bug: dynamic ruleset round-trip assumption unverified — CT-recomputed fields would cause perpetual diffs

[2000game]

Found by the PR #18 three-agent review (engine reviewer). Convergence risk at the CT-server-behavior boundary.

What happens

A no-op plan requires deepEqual(desired.dynamic, actual.dynamic); normalizeRuleset strips only the two timestamp keys and normalizes the query subtree. The other RuleSet-level fields — description, shorty, importance, personIdFieldName, process — must round-trip byte-for-byte. Fixtures show auto-generated-looking values (e.g. "Automatische Mitgliedschaft für Gruppe #683"); if CT overwrites or normalizes any of these on PUT (plausible for description/shorty, which reference a group id the user can't know before create), an affected dynamic group updates on every apply forever.

The existing live round-trip test writes back CT's own GET output, so it is drift-free by construction and cannot catch a user-authored ruleset diverging from what CT stores.

Fix sketch

Pin CT's write-back behavior with a live-gated test that PUTs a user-authored ruleset (custom description/shorty) and asserts the subsequent GET matches. If CT recomputes fields, extend the normalizer to drop/canonicalize them (like the timestamps) so plans converge.

Acceptance

• Live-gated test: user-authored ruleset → apply → re-plan is a no-op on eqrm-dev.
• Any CT-recomputed fields identified are normalized out of the diff.

[2000game]

Status (autonomous wave, 2026-07-09): the live-gated round-trip test is merged (PR #59) — tests/dynamic.integration.test.ts PUTs a user-authored ruleset, asserts field-by-field normalized round-trip equality (failure output names exactly which fields CT rewrote), asserts plan-no-op post-PUT (desired status tracks the group's live status), and restores the prior ruleset in a finally. Skipped by default.

Remaining (needs credentials, one-time): run it against eqrm-dev — CT_LIVE=1 CT_LIVE_WRITE=1 CT_LIVE_WRITE_HOST=<dev host> CT_DYNAMIC_FIXTURE_GID=<disposable dynamic group id> npx vitest run tests/dynamic.integration.test.ts (see docs/dynamic-groups.md). If it fails naming rewritten fields → extend normalizeRuleset per the failure output, then close. Caveats from review: a status-only delta = TOCTOU race, not a #36 finding; a status "none" fixture group false-fails (use a real dynamic group).

[2000game]

Handoff state (2026-07-10, end of autonomous session):

• The bug: dynamic ruleset round-trip assumption unverified — CT-recomputed fields would cause perpetual diffs #36 PROPERTY is demonstrated live: the dev rehearsal applied 7 user-authored (file-based, prod-captured) rulesets to eqrm-dev and the re-plan is a clean no-op — CT 3.134.1 rewrote nothing. (This also live-proved the corrected PUT envelope {dynamicGroupRuleSet: [ruleset]}, bug(dynamic): ruleset PUT sends bare object — CT expects the [RuleSet] array envelope (server 500/TypeException) #77/fix(dynamic): correct ruleset PUT envelope — { dynamicGroupRuleSet: [ruleset] } #80.)
• Formal gated test: the write-back round-trip PASSED live against dev fixture group 1698 (skidscheckinopsmz). The bug: dynamic ruleset round-trip assumption unverified — CT-recomputed fields would cause perpetual diffs #36 pin itself needed a payload fix (CT validates ruleset content on PUT: process requires the groupOnly/queryResultOnly/groupAndQueryResult shape; boolean → 0) — fixed in PR test(dynamic): schema-valid authored payload for the #36 gated round-trip pin #81 (open, handoff).
• To close: merge PR test(dynamic): schema-valid authored payload for the #36 gated round-trip pin #81, then run CT_HOST=https://eqrm-dev.church.tools CT_LIVE=1 CT_LIVE_WRITE=1 CT_LIVE_WRITE_HOST=https://eqrm-dev.church.tools CT_DYNAMIC_FIXTURE_GID=1698 npx vitest run tests/dynamic.integration.test.ts. Expect: write-back ✅, bug: dynamic ruleset round-trip assumption unverified — CT-recomputed fields would cause perpetual diffs #36 pin ✅; the older 'committed fixture matches instance' test will fail against 1698 by design (it wants its original prod fixture group — needs a skip-guard or its own GID; noted in PR test(dynamic): schema-valid authored payload for the #36 gated round-trip pin #81's body).

[2000game]

Closed by PR #81 (merged as 5706eb8).

Live verification run just now against eqrm-dev fixture group 1698:

CT_HOST=https://eqrm-dev.church.tools CT_LIVE=1 CT_LIVE_WRITE=1 \
CT_LIVE_WRITE_HOST=https://eqrm-dev.church.tools CT_DYNAMIC_FIXTURE_GID=1698 \
npx vitest run tests/dynamic.integration.test.ts -t pin
→ ✓ 'a user-authored PUT round-trips through GET, and a plan built from it is a no-op' (1 passed)

Both acceptance criteria met:

• Live-gated test: user-authored ruleset (custom description/shorty/importance) → apply → re-plan is a no-op on eqrm-dev.
• No CT-recomputed RuleSet fields needed normalizing — the field-by-field compare found zero rewrites, so normalizeRuleset already covers everything (the two timestamps + query subtree). CT did not overwrite description/shorty/importance/personIdFieldName/process.

The property is also corroborated by the dev rehearsal itself: 7 user-authored file-based rulesets applied to dev, re-plan clean no-op.