CR-2026-219 — The recall count stops claiming completeness over a slice (B-104) — v0.1

Version. 0.1 Date. 2026-08-15 Author. Claude.ai, drafted against the B-104 build-list entry (v0.87+, filed from inspection-briefs/loomworks-personal-memory-probe-duplicate-shane-v0_1). Repository. loomworks-engine. Build-list item. B-104 (full close).

Plain-language summary

When the Companion says "the N things I have recorded about you," N is the length of a 500-row page, while the same call's true total is received and ignored. Today the true count is 8, so the sentence is honest. At 501 it would silently claim completeness over a truncated set — a false claim of the truthfulness class, waiting on volume. The fix is one comparison: when the total exceeds the page, the sentence says "the N most recent" instead of claiming the whole. No decision needed from the Operator.

0. Established facts (from the filing; verify current at Step 0)

• The recall path counts len() of a _PERSONAL_RECALL_LIMIT = 500 page (router.py:1717-ish at filing time) while the same list_assertions call's total is received and unused.
• CR-2026-129's recall-truthfulness discipline holds for what it loads; this is the claim-over-the-limit edge it never met.
• This is a composed claim and takes the standing note's full gate: composed against the fact, or confessing the limit — never complete-sounding and wrong.

1. Scope

In scope. The comparison and the two sentence forms: total <= len → the existing complete-count sentence, unchanged; total > len → the "N most recent" form. Tests per §3.

Out of scope. The limit's value. Pagination of recall. B-103. Any other sentence the recall path composes.

2. Steps

Step 0 — branch and verify. Branch cr-2026-219-recall-count-limit-edge. Verify CR-2026-219 unclaimed in loomworks-record; HALT if claimed. Locate the current count site and confirm total is available in scope as filed. Report the location. HALT if the shape has changed.

Step 1 — the comparison. total > len(page) selects the "N most recent" sentence form; otherwise the existing sentence, byte-identical.

Step 2 — tests.

• total == len → the existing sentence, unchanged (the today-case, 8 of 8).
• total > len → the "most recent" form with the page length as N — driven by a stubbed total, not by writing 501 rows.
• Each observed failing against pre-fix code before being trusted.

Step 3 — quality gates. ruff zero-tolerance; mypy message-level non-regression.

Checkpoint A — after the green run, before tag and push. Gate string: GATE A APPROVED — PROCEED for CR-2026-219. No live over-limit verification is expected — production truth is 8 rows; the stubbed test is the evidence, and fabricating 493 personal assertions to walk the edge would violate the synthetic-only discipline for no gain. The report says so plainly rather than papering it.

3. Acceptance

1. The over-limit sentence names its slice; the under-limit sentence is byte-identical to pre-fix.
2. Suite green at the zero-failures baseline; ruff clean; mypy at baseline.
3. Build list closes B-104 (new file alongside the current highest version — verify at write time, do not assume).

4. Discipline

Stage by explicit path, never git add -A. Re-verify HEAD before each write. Halt before every push. One session, one repository. An early gate string is asked back, not banked.

DUNIN7 — Done In Seven LLC — Miami, Florida CR-2026-219 — recall count limit edge — v0.1 — 2026-08-15