Version. 0.2
Date. 2026-08-17
Status. Standing. This is the canonical statement of the rule. Everything else cites it.
Author. Claude Code (CR-2026-225 §8.3, Operator gate approval 2026-08-17). Operator: Marvin Percival.
Occasion. CR-2026-225 (engagement semantic recall) merged to engine main (5954c70, tag cr-2026-225-engagement-semantic-recall). The retrieval limit stops being the honesty boundary; the answer's own scope statement becomes it.
Supersedes. v0.1 (2026-08-06), which stands as sibling — the window-conditional rule it stated was right for the recency-window retrieval that existed then, and is preserved below, not smoothed.
The Companion's question-answering may be demonstrated live, in front of a prospect or a customer, on any engagement of any size — in an environment that is actually running CR-2026-225 with its backfill complete.
In an environment that is not — not yet migrated, or migrated with the embedding backlog outstanding — the v0.1 rule continues to govern: demonstrable only for an engagement holding fewer notes than the retrieval window; above the window, not.
And the universal commons is excluded regardless, in every environment. It is excluded from recall's embedding by CR-2026-225 itself (its size grows with adoption rather than with work), so a commons demonstration would demonstrate the fallback, not the product.
The v0.1 rule existed because retrieval was wholesale recency — the fifty most recent notes, no search, so a question about something said early failed silently above the window, and honestly bounded is a harder thing to show a stranger than complete.
CR-2026-225 replaces the mechanism. The question is embedded and the engagement's notes are searched by meaning — current, superseded, and retracted versions included — the readout is composed by the server over exactly the retrieved set, each item carries its provenance, corrections are surfaced as corrections with their successors, and the answer states its own retrieval scope ("searched all N notes on record … showing the K most relevant"). The honesty boundary moved from a hidden window to a stated sentence. That is what a demonstration can stand on.
The condition is not decoration. A deployment that has merged the code but not run the backfill answers from the loud fallback (the recency window, explicitly marked "this was not a semantic search"), and an environment still on pre-225 code answers from the silent-window mechanism v0.1 was written against. The bar lifts where the mechanism actually runs — per environment, verified, not assumed. As of this note's date, no environment qualifies yet: the dev database's backlog (130 note versions across 8 engagements) awaits the embedding key's provisioning, and the serving perimeter awaits its restart onto the merged code. The rule is written for the state the environments are entering, and the check below is what makes the claim per-environment honest.
cr-2026-225-engagement-semantic-recall, migration 0111 applied, restarted onto that code.)Unchanged from v0.1: it does not bar discussing recall in any depth with anyone; it does not bar demonstrating anything else (rooms, provenance walk, contribution, downloads, the record); and it does not describe a defect — a fallback that names itself is the product behaving correctly.
> The Companion's question-answering may be demonstrated live, in front of a prospect or a customer, for an engagement holding fewer notes than the retrieval window. Above the window, it may not. And the universal commons is excluded regardless of its size.
Right when written (2026-08-06, after B-11 pinned answer determinism and attached sources); superseded in mechanism by CR-2026-225, and still the governing form for any environment not yet running that CR with backfill complete.
Cite this note; do not restate the rule. A document that states it again creates a second lineage, which is the thing this note exists to end. When the rule changes, it changes here.
DUNIN7 — Done In Seven LLC — Miami, Florida Loomworks — demonstrating the Companion's answers (standing note) — v0.2 — 2026-08-17 The honesty boundary moved from a hidden window to a stated sentence.