[LKPR-140] Dashboard V2 real E2E (FE+BE) suite + 5 regression fixes

[jessinra-megumi-dev]

LKPR-140 — Dashboard V2 real end-to-end (FE + BE) test suite

Replaces the previous false-confidence E2E setup (preview-mode + test.skip-on-empty guards) with a real seeded backend + live frontend Playwright suite. The browser hits the Vite dev origin, which proxies /api/* to a genuine FastAPI backend seeded with deterministic fixtures. Graph feature (LKPR-138) excluded per scope.

Result

48/48 functional tests green. 8 @visual tests gated off pending Linux baselines (see caveat). Full suite verified repeatedly on fresh seeds.

⚠️ 5 latent product bugs surfaced (the entire point of real E2E)

The old preview-mode + skip guards were masking a non-functional dashboard. Fixed inside this ticket (option A — ACs were unachievable otherwise):

1. AppShell overwritten (a839f1f, LKPR-128) — every page rendered nav-only since Jul 12; {@render children()}, TopBar, Toast, CommandPalette all dropped. Restored.
2. memories replaceState crash — URL-object clone threw on the memories page.
3. syncUrl() pre-router crash — sessions/review/links hung on "Loading…" (replaceState threw in mount effect before the router initialized).
4. id/lore_id contract mismatch — memory detail drawer fetched /api/memories/undefined → 404, drawer never opened.
5. SessionDrawer role="complementary" — a modal drawer mis-declared as a sidebar landmark; unreachable by the dialog role, mis-announced to screen readers. Now role="dialog" aria-modal="true" on a <div> (matches MemoryDetailDrawer).

Test-infrastructure fixes

• Seed idempotency — seed.py wipes its data dir up front; without it, re-runs tripped the dedup threshold → 0 inserts → whole webServer boot aborted.
• Port defaults offset to 7787/7788 — 7777 is the conventional live dev-dashboard port; with reuseExistingServer a plain npx playwright test would silently hijack it and run against the wrong data. CI unaffected (spawns fresh servers).
• Hydration-race hardening — click tbody tr.clickable (skip aria-hidden skeleton rows); wait for hydration before tab clicks; expect.toPass() retry for the onMount-attached command-palette hotkey.
• Cross-platform palette modifier — press the modifier the app listens for (from navigator.userAgentData.platform), not Playwright's host-derived ControlOrMeta.
• Query reactivity + cold-start — wait for bind:value flush before Enter; 25s result timeout for the one-time embedding/BM25 cold start.
• Skip-guards → hard assertions — links & sessions drawer tests now fail loudly instead of silently skipping on empty data.
• CI — new playwright-dashboard job on ubuntu-latest (seeds + runs full suite via dual webServer).

⚠️ Caveat — visual baselines deferred

Playwright suffixes screenshot baselines with the OS (-linux.png). CI-environment (Linux) baselines can't be generated here (Docker daemon unavailable), so the 8 @visual tests are gated behind RUN_VISUAL=1 to keep the suite green everywhere. Follow-up: generate Linux baselines in a container (mcr.microsoft.com/playwright), commit them, flip CI to RUN_VISUAL=1.

Notes

• 6 pre-existing svelte-check errors remain in untouched files (DataTable _children/_selectedRows, RelationshipDrawer.test vitest globals, query page type) — out of scope, not introduced here.
• Commit history is granular (18 focused commits, code-first) for easy review/bisect.

[github-actions]

Coverage report

• Total: 78%
• Workflow run: 30247614513
• HTML artifact: coverage-html

The report is also available in the GitHub Actions job summary.

[Jessinra]

@coderabbitai help check for major medium issue

[coderabbitai]

Tip

For best results, initiate chat on the files or code changes.

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo '=== PR metadata ==='
gh pr view 341 --json number,title,baseRefName,headRefName,changedFiles,additions,deletions,url

echo '=== Changed files ==='
gh pr diff 341 --name-only

echo '=== Diff stat ==='
gh pr diff 341 --stat

echo '=== Repository status and merge-base diff summary ==='
git status --short
base="$(git merge-base HEAD origin/$(gh pr view 341 --json baseRefName --jq .baseRefName))"
git diff --summary "$base"...HEAD

echo '=== Outlines: changed source/config files ==='
git diff --name-only "$base"...HEAD |
  while IFS= read -r f; do
    case "$f" in
      *.ts|*.svelte|*.py|*.yml|*.yaml|*.json)
        [ -f "$f" ] || continue
        echo "--- $f ---"
        ast-grep outline "$f" --items all 2>/dev/null | head -n 160 || true
        ;;
    esac
  done

