[pedapudi]

Summary

PR 3 of 5 for the plan-descriptive-growth feature (goldfive#423). Wires harmonograf end-to-end so operators can see which tasks the framework discovered reactively at delegation-observed time vs. tasks the planner authored up front. PR 1 added the data model (Task.discovered: bool + discovery_identity_hash); this PR is the rendering layer.

• Plan view: discovered task cards render with a dashed tertiary-color border + a small "DISC" pill in the bottom-right. Non-discovered cards render byte-identically to before (back-compat baseline test pins this). A "+N discovered" badge surfaces on the panel summary when the count is non-zero.
• Interventions panel: INFO-severity NEW_WORK_DISCOVERED rows (framework-synthesised, per design doc §4.6) render at 60% opacity via data-observational="true". A "Hide observational discoveries (N)" toggle in the header lets users hide them entirely (off by default — users SHOULD see them). WARNING-severity discoveries (sub-agent-authored via report_new_work_discovered) render normally.
• Server round-trip: convert.py + storage/sqlite.py now pass discovered + discovery_identity_hash through pb → storage → pb in both directions. Without this the frontend would see the markers on initial wire but lose them on reconnect via GetSessionPlanHistory.
• Proto stubs: regenerated frontend/src/pb/** via make proto-ts. The PR 1 bump on main updated only the submodule pointer and forgot to regen the stubs; this PR catches them up. Also surfaces the other goldfive PRs bundled into the same submodule pointer (JUSTIFIED_DEVIATION, CAPABILITY_MISMATCH, CONFUSION removal).

Visual rendering changes

Plan view (bottom of harmonograf UI) — before this PR, all task cards rendered uniformly. After this PR:

• A discovered task card gains a dashed border (orange/tertiary) and a small "DISC" pill at bottom-right.
• A "+1 discovered" pill appears next to the plan summary header on the right side of the panel.
• Plans with no discovered tasks render unchanged.

Interventions panel — before this PR, all NEW_WORK_DISCOVERED rows rendered identically regardless of severity. After this PR:

• An INFO-severity row renders at lower contrast (0.6 opacity).
• The header gains a "Hide observational discoveries (N)" checkbox when at least one such row exists.
• Toggling on hides only the observational rows; WARNING rows stay visible.

Browser-verified against a synthesised test DB with tasks.discovered=1 rows. Screenshot showed the dashed first-card border, "+1 discovered" pill, NEW_WORK_DISCOVERED row in the interventions list, and the filter toggle all rendering as designed.

Per-component breakdown

File	Change
frontend/src/gantt/types.ts	Task.discovered?: boolean
frontend/src/rpc/goldfiveEvent.ts	convertGoldfiveTask passes discovered through
frontend/src/components/TaskStages/TaskStagesGraph.tsx	--discovered class + DISC pill SVG group
frontend/src/components/TaskStages/TaskStagesGraph.css	dashed-border + DISC badge styles
frontend/src/components/TaskStages/TaskPlanPanel.tsx	+N discovered badge next to summary
frontend/src/components/Interventions/InterventionsList.tsx	isObservationalDiscovery + filter toggle + data-observational
frontend/src/components/Interventions/InterventionsList.css	filter chrome + 0.6-opacity observational row
frontend/src/pb/**	regenerated proto stubs (PR 1 bump + same-batch goldfive PRs)
server/harmonograf_server/convert.py	both directions of Task round-trip
server/harmonograf_server/storage/sqlite.py	new columns + ALTER TABLE migration

Test coverage

• TaskStagesGraph.test.tsx — 5 new tests in a discovered-task rendering (goldfive#423) block:
  • Baseline: plan with no discovered tasks renders no DISC badge / no accent class.
  • Single discovered task gets the class + badge + data-discovered="true" + data-testid.
  • Mixed plan: only discovered tasks get flagged.
  • Legacy fixture (no discovered field at all) renders as non-discovered.
  • Tooltip hint includes "discovered at runtime".
• InterventionsList.test.tsx — 5 new tests in a observational discovery filter (goldfive#423) block:
  • INFO-severity new_work_discovered rows flag data-observational.
  • WARNING-severity rows do NOT.
  • Toggle is suppressed when no observational rows exist.
  • Toggle defaults to off; observational rows visible.
  • Flipping the toggle hides only observational rows; actionable rows stay.
• goldfiveEvent.test.ts — 1 new test for Task.discovered round-trip through convertGoldfiveTask (forecast vs. discovered fixtures).
• test_convert_round_trip.py — 4 new tests + 1 extended:
  • test_pb_to_storage_preserves_discovered_overlay (pure converter)
  • test_storage_to_pb_preserves_discovered_overlay (inverse)
  • test_storage_round_trip_preserves_discovered_overlay (parametrised over memory + sqlite)
  • test_storage_round_trip_legacy_plan_defaults_discovered_false (legacy frame → defaults)
  • Extended test_sqlite_migration_adds_columns_no_data_loss to assert the discovery fields too.

Test plan

• pnpm test — all new frontend tests pass (722 passing; 2 pre-existing failures in TrajectoryView.headerMath.test.tsx are NOT caused by this PR — verified by stashing the diff and re-running).
• pnpm exec tsc --noEmit — clean.
• pnpm exec eslint on changed files — clean.
• pnpm build — clean.
• pytest server/tests/ — 453 passing (including 15/15 in test_convert_round_trip.py).
• Browser verification: stack running against a worktree-isolated test DB; plan view shows the discovery accent + "+1 discovered" badge; interventions panel shows the toggle + dimmed row.

Constraints honoured

• No changes to frontend/src/gantt/renderer.ts or frontend/src/components/Gantt/ (recent fix(frontend/gantt): delegation-observed arrow tail anchors at source span end (closes #241) #285/fix(frontend/gantt): fitAll ignores nowMs when spans exist; freezeAt for completed sessions (closes #286) #287 areas).
• No submodule bump — third_party/goldfive pointer unchanged.
• Existing plan view renders identically with no discovered tasks (back-compat baseline test).
• Legacy proto frames without the field render as non-discovered (legacy-fixture test).
• No new console warnings / errors on load (verified in browser session).