refactor(dashboard): simplify goal prioritization + O(facts) goals-API dedup (#2695)

[rysweet]

Summary

Follow-up refactor/perf on the #2695 Goals-tab work (hierarchy + differentiated, priority-ordered goals, merged in #2714). This PR is additive and behavior-preserving — it simplifies and speeds up the already-shipped feature, it does not change what the operator sees.

Two changes:

1. refactor(goal-curation): drop the never-populated user_facing prioritization signal.
   PrioritizationSignals.user_facing was a structured, caller-supplied HashSet that no production caller ever populated (the decompose path only ever built depends_on). Carrying a dead signal through the pure prioritize() scorer added an always-false branch and an empty-set plumb on every call site. Removing it shrinks PrioritizationSignals to the one signal that is actually fed (depends_on) — no observable change to output because the field was always empty. Prefer structured goal-graph data over dead configuration (G3).

2. perf(dashboard): make the /api/goals fact-dedup O(facts) instead of O(facts×board).
   goals_at merges meeting/action/decision facts from cognitive memory into the backlog, skipping any fact whose id is already listed. The old check re-scanned the entire active + backlog list (with a serde-map lookup per element) for every fact — quadratic on an endpoint that is polled on every dashboard refresh. Build an O(1) HashSet id index once and insert newly-listed fact ids as we go, so later facts still dedup against earlier ones. Identical output, no quadratic scan.

Reconciled with the just-merged Goals tab (#2695/#2714): one coherent tab that shows hierarchy (structured parent_goal_id nesting), priority-ordered goals (ascending, stable id tiebreak, visible tiers + priority_explicit provenance), and real per-goal status (status_progress).

Rebased onto latest origin/main. Additive; no "Bridge" names; no new stray prints; no silent fallbacks; structured goal-graph data preferred over parsing (G3).

Step 13: Local Testing Results

Detected toolchains:

• Rust / Cargo, single crate simard (edition 2024). Evidence: Cargo.toml ([package] name = "simard", default-run = "simard") + Cargo.lock at repo root; changed files are all .rs under src/goal_curation/ and src/operator_commands_dashboard/. Per the qa-team skill, a Cargo.toml-at-root repo uses cargo test as the outside-in harness (no gadugi framework required).

Chosen validation strategy:

• The user-visible boundary is the dashboard /api/goals JSON contract (served by goals_at) plus the goal prioritization pass and decompose path. I exercised both the native hermetic consumer-boundary tests (which call goals_at(state_root) — the exact fn the HTTP handler calls — and assert on the JSON a browser receives) and the real compiled simard binary end-to-end: the goal CLI against a durable store and the live dashboard serve HTTP server via curl.

Scenario 1 — Simple: prioritization pass + real CLI shows differentiated, visible priorities

Command: cargo test --lib goal_curation::tests_prioritize and simard goal add <p> <desc> + simard goal set-priority + simard goal list (real binary, temp SIMARD_STATE_ROOT)
Result: PASS
Output:

test result: ok. 10 passed; 0 failed  (goal_curation::tests_prioritize)
  - prioritize_differentiates_a_flat_undifferentiated_set ... ok
  - prioritize_leaves_explicitly_set_priorities_intact ... ok
  - prioritize_bands_every_rescored_priority_into_1_through_5 ... ok
  - prioritize_ranks_a_blocker_above_a_non_blocker ... ok

$ simard goal list
ID                              PRIORITY  STATUS       DESCRIPTION
fix-auth-outage-blocking-users  p1        not-started  Fix auth outage blocking users
tidy-up-doc-comments-someday    p5        not-started  Tidy up doc comments someday
refactor-config-loader          p3        not-started  Refactor config loader

Priorities are visible and differentiated (p1/p3/p5), not a flat default.

Scenario 2 — Complex: dashboard /api/goals HTTP boundary — hierarchy field + priority-ordered + real status + provenance + dedup

Command: cargo test --lib operator_commands_dashboard::tests_goals_crud goal_curation::tests_decompose and a live simard dashboard serve + curl -H "Authorization: Bearer …" /api/goals
Result: PASS
Output (hermetic — the exact JSON contract a browser consumes):

test result: ok. 45 passed; 0 failed  (tests_goals_crud)
  - goals_active_ordered_by_priority_ascending_with_stable_id_tiebreak ... ok
  - goals_expose_parent_goal_id_for_hierarchy ... ok        # child → parent_goal_id="umbrella"; parent → null
  - goals_expose_priority_explicit_provenance ... ok
  - goals_exposes_distinct_status_progress_per_lifecycle_state ... ok
test result: ok. 21 passed; 0 failed  (tests_decompose)     # differentiates decomposed siblings via depends_on

Output (live HTTP server against a real durable store, 5 goals; note priority-ascending order, real status, and priority_explicit provenance):

$ curl -s -H "Authorization: Bearer …" http://localhost:8100/api/goals
p1  hotfix-startup-crash     parent=null  status=NotStarted  explicit=false
p2  add-rate-limiting        parent=null  status=NotStarted  explicit=false
p3  overhaul-auth-subsystem  parent=null  status=NotStarted  explicit=false   # p3 tie → stable id order
p3  refactor-config-loader   parent=null  status=NotStarted  explicit=false
p4  tidy-doc-comments        parent=null  status=NotStarted  explicit=true    # operator set-priority honored
active_count=5   (no goals dropped/duplicated — dedup path exercised)

Ordered by priority (highest first) with a stable id tiebreak; each goal's priority is visible; priority_explicit provenance and parent_goal_id hierarchy edge are exposed; status_progress carries real per-goal lifecycle status. The non-null parent_goal_id nesting case (goal decompose needs an LLM/recipe-runner, unavailable offline) is proven by the goals_expose_parent_goal_id_for_hierarchy hermetic test, which drives the same goals_at code with a properly-seeded store.

Regression + CI gates (local)

cargo fmt --all -- --check ......................... clean
cargo clippy --all-targets --all-features --locked . 0 warnings
cargo test --lib goal_curation .................... 289 passed; 0 failed
cargo test --lib operator_commands_dashboard ...... 534 passed; 0 failed
pre-push (fmt + release test subset + clippy -D warnings) ... all Passed

Related: #2695, #2714

Co-authored-by: Copilot 223556219+Copilot@users.noreply.github.com

[github-actions]

📊 Coverage Summary

Generated by cargo llvm-cov --workspace --summary-only (nightly, excluding test files)

Module	Lines	Covered	Coverage
Total	146810	122468	83.4%

_{Coverage data from CI run. Test files matching tests?/ are excluded from line counts.}