Expectation against the review (written after reading it, 2026-10-11)#

The expectation (expectation-before-review-769.md, IAIP commit 2413a69) named seven things. The review (opus-review-769.md, Opus 5.5, 27 findings) is the other half.

expected	found	which half
listTraces dead tags branch, heuristic limit	F4, ranked 7	both
empty Sessions card for readers	F23	both
host paths in the pure schema	F7, ranked 5	both
reach: node metadata must not reach a reader	held (verified) — but F6: the doors fail open when the audience lookup throws, and F10: visibility follows the first holder	review
sessions merge order-independence	F20	both
rise open redirect bounded to github.com	not a finding; confirmed by reading	mine
hand-built ArtefactAddress in the room drifting from reference-ui	not found as such; F24 (serial lookups, loose gate) was the cost there	review

Not predicted, and the three that mattered most:

F1 ./references imported node:fs through rise-dsl, so @miadi/reference-ui no longer bundled for a browser; the tests (all Node) could not see it. The fix is a server-only module at the package root. The check is now one esbuild line.
F2 a versioned prompt address (…@2) asked the wheel for a node id with the version; the door writes it without. Every chip for a pinned prompt was unresolved.
F3 traceName is in the trace_context field group; without it the trace is named after its root span. The fake in the test answered traceName regardless of fields (F26), which is why the suite passed.

What this calibrates: my expectation found the design-level defects (reach, purity, dead code) and missed the ones that live between a test and the thing it fakes. A second opinion that runs the built artefact against a browser bundler and a real wheel finds what unit tests on fakes cannot. The scoring practice (score-proposal.mjs) measures proposals; this file is the first measurement of a review against its expectation, and the next review's expectation should name a fake-boundary check explicitly.

Fixed before publish (commit on jgwill/Miadi main, #769): F1–F9, F11, F12, F14, F16–F20, F22–F24, F26, F27. Later, named in Episode 552: F10 (visibility by any visible holder), F13 (prompt names with folders), F15 (made_by names operator sessions: a reach rule to decide), F21 (docs), F25 (framer-motion "latest").