either, because
203+
B1-F2-F2 changed no production semantics.
186204

187205
## P-037 conformance baseline at T_B
188206

189-
Both re-run fresh at `cfe7e19c4f6a9f4d2d7606bcc6f288b52f196bcd`, `OWEN_
207+
Both re-run fresh at `fdde0a026d264ceff0e44f8c86e7d059207f67e6`, `OWEN_
190208
RUST_CORE` pointed at the qualified `own-cli` release build:
191209

192210
- `python scripts/p037_controls.py --engine both`: `RESULT: all match` (the
@@ -195,15 +213,16 @@ RUST_CORE` pointed at the qualified `own-cli` release build:
195213
- `python scripts/p037_fact_shapes.py check --engine both`: `RESULT: all 55
196214
shape(s) match their recorded facts and verdicts`.
197215

198-
Identical results to both prior `R_B` takes' own recorded conformance
199-
state. Also reconfirmed at this same head: `python scripts/p037_controls.py
200-
--post-a1 --engine both` stays fully RED (`RESULT: mismatch`) on all four
201-
controls, both engines -- no semantic treatment occurred as part of this
202-
boundary fix. Recorded here as the P-037 conformance state at the new
203-
T_B, per the B1 brief's own instruction to measure and record it, not to
204-
chase or repair it as part of this baseline.
216+
Identical results to all three prior `R_B` takes' own recorded
217+
conformance state. Also reconfirmed at this same head: `python
218+
scripts/p037_controls.py --post-a1 --engine both` stays fully RED
219+
(`RESULT: mismatch`, exit code 1, all four controls disagreeing with the
220+
post-A1 target on both engines) -- no semantic treatment occurred as part
221+
of this seam-widening fix. Recorded here as the P-037 conformance state at
222+
the new T_B, per the B1 brief's own instruction to measure and record it,
223+
not to chase or repair it as part of this baseline.
205224

206-
Run time/date: 2026-09-24, ~04:26-05:10 UTC, operator run.
225+
Run time/date: 2026-09-24, operator run.
207226

208227
This commit is evidence-only: it updates exactly five files under
209228
`docs/evidence/` (these four JSON artifacts plus this manifest, all
@@ -212,5 +231,5 @@ instrument, treatment, spec, tooling, or CI file is touched.
212231
`docs/evidence/p037-b-epoch.json`'s `named_later.T_B`/`R_B` are not
213232
updated by this commit -- that is the naming commit's own, separate act,
214233
after this evidence is itself terminal-green, mirroring exactly how
215-
`df9a433` preceded `bfa8647` the first retake, and `40cb9b4` preceded
216-
`2ed4d92` before that.
234+
`df9a433` preceded `bfa8647`, and `069d5bc` preceded `5579d80`/`f8ffd42`,
235+
before it.

0 commit comments

Comments
 (0)