docs: goal 0057 queued -- live-sync audit, no reload ever (owner-directed) (#112)

Claude-Session: https://claude.ai/code/session_01FJ8wStsHyu7XPLTspNjMnQ

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>

Author: alicoding

docs/goals/0057-live-sync-audit.md

@@ -0,0 +1,61 @@
+# 0057 — Live-sync audit: no surface ever needs a reload
+
+**Raised:** 2026-08-14, owner-directed after hitting the runs-panel
+staleness live ("the problem is that it is not real time — why do we
+have to reload our app ever in any part; can we now look at other
+parts of the app to make sure we don't have similar issues").
+Queue position: ahead of 0052 (owner-picked by asking now).
+
+## The bug class this generalizes
+
+The runs-panel instance (fixed in its own PR, with regression test
+`TestRunWorkflow_EmitsRunDataEventOnStartAndCompletion`): a surface
+fetches once on mount, stays mounted across tab switches, and relies
+on `mill-data-changed{entity}` to refresh — but one or more mutation
+paths for that entity never emit. Result: data a user watches goes
+stale, or appears empty, until a full app reload. Goal 0017
+established the event and swept the then-known surfaces; this goal
+audits systematically instead of waiting for the next live hit.
+
+The principle (goal 0017's, now stated as the app-wide invariant):
+**a user must never need to reload Mill to see the current state of
+anything the app displays.** Every displayed collection/state either
+(a) subscribes to the entity events its data derives from, with every
+mutation path emitting them, (b) deliberately polls (the in-flight
+run detail's honest-only-path pattern), or (c) is provably immutable
+for the session. Anything else is a gap.
+
+## Plan
+
+1. **Inventory (delegated, read-only):** complete matrix of
+   emitters (`dataevent.Emit` + other Go→frontend events), frontend
+   `Events.On` subscribers, fetch-once-no-subscription components,
+   and mutating service methods without emits.
+2. **Judgment pass (main session):** classify every candidate as
+   real gap / deliberate poll / session-immutable, recorded in this
+   file. Surfaces where the entity string doesn't exist yet get one
+   named (the dataevent entity vocabulary is the contract).
+3. **Fix the real gaps** — emits at mutation chokepoints (the
+   runs-panel fix's shape), subscriptions where a panel listens to a
+   subset of its entities; each fix carries a TestHook-seam
+   regression test per testing.md. Cluster into one PR if small,
+   or per-surface PRs if not.
+4. **Prevention:** extend the goal-0017 per-service emit-test
+   pattern to any mutating service that has none, so a future
+   mutation path can't land emit-less unnoticed. No new framework —
+   the dataevent seam already exists; this is coverage, not
+   machinery.
+
+## Acceptance (checkable)
+
+- [ ] The inventory matrix (or a distilled gap table) is recorded in
+      this file, every fetch-once surface classified as
+      gap / deliberate-poll / session-immutable with a reason.
+- [ ] Every classified gap is either fixed (emit + subscription +
+      regression test at the TestHook seam) or explicitly rejected
+      with a reason recorded here.
+- [ ] Every mutating service package has emit coverage in its tests
+      for the methods that mutate displayed state.
+- [ ] No surface in the app requires a reload to reflect a mutation
+      Mill itself performed — spot-checked live on at least the
+      surfaces the inventory flagged most suspicious.

