S1 — Name all three packages where the wheel is described, and check it (closes C-96)

.github/workflows/publish_package.yml

@@ -9,8 +9,12 @@ name: Publish Package
 #   Publisher: GitHub  |  Owner: views-platform  |  Repo: views-frames
 #   Workflow:  publish_package.yml  |  Environment: (leave blank)
 #
-# The single `views-frames` wheel ships BOTH packages (`views_frames` +
-# `views_frames_summarize`) — one PyPI project, `pip install views-frames`.
+# The single `views-frames` wheel ships all three packages — `views_frames`,
+# `views_frames_summarize` and `views_frames_reconcile` — as one PyPI project, so
+# `pip install views-frames` makes all three importable. The authority is
+# `[tool.hatch.build.targets.wheel] packages` in pyproject.toml. This comment named
+# only the first two until 2026-08-18, having been written before the third package
+# shipped in v1.7.0. `docs/validate_docs.sh` check 10 now keeps the two in step.
 
 on:
   release:

docs/guides/publishing-to-pypi.md

@@ -110,7 +110,7 @@ Trusted-Publishing workflow (recommended) **or** create a new token **scoped to
 ```bash
 rm -rf dist && uv build
 uvx --from twine twine check dist/*            # both files must say PASSED
-# sanity: BOTH packages + their py.typed are in the wheel
+# sanity: all three packages + their py.typed are in the wheel
 python3 -c "import zipfile,glob; ns=zipfile.ZipFile(glob.glob('dist/*.whl')[0]).namelist(); \
 print([n for n in ns if n.endswith('py.typed')])"   # expect all three packages' py.typed
 

docs/validate_docs.sh

@@ -284,6 +284,63 @@ fi
 
 fi  # ../src exists
 
+# 10. The documents that describe the CURRENT wheel name every package it ships, and
+#     none of them claims a stale count (register: the package-count drift class).
+#
+#     Two halves, because neither catches the other's case. `.github/workflows/
+#     publish_package.yml` named two of three packages while never using a count word,
+#     so only the completeness half sees it; `docs/guides/publishing-to-pypi.md` said
+#     "BOTH packages" in one comment while naming all three elsewhere in the same file,
+#     so only the literal half sees it. Both were live on 2026-08-18.
+#
+#     SCOPED ON PURPOSE to documents that describe the wheel as it is today. ADRs,
+#     postmortems, the register and the CHANGELOG state counts that were true when
+#     written — "Two packages to maintain" (ADR-017), "shipped in both packages" (a
+#     dated C-23 resolution) — and flagging those would make this check noise, which
+#     is how a check gets deleted. Same reasoning as check 3, which restricts itself
+#     to the constitutional ADRs.
+#
+#     WHAT IT DOES NOT CATCH: a novel phrasing that omits a package in a file which
+#     names all three somewhere else. The completeness half is per-file, not per-claim.
+echo "--- Checking the wheel-describing documents against pyproject.toml ---"
+before=$errors
+WHEEL_DOCS="../.github/workflows/publish_package.yml ../docs/guides/publishing-to-pypi.md"
+if [ ! -f "../pyproject.toml" ]; then
+    echo "  ERROR: no ../pyproject.toml; cannot read the wheel's package list"
+    errors=$((errors + 1))
+else
+    # The source of truth: [tool.hatch.build.targets.wheel] packages = ["src/a", "src/b"]
+    pkgs=$(grep -m1 '^packages = \[' ../pyproject.toml | grep -oE '"[^"]+"' | tr -d '"' | sed 's|.*/||')
+    pkg_count=$(printf '%s\n' "$pkgs" | grep -c .)
+    if [ "$pkg_count" -eq 0 ]; then
+        echo "  ERROR: could not read [tool.hatch.build.targets.wheel] packages from pyproject.toml"
+        errors=$((errors + 1))
+    fi
+    for doc in $WHEEL_DOCS; do
+        if [ ! -f "$doc" ]; then
+            echo "  ERROR: $doc is missing; it is one of the documents that describes the wheel"
+            errors=$((errors + 1))
+            continue
+        fi
+        # (a) completeness — every shipped package is named
+        for pkg in $pkgs; do
+            grep -qE "\b${pkg}\b" "$doc" || {
+                echo "  ERROR: $doc does not name '$pkg', which the wheel ships"
+                errors=$((errors + 1))
+            }
+        done
+        # (b) literal — no stale count word
+        stale=$(grep -niE '\b(both|two) packages\b' "$doc" || true)
+        if [ -n "$stale" ]; then
+            echo "  ERROR: $doc claims a stale package count (the wheel ships $pkg_count):"
