fix(facts-lock): 0 frozen facts reads PENDING, not "agree with all 0" (#519)

CHANGELOG.md

@@ -19,6 +19,7 @@ Sections dated before 2026-09-19 predate the cycle and stay as they are. Named m
 
 ## Unreleased
 
+- fix(facts-lock): **`check` no longer reports "N BRD(s) agree with all 0 frozen fact(s)" when nothing was compared** (#519). A corpus whose BRDs carry identifiers only under keys the harvest does not read (`domainEntities[].name`, `pages[].name`) froze an empty lock, and `check` printed what reads as a pass. With 0 frozen facts, `check` now prints `PENDING facts-lock: 0 identifiers found under any known key in N BRD(s): the lock is empty and check proves nothing (keys read: ...)`, or, when identifiers exist but no lock was built, `PENDING ... run build first; check compared nothing`; `build` prints the same empty-lock line when it froze 0 facts and found no conflicts. Exit codes are unchanged (an exit-3 "nothing measured" and harvesting `name` keys are left for a separate change). Fixture: `test-facts-lock.sh` (the issue's repro: 3 fail on master; 19/19). — AlvinnYeo (#519)
 - fix(journey-runner): **the RUNG 1 landing guard now waits up to its 12 s for `ui.ready`** (#536). It called `locator.isVisible({ timeout: 12000 })`, and Playwright ignores that timeout and answers at once, so a page that rendered more than 300 ms after the per-action settle (a progress bar, a microflow before the page opens, a step ending in a fixed `settleMs`) read `reached: FAIL` and every later rung of the step INVALID. It now uses `waitFor({ state: 'visible', timeout: 12000 })`, and the NOT-REACHED screenshot is taken after that wait. Only the landing guard changes; the same no-wait call in `act()` (`click`, lines 184/210/222), the named-control probe and `helpers.js`/`monkey.js`/`full-app-walkthrough.js` are left for a separate change. **Behaviour change to check in your project:** a journey whose ready selector never appears now spends 12 s per such step before it reports NOT-REACHED. Verified by inspection and a Playwright probe (hidden element shown after 2 s: `isVisible` false in 74 ms, `waitFor` true in 2.3 s); no fixture drives the runner's step loop. — AlvinnYeo (#536)
 - fix(build loop, exec): **the commit guidance stages `javasource/` and `javascriptsource/`, and `exec.sh` prints a staging hint when they changed** (#540). `create java action ... as $$` writes `javasource/<module>/actions/*.java` beside the model; the skill's example (`git add App.mpr mprcontents/ mdlsource/<script>.mdl`) committed the model without the body, which builds on the machine that ran exec and not in a clone. `iterative-build-loop.md` (both `git add` examples and the sentence after the exec one, +44 words, on-demand file: baseline 73828 before and after) now names both folders and says to drop one the app lacks (git refuses the whole add on a missing pathspec). After a clean exec, `exec.sh` runs one `git status --porcelain` over the model folder's `javasource/` and `javascriptsource/` and, when either has changes, prints `Action sources changed outside the model ...: stage them with the .mpr.` with up to 5 paths; never a failure, no mxcli call, exit codes unchanged. **Behaviour change to check in your project:** `exec.sh` success output can carry that hint. Fixture: `test-bug07-08.sh` A2/A3 (A2 fails on master; 48/48). — AlvinnYeo (#540)
 - fix(page-fidelity): **a wireframe's `widget-datagrid-content` grid mimic is no longer taken as the page body** (#537). `contentOf()`'s `.main` and `.content` fallbacks matched `\bmain\b` / `\bcontent\b`, and `\b` treats `-` as a boundary, so a hyphenated class such as `widget-datagrid-content` won, the page was reduced to the inside of one grid, and the title and buttons outside it scored `headings 0/0  actions 0/0` (or UNMEASURED). Both fallbacks now match the whole class word with the existing `cls()` helper (still `div` only), as the popup rules already did. **Behaviour change to check in your project:** text-match scores (and the `docs/PAGE-FIDELITY.tsv` rows they append) can move for a wireframe with no `<main>`/`wf-screen`/`wf-shell`/`mockup-frame` whose first `main`/`content` match was a hyphenated class such as `main-area` or `x-content`: it is now scored from the next rule (often `<body>`). Fixture: `test-page-fidelity-mocks.sh` (the issue's wireframe: fails 3 on master; 83/83). — AlvinnYeo (#537)

bin/facts-lock.sh

@@ -212,6 +212,8 @@ for ns, clusters in sorted(harvest.items()):
             conflicts.append((ns, ck, {v: sorted(spellings[v]) for v in variants}))
 