docs/goals/BACKLOG.md

@@ -56,8 +56,17 @@ accepted)**
    in the workflow"; generic without being engineer-only; run
    history already says step). UI copy sweep here; wire names land
    inside 0052 (sequenced before it so the contract never freezes
-   the rejected word); internal identifiers untouched
-6. [ ] [0052 — Contract maturity: generated schema + versioned envelope](0052-contract-surface.md)
+   the rejected word); internal identifiers untouched.
+   IMPLEMENTED 2026-08-14 (PR #110 merged); box 3 (0052 wire names)
+   stays open pending 0052
+6. [ ] [0057 — Live-sync audit: no reload, ever](0057-live-sync-audit.md)
+   — owner-directed 2026-08-14 after the runs-panel staleness hit
+   live ("why do we have to reload our app ever in any part");
+   generalizes that bug class: inventory every emit/subscription/
+   fetch-once surface, fix real gaps with TestHook-seam regression
+   tests, emit-coverage tests per mutating service. Queue-jumped
+   ahead of 0052 by the owner's ask
+7. [ ] [0052 — Contract maturity: generated schema + versioned envelope](0052-contract-surface.md)
    — owner-directed 2026-08-13 ("follow the mature pattern") after
    the far-side contact: schema-first discipline adopted as generated
    JSON Schema from Go types (protobuf/gRPC/OpenAPI rejected with
@@ -66,19 +75,19 @@ accepted)**
    coordinates schema-id evolution semantics with 0046's ADR.
    Sequenced directly after 0044 — the dry run sharpens what the
    contract must express
-7. [ ] [0054 — Step designer + declare-vs-code boundary](0054-step-designer.md)
+8. [ ] [0054 — Step designer + declare-vs-code boundary](0054-step-designer.md)
    — owner-directed 2026-08-13: declare new step types in-product
    (v1 = promote HTTPRequest operations / MCP tools / child
    workflows into named palette types, stored as data) + the ADR
    ratifying WHEN code is genuinely required (new engine) vs
    declaration suffices (naming/mapping over existing engines).
    After 0052 — declared types must be first-class in the generated
    contract from day one; benefits from 0047's palette facet
-8. [ ] [0055 — Canvas note block](0055-canvas-note-block.md)
+9. [ ] [0055 — Canvas note block](0055-canvas-note-block.md)
    — owner-raised 2026-08-14 (n8n sticky-note precedent): annotation
    in the authoring space; not a step (no ports/execution); rides
    the workflow envelope, so coordinates with 0052's schema
-9. [ ] [0056 — Workbench boundary definition](0056-workbench-boundary.md)
+10. [ ] [0056 — Workbench boundary definition](0056-workbench-boundary.md)
    — owner-raised 2026-08-14 ("we never truly defined what Mill
    boundary to be"): the POSITIVE product statement — capture lands
    somewhere useful, grouped, findable, transformable, every
@@ -87,14 +96,14 @@ accepted)**
    contacts/links differ only in schema+routing (one collections
    capability, never N hardcoded verticals — anti-goal recorded);
    build goals queue after its verdict
-10. [x] [0045 — v0.2.0 + release cadence](archive/0045-release-cadence.md)
+11. [x] [0045 — v0.2.0 + release cadence](archive/0045-release-cadence.md)
    — DELIVERED 2026-08-13: cadence rule LOCKED in SPEC §1.3
    (tag-on-capability-goal-archive; the asset is the launchable
    mill.app zip, clone-build stays the documented dev path); v0.2.0
    live and attestation-verified. The cut surfaced and fixed three
    more latent defects (raw-binary asset, release.yml invalid YAML —
    now gated by workflow-lint, GPL-3 setup-task action wrapper)
-11. [ ] [0051 — Run-analytics dashboard v2](0051-run-analytics-dashboard.md)
+12. [ ] [0051 — Run-analytics dashboard v2](0051-run-analytics-dashboard.md)
    — owner-directed 2026-08-13; research DELIVERED same day (n8n/
    Zapier/Make/Windmill/Temporal/Airflow/Power Automate survey): Home
    already covers most of the converged set; ships avg-duration KPI +
@@ -103,17 +112,17 @@ accepted)**
    Activity; two data-model gaps recorded as deferred (per-step
    timestamps, trigger fire-log). Slotted ahead of 0049/0050 per the
    UX-first tiebreak
-12. [ ] [0049 — Docs anti-rot mechanics](0049-docs-anti-rot.md)
+13. [ ] [0049 — Docs anti-rot mechanics](0049-docs-anti-rot.md)
    — owner-ratified 2026-08-13 from the two-agent docs survey: path-
    reference checker (lefthook+CI), generated ADR index with
    drift-fail, revive package-comments floor, AGENTS.md interop,
    README Spec-view fix; rejected list recorded in the goal file
-13. [ ] [0050 — Codebase structure audit](0050-codebase-structure-audit.md)
+14. [ ] [0050 — Codebase structure audit](0050-codebase-structure-audit.md)
    — owner-raised 2026-08-13 (OSS flat-structure critique): audit
    layout against official Go guidance + what OSS reviewers actually
    flag, verdict table + a citable layout ADR; only audit-justified
    moves, no conformance churn
-14. [ ] [0046 — Schema evolution](0046-schema-evolution.md) —
+15. [ ] [0046 — Schema evolution](0046-schema-evolution.md) —
    owner-raised from a real regulated-platform incident
    (rename-forbidden / retype-permanent / live-referenced decisions
    forcing a ~30-version manual cleanup); research delivered
@@ -123,7 +132,7 @@ accepted)**
    Mill's Configure entities are live-referenced (a pinned workflow
    still resolves TODAY's decision definition) and deletes have no
    reference-integrity check
-15. [ ] [0047 — Node audience/complexity facet](0047-node-audience-facet.md)
+16. [ ] [0047 — Node audience/complexity facet](0047-node-audience-facet.md)
    — owner-proposed; session recommendation recorded: function stays
    the primary palette grouping, audience/complexity becomes a
    NodeType metadata field + progressive-disclosure facet (not a