+            printf '    %s\n' "$stale"
+            errors=$((errors + 1))
+        fi
+    done
+fi
+[ "$errors" -eq "$before" ] && echo "  OK (wheel ships $pkg_count packages; checked $(printf '%s' "$WHEEL_DOCS" | wc -w) wheel-describing documents)"
+
+
 echo ""
 if [ "$errors" -gt 0 ]; then
     echo "=== FAILED: $errors issue(s) found ==="

reports/technical_risk_register.md

@@ -5,9 +5,9 @@
 | Project           | views-frames                         |
 | Owner             | VIEWS platform maintainers           |
 | Last Updated      | 2026-08-18                           |
-| Total Concerns    | 92                                   |
+| Total Concerns    | 93                                   |
 | Open Concerns     | 10                                   |
-| Resolved Concerns | 82                                   |
+| Resolved Concerns | 83                                   |
 | Disagreements     | 12                                   |
 
 ---
@@ -441,6 +441,54 @@ Cross-refs: C-47 (eval provenance kept out of the generic header — the precede
 
 > Resolved 2026-07-31 by **ADR-027** (Epic #208 / S1 #209) — the #113 decision.
 
+### C-96: the wheel's package count drifted in the documents that describe it — RESOLVED
+
+| Field | Value |
+|-------|-------|
+| ID | C-96 |
+| Tier | 4 |
+| Resolved | 2026-08-18 |
+| Resolution | Both live sites corrected, and `docs/validate_docs.sh` check 10 added so the documents that describe the wheel are held against `[tool.hatch.build.targets.wheel] packages`. |
+| Source | pre-release falsification audit (2026-08-18, F3) — an unpredicted finding |
+| Cross-refs | C-74 (a check that does not run), C-86 (the README tree, the same drift one document over), C-77 (the evidence discipline this entry is an instance of), ADR-023 (the release that made the count three). |
+
+The wheel has shipped three packages since v1.7.0. Two documents still said two:
+
+- `.github/workflows/publish_package.yml` — a comment in the workflow that **actually
+  publishes**, naming two of the three. It never used a count word, so only a completeness
+  check sees it.
+- `docs/guides/publishing-to-pypi.md` — `# sanity: BOTH packages + their py.typed`, sitting
+  **two lines above** a comment corrected earlier the same day to say "all three". The file
+  names all three packages elsewhere, so only a literal check sees it.
+
+**Tier 4 — no correctness impact.** Both are comments. The reason this is registered at all is
+the second one: that file was corrected on 2026-08-18 and the class was reported as fixed after
+a grep for one remembered phrasing. This is the fourth time this register has recorded a fix
+that reached the instance rather than the class, and the first time the miss was inside a file
+the author had already edited.
+
+**Check 10 has two halves because neither catches the other's case** — completeness (every
+package in the wheel target is named) misses the guide, literal (no stale count word) misses
+the workflow. It is scoped to the documents describing the *current* wheel and excludes ADRs,
+postmortems, this register and the CHANGELOG, all of which state counts that were true when
+written: ADR-017's "Two packages to maintain", C-23's "shipped in both packages". Flagging
+those would make the check noise, and a noisy check gets deleted. Same scoping reasoning as
+check 3, which restricts itself to the constitutional ADRs.
+
+Mutation-tested in five directions: a stale count word reintroduced, a package dropped from
+either document, a **fourth** package added to the wheel target and left undocumented (the
+generalisation test), and an unreadable package list. All five error.
+
+It then caught a real regression within minutes of being written: a `git checkout` used to
+clean up after the first mutation discarded the still-uncommitted fix to the workflow, and the
+check failed on the next run rather than the mistake reaching a commit.
+
+**What it does not catch**, stated in the check's own comment rather than left to be found: a
+novel phrasing that omits a package in a file which names all three somewhere else. The
+completeness half is per-file, not per-claim.
+
+---
+
 ### C-84: the physical-architecture standard described a repository that no longer existed — RESOLVED
 
 | Field | Value |