docs: correct the stale delivery-model and test-count claims

The evidence pointers fixed in 1aedd52 were not the only records left behind
when the release bundle was committed. Two more carried the same drift.

README.md described keeping raw run directories with an external evidence
package and said a source clone could not carry the evidence. The bundle is
committed, so the README now points at the archive and its .sha256.txt pin,
while keeping the boundary explicit: verifying the hashes shows the bundle is
internally consistent, not that the producer or host is trustworthy.

README.md also published its "current verified result" as 95 tests run and 94
passed, directly under the commands a reader is told to run. The tree runs 98
and 97 since the DOCTYPE guard landed, so following the README produced a
mismatch. It now reads 98/97 and explains why the release records differ.

docs/v1-local-validation.md said the artifacts were not committed to Git, and
pinned the pre-consolidation archive digest 5feae77e... while naming the
committed archive by path. Both digests are now recorded - the packed-on-host
value as history and the committed 7b9a5a56... as the value to verify against.

The 95/94 figures inside the validation record and the release manifest are
deliberately unchanged. They describe the validated tree, not this one, and
the record now says so in its opening paragraph.

Documentation only. No source, evidence or manifest values were altered.

Author: rootverdict

README.md

@@ -64,9 +64,14 @@ telemetry. The detection run produced a 63-file SHA256 evidence manifest; the
 benign run produced a 12-file manifest. Timeout calibration passed all 20 of
 20 baseline runs, and contract export passed.
 
-The raw run directories are intentionally ignored by Git. Keep them with the
-external evidence package when the VM result needs to be independently
-reviewed; a source clone alone cannot prove a historical Windows telemetry run.
+The release bundle is committed as `artifacts/detfuzz-v1.0.1-release.zip`, with
+its digest pinned in `artifacts/detfuzz-v1.0.1-release.sha256.txt`, so a clone
+carries the evidence these results rest on. Extracted run directories and
+intermediate build output stay ignored by Git.
+
+Verifying those hashes shows the bundle is internally consistent and unmodified
+relative to its recorded manifest. It does not authenticate the producer or the
+host, and a clone alone cannot prove a historical Windows telemetry run.
 See [`docs/v1-local-validation.md`](docs/v1-local-validation.md) and
 [`docs/evidence-checklist.md`](docs/evidence-checklist.md).
 
@@ -251,11 +256,16 @@ python -m unittest discover -s tests
 The current verified result is:
 
 ```text
-95 tests run; 94 passed and 1 expected dependency-path test skipped
+98 tests run; 97 passed and 1 expected dependency-path test skipped
 Ruff: all checks passed
-mypy: no issues found
+mypy: no issues found in 31 source files
 ```
 
+The v1.0.1 release was validated at 95 tests; the three additional tests cover
+the DOCTYPE rejection added after that release. `docs/v1-local-validation.md`
+and the release manifest retain the 95/94 figures deliberately, as a
+point-in-time record of the validated run rather than a claim about this tree.
+
 ## V1 boundary
 
 V1 intentionally covers one safe encoded-command PowerShell rule shape and

docs/v1-local-validation.md

@@ -1,9 +1,13 @@
 # DetFuzz V1 Local Validation
 
 This record captures the latest V1 end-to-end validation completed on
-2026-08-13. Raw artifacts are retained under the ignored
-`artifacts/detfuzz-v1.0.1-release/` directory and in the external portfolio
-archive; they are not committed to Git.
+2026-08-13. It is a point-in-time record: the figures below describe the tree
+that was validated, and they are deliberately not revised as the repository
+moves on.
+
+The release bundle is committed as `artifacts/detfuzz-v1.0.1-release.zip`, with
+its digest pinned in `artifacts/detfuzz-v1.0.1-release.sha256.txt`. The
+extracted `artifacts/detfuzz-v1.0.1-release/` directory stays ignored by Git.
 
 ## Environment
 
@@ -97,21 +101,28 @@ to the next whole second before adding the full `10s` safety margin. The final
 - mypy: no issues found in 31 source files.
 - PowerShell demo helper: parser validation passed.
 
-## External evidence package
+## Evidence package
 
-- Archive: `artifacts/detfuzz-v1.0.1-release.zip`.
-- SHA256:
+- Archive: `artifacts/detfuzz-v1.0.1-release.zip`, committed to Git.
+- SHA256 as packed on the validation host, 2026-08-13:
   `5feae77e09e35323cf4a9384f7efc3353e61416714d5a8a86923b7f5a2e331ab`.
+- SHA256 of the committed archive:
+  `7b9a5a56ce9fc383640b50b75da21278ee7d1340de2a9b390779c9f5527e2e38`.
 - Archive entries: 114 files.
 - Legacy or temporary entries: 0.
 
+The two digests differ because the bundle was repacked when the artifacts were
+consolidated and committed. Verify against
+`artifacts/detfuzz-v1.0.1-release.sha256.txt`, which pins the committed archive
+and is the authoritative reference; the first digest is retained as history.
+
 The archive stores files only; empty per-case working directories are not
 retained because every marker and telemetry artifact is copied into the
 `evidence/` tree that the manifest hashes.
 
 The source repository records these identities and results but cannot prove the
-historical Windows run without the external archive. Recompute the archive and
-evidence-manifest hashes before relying on the result.
+historical Windows run on its own. Recompute the archive and evidence-manifest
+hashes before relying on the result.
 
 ## Reproducibility