 # --- act ----------------------------------------------------------------------------------------
+EMPTY_MSG = ("PENDING facts-lock: 0 identifiers found under any known key in %d BRD(s): the lock is empty "
+             "and check proves nothing (keys read: " + ", ".join(k for ks in KEYED.values() for k in ks) + ", *_CODE_SET)")
 if CMD == "build":
     for ns, ck, canon, srcs in new_facts:
         frozen.setdefault(ns, {})[ck] = {"canonical": canon, "sources": srcs}
@@ -229,6 +231,8 @@ if CMD == "build":
     total = sum(len(v) for v in frozen.values())
     print("facts-lock: %d fact(s) frozen across %d namespace(s), from %d BRD(s)" % (total, len(frozen), len(brds)))
     print("            %s" % LOCK)
+    if total == 0 and not conflicts:
+        print(EMPTY_MSG % len(brds))
 
 for src, err in unreadable:
     print("  UNREADABLE  %s — %s (its facts are NOT in the lock)" % (src, err))
@@ -253,7 +257,13 @@ if CMD == "check":
         print("  UNFROZEN   %s/%s is contested (%s) — build has not resolved it" % (ns, ck, "/".join(sorted(sp))))
     if not deviations and not conflicts:
         total = sum(len(v) for v in frozen.values())
-        print("facts-lock: %d BRD(s) agree with all %d frozen fact(s)" % (len(brds), total))
+        if total == 0:
+            # 0 compared is not agreement (#519): a lock with no facts proves nothing.
+            print(EMPTY_MSG % len(brds) if not harvest else
+                  "PENDING facts-lock: the lock holds 0 facts while %d identifier(s) were found in %d BRD(s): run build first; check compared nothing"
+                  % (sum(len(c) for c in harvest.values()), len(brds)))
+        else:
+            print("facts-lock: %d BRD(s) agree with all %d frozen fact(s)" % (len(brds), total))
 
 if unreadable:
     sys.exit(1)

tests/wave2/test-facts-lock.sh

@@ -119,6 +119,33 @@ esac
 bash "$SUBJECT" "$P" show >/dev/null 2>&1
 [ "$?" -ne 0 ] && ok "an unreadable BRD makes the run non-zero" || bad "exited 0 despite not reading a BRD"
 
+# --- 0 facts is not agreement (#519) -------------------------------------------------------------------
+# The issue's repro: BRDs that carry identifiers only under `name` keys give the harvest nothing.
+# check used to print "2 BRD(s) agree with all 0 frozen fact(s)", which reads as a pass over a
+# comparison that never happened. Exit codes are unchanged; the line must say PENDING, not agree.
+Z="$WORK/zero"; mkdir -p "$Z/analysis/x/knowledge-base/brd"
+echo '{"id":"F001","domainEntities":[{"name":"PaymentPlan"}],"pages":[{"name":"PaymentPlan_Overview"}]}' > "$Z/analysis/x/knowledge-base/brd/F001-a.brd.json"
+echo '{"id":"F002","domainEntities":[{"name":"Paymentplan"}],"pages":[{"name":"PaymentPlan_Overview"}]}' > "$Z/analysis/x/knowledge-base/brd/F002-b.brd.json"
+ZC=$(bash "$SUBJECT" "$Z" check 2>&1)
+case "$ZC" in
+  *"agree with all 0"*) bad "check over 0 harvested facts still reads as agreement"; echo "        $ZC" | head -2 ;;
+  *"PENDING facts-lock: 0 identifiers found under any known key in 2 BRD(s)"*) ok "check over 0 harvested facts says PENDING, not agree" ;;
+  *) bad "check over 0 harvested facts printed neither agree nor PENDING"; echo "        $ZC" | head -2 ;;
+esac
+ZB=$(bash "$SUBJECT" "$Z" build 2>&1)
+case "$ZB" in
+  *"PENDING facts-lock: 0 identifiers found"*) ok "build that freezes 0 facts says the lock is empty" ;;
+  *) bad "build froze 0 facts and did not say the lock is empty"; echo "        $ZB" | head -3 ;;
+esac
+# A lock not yet built over a corpus that does have identifiers: check says build first.
+N="$WORK/notbuilt"; mkdir -p "$N/analysis/x/knowledge-base/brd"
+echo '{"id":"F001","domainEntities":[{"entity":"PaymentPlan"}]}' > "$N/analysis/x/knowledge-base/brd/F001-a.brd.json"
+NC=$(bash "$SUBJECT" "$N" check 2>&1)
+case "$NC" in
+  *"run build first"*) ok "check with no lock built says run build first, not agree" ;;
+  *) bad "check with no lock built did not say run build first"; echo "        $NC" | head -2 ;;
+esac
+
 # --- usage --------------------------------------------------------------------------------------------
 bash "$SUBJECT" 2>/dev/null; [ "$?" -eq 2 ] && ok "no argument exits 2" || bad "no argument did not exit 2"
 bash "$SUBJECT" "$P" nonsense 2>/dev/null; [ "$?" -eq 2 ] && ok "unknown subcommand exits 2" || bad "unknown subcommand accepted"