Length of output: 3093

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

base="$(git merge-base HEAD origin/main)"

echo '=== Local diff stat ==='
git diff --stat "$base"...HEAD

echo '=== Runtime product fixes ==='
git diff --no-ext-diff --unified=80 "$base"...HEAD -- \
  src/dashboard_v2/src/lib/api/memories.ts \
  src/dashboard_v2/src/lib/components/overlays/SessionDrawer.svelte \
  src/dashboard_v2/src/lib/components/shell/AppShell.svelte \
  src/dashboard_v2/src/lib/components/shell/NavRail.svelte \
  src/dashboard_v2/src/lib/constants/routes.ts \
  src/dashboard_v2/src/routes/links/+page.svelte \
  src/dashboard_v2/src/routes/memories/+page.svelte \
  src/dashboard_v2/src/routes/review/+page.svelte \
  src/dashboard_v2/src/routes/sessions/+page.svelte \
  src/lorekeeper/dashboard/routes/memories.py \
  src/lorekeeper/domains/memory/repository.py

echo '=== E2E/CI configuration ==='
git diff --no-ext-diff --unified=100 "$base"...HEAD -- \
  .github/workflows/ci.yml \
  src/dashboard_v2/playwright.config.ts \
  src/dashboard_v2/vite.config.ts \
  src/dashboard_v2/tests/seed.py \
  src/dashboard_v2/.gitignore \
  tests/test_backend_coverage.py

echo '=== Test changes ==='
git diff --no-ext-diff --unified=35 "$base"...HEAD -- \
  src/dashboard_v2/tests/links.spec.ts \
  src/dashboard_v2/tests/memories.spec.ts \
  src/dashboard_v2/tests/metrics.spec.ts \
  src/dashboard_v2/tests/query.spec.ts \
  src/dashboard_v2/tests/review.spec.ts \
  src/dashboard_v2/tests/sessions.spec.ts \
  src/dashboard_v2/tests/shell.spec.ts \
  src/dashboard_v2/tests/visual.spec.ts

Length of output: 118728

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo '=== Candidate file locations ==='
fd -t f . src tests | rg '(^|/)(server|__main__|metrics|heatmap|handler|app|home|seed)\.(py|svelte|ts)$|dashboard'

echo '=== Structural outlines ==='
for f in \
  src/dashboard_v2/tests/seed.py \
  src/dashboard_v2/src/routes/metrics/+page.svelte \
  src/dashboard_v2/src/routes/memories/+page.svelte \
  src/lorekeeper/dashboard/handler.py \
  src/lorekeeper/dashboard/app.py \
  src/lorekeeper/server.py; do
  if [ -f "$f" ]; then
    echo "--- $f ---"
    ast-grep outline "$f" --items all 2>/dev/null | head -n 220 || true
  fi
done

echo '=== Metrics/heatmap implementation and writes ==='
rg -n -i -C 5 \
  'hm-cell|heatmap|tool.?call|record.*metric|MetricsStore|metrics_store|tabindex' \
  src/dashboard_v2/src src/lorekeeper tests \
  -g '*.{ts,svelte,py}'

echo '=== Dashboard CLI and port behavior ==='
rg -n -C 5 \
  'ArgumentParser|add_argument\(.+port|uvicorn|def main\(|__main__|LORE_DASH_RELOAD' \
  src/lorekeeper -g '*.py'

echo '=== Memory list/detail serialization and ID fields ==='
rg -n -C 6 \
  'serialize_memory|list_memories_paginated|def get_memory|lore_id|["'\'']id["'\'']' \
  src/lorekeeper/dashboard src/lorekeeper -g '*.py'

Length of output: 193570

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

echo '=== Dashboard memory update path ==='
rg -n -A30 -B8 'def update_memory|update_memory_fields|def update_memory' \
  src/lorekeeper/dashboard/handler.py \
  src/lorekeeper/processors/memory.py \
  src/lorekeeper/domains/memory \
  src/lorekeeper -g '*.py'

echo '=== Drawer edit fields and save payload ==='
rg -n -A18 -B12 \
  'onSave|edited|content|description|drawer-title|save' \
  src/dashboard_v2/src/lib/components/overlays/MemoryDetailDrawer.svelte \
  src/dashboard_v2/src/lib/components/overlays/types.ts \
  src/dashboard_v2/src/lib/api/memories.ts

echo '=== Metrics page and interactive heatmap ==='
fd -t f . src/dashboard_v2/src | rg '(metrics|heatmap|Activity).*\.(svelte|ts)$' | while read -r f; do
  echo "--- $f ---"
  rg -n -A12 -B8 'hm-cell|role=.button|tabindex|tooltip|tool-calls|fetch' "$f" || true
done

echo '=== Metrics written by seeded operations ==='
rg -n -A8 -B8 'increment_metric|increment_metric_safe' \
  src/lorekeeper/processors/reflection.py \
  src/lorekeeper/processors/memory.py \
  src/lorekeeper/domains/reflection/service.py \
  src/lorekeeper/domains/suggestion/sweep.py \
  src/lorekeeper/platform/metrics/repository.py

echo '=== Exact changed-line locations ==='
nl -ba src/dashboard_v2/src/routes/memories/+page.svelte | sed -n '95,175p'
nl -ba src/lorekeeper/dashboard/handler.py | sed -n '124,150p'
nl -ba src/lorekeeper/domains/memory/repository.py | sed -n '195,230p'

Length of output: 126454

---

@Jessinra, I found one medium-severity issue and no major issues in the reviewed changes.

