facts-lock.sh harvests no identifiers from BRDs that use name keys, then check reports "agree with all 0 frozen fact(s)" #519

File: bin/facts-lock.sh lines 91-95 (KEYED), 100-116 (walk), 249-256 (check summary) · Toolkit: eec646e · Found: 21 Oct 2026, while running the parallel-BRD freeze step

Summary

facts-lock.sh build freezes an identifier only when it sits under one of these keys (lowercased):
operationid, operationids, operation, entity, entityname, targetentity, aggregate, page, pagename,
pageid (plus *_CODE_SET tokens anywhere). BRDs whose arrays of objects carry the identifier as name
(domainEntities[].name, microflows[].name, pages[].name) give it nothing to harvest. The lock is written with
"facts": {}, and check then prints:

facts-lock: 8 BRD(s) agree with all 0 frozen fact(s)
exit 0

That reads as a pass. Nothing was compared. A later BRD that respells an entity, page or microflow will not be caught,
which is the job the script exists for (its header: N agents re-deriving the same identifiers).

The toolkit disagrees with itself on the key

• tests/wave2/test-facts-lock.sh:45,50 fixtures use domainEntities[].entity.
• bin/coverage-check.sh:172 documents pointer cells as /domainEntities/0/name, and enumerates leaves that way
  (:220-222).

So a BRD that satisfies coverage-check.sh can be invisible to facts-lock.sh.

Reproduce

mkdir -p P/analysis/x/knowledge-base/brd
echo '{"id":"F001","domainEntities":[{"name":"PaymentPlan"}],"pages":[{"name":"PaymentPlan_Overview"}]}' > P/analysis/x/knowledge-base/brd/F001-a.brd.json
echo '{"id":"F002","domainEntities":[{"name":"Paymentplan"}],"pages":[{"name":"PaymentPlan_Overview"}]}' > P/analysis/x/knowledge-base/brd/F002-b.brd.json
bin/facts-lock.sh P build
  facts-lock: 0 fact(s) frozen across 0 namespace(s), from 2 BRD(s)          exit 0
bin/facts-lock.sh P check
  facts-lock: 2 BRD(s) agree with all 0 frozen fact(s)                      exit 0

Same two files with every "name" key renamed to "entity" (run in a second scratch folder):

bin/facts-lock.sh Q build
  facts-lock: 1 fact(s) frozen across 1 namespace(s), from 2 BRD(s)
  CONFLICT  entities/paymentplan — 2 spellings, nothing frozen:
      PaymentPlan    F001-a.brd.json
      Paymentplan    F002-b.brd.json                                         exit 1

Same content, same disagreement; only the key name differs. (Run 21 Oct 2026 in $TMPDIR, toolkit eec646e.)

Expected

1. Names under name inside the BRD's identifier-bearing arrays are harvested, into a namespace named for the array
   (entities for domainEntities, pages for pages, microflows for microflows, operations for
   integrations).
2. A build or check that freezes or compares 0 facts over N > 0 BRDs is not reported as agreement. Say
   "0 identifiers found under any known key in N BRD(s): the lock is empty and check proves nothing", and exit
   non-zero (or 3, "nothing measured"), so a gate cannot count it as green.

Suggested fix (ranked)

1. In walk(), carry the parent array's key: when the current key is name and the parent is one of
   domainEntities, entities, pages, microflows, nanoflows, integrations, harvest it into that
   namespace. Keeps the "identifiers only, no prose" rule: still len <= 80 and no spaces.
2. Add the empty-lock guard to build and check (expected point 2).
3. Add a fixture with name-keyed BRDs to test-facts-lock.sh, and one asserting that an empty harvest is not "agree".

How to retest

The two-file repro above: expect CONFLICT entities/paymentplan after the fix, and an empty-lock warning (not
"agree") on any corpus where nothing was harvested.