• Medium — memory edits/deletes leave live search stale. The newly wired drawer now PATCHes editable title, description, and content, but MemoryWriteService.update_memory_fields() only updates SQLite and commits. It does not invalidate/rebuild the in-memory memory/BM25 cache or replace the corresponding LanceDB vector. Consequently, a user can save changed content and the table/detail API reflects it, while Query still returns/ranks it using the old indexed content until a server restart. The newly enabled permanent delete has the same problem: it deletes only the SQL row without removing the vector or invalidating cache state, so a previously cached/indexed memory can remain searchable. Refresh the cache and update/remove the vector as part of these write paths, and add an E2E/API regression test that edits content (and deletes a memory) then verifies search no longer returns the stale representation.

[Copilot]

Pull request overview

This PR replaces the previous "false-confidence" Dashboard V2 Playwright setup (preview-mode + test.skip-on-empty guards) with a real seeded backend + live frontend E2E suite: the browser hits the Vite dev origin, which proxies /api/* to a FastAPI backend seeded with deterministic fixtures. Wiring up a genuinely functional stack surfaced (and this PR fixes) five latent product regressions in the dashboard shell, memory write path, and a11y — the changes therefore touch core product code, not just test infra.

Changes:

• Test infrastructure: new seed.py fixture seeder, dual webServer in playwright.config.ts, Vite /api proxy, offset ports (7787/7788), hydration/cold-start hardening, skip-guards → hard assertions, and a new playwright-dashboard CI job.
• Product regression fixes: restored the overwritten AppShell (children/TopBar/Toast/CommandPalette), fixed replaceState crashes (URL-object clone + pre-router guards), wired the memory drawer save/delete round-trip with id/lore_id bridging, and corrected SessionDrawer from role="complementary" to role="dialog" aria-modal="true".
• Backend: added source_type to the MemoryUpdate model and the repository allowed-field set, with pytest coverage for the PATCH round-trip.

Review notes: The frontend fixes are consistent with existing patterns (SessionDrawer now matches MemoryDetailDrawer; the AppShell imports all resolve; fetchHealth().pending_suggestions feeds the live NavRail badge). CI action versions (@v7) match the rest of ci.yml. One moderate bug remains: the seed's explicit fixture links use invalid relation_type values, so they are silently dropped.

Reviewed changes

Copilot reviewed 27 out of 27 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/dashboard_v2/tests/seed.py	New in-process fixture seeder; explicit links use invalid relation types (related/supports) — silently dropped.
src/dashboard_v2/playwright.config.ts	Dual webServer (backend seed+serve, Vite dev proxy), offset ports.
src/dashboard_v2/vite.config.ts	/api dev proxy to the FastAPI backend, port overridable via env.
.github/workflows/ci.yml	New playwright-dashboard job (Python+Node+HF setup, npm ci, Playwright run, artifact upload).
src/dashboard_v2/src/lib/components/shell/AppShell.svelte	Restored full shell (children/TopBar/Toast/CommandPalette, hotkey, health badge).
src/dashboard_v2/src/lib/components/shell/NavRail.svelte	Adds pendingReview prop; overlays live Review badge count.
src/dashboard_v2/src/lib/constants/routes.ts	Removes hardcoded badge: 8 on the Review route.
src/dashboard_v2/src/routes/memories/+page.svelte	Wires drawer save/delete, bridges id/lore_id, fixes replaceState.
src/dashboard_v2/src/lib/api/memories.ts	Adds updateMemory/deleteMemory helpers.
src/dashboard_v2/src/routes/{links,sessions,review}/+page.svelte	Guards replaceState against pre-router crash.
src/dashboard_v2/src/lib/components/overlays/SessionDrawer.svelte	Modal drawer now role="dialog" aria-modal="true".
src/lorekeeper/dashboard/routes/memories.py	Adds source_type to MemoryUpdate.
src/lorekeeper/domains/memory/repository.py	Adds source_type to allowed update fields.
tests/test_backend_coverage.py	New PATCH round-trip tests (title, source_type, 404).
src/dashboard_v2/tests/*.spec.ts	Skip-guards → hard assertions, palette modifier, query/hydration timing.
src/dashboard_v2/tests/visual.spec.ts + .gitignore	Gate @visual behind RUN_VISUAL=1; un-ignore baseline dir.
docs/plans/...md, backlogs/ready/LKPR-140...md	Planning + backlog docs (describe original 7777/7778 + globalSetup approach).