Skip to content
Crosswalker
🚧 Early alpha — building the foundation. See the roadmap →

v0.1.6 test-status update — Phase 3.6 signed off; Phase 2 / 3 / 3.5 untested manually

Created Updated

Why this log exists

Section titled “Why this log exists”

Mid-milestone status accounting. Phases keep shipping; manual testing is necessarily lagging (one human, multiple parallel build streams). This log answers the question a fresh agent or a future session would ask: what’s verified, and what’s still riding on assumptions?

The same v0.1.6 milestone page maintains a structured table of build + test status per phase. This log adds the context that doesn’t fit a table — the why behind each gap, and the manual flows + known issues that warrant attention.

Manual-test sign-offs since the last status log (2026-05-11)

Section titled “Manual-test sign-offs since the last status log (2026-05-11)”

Phase 1 — Recipe query: block schema ✅ Signed off

Section titled “Phase 1 — Recipe query: block schema ✅ Signed off”

User walked Scenario 1 (existing recipes still work). The 2026-05-11 wizard “0 notes generated” bug fix held — 12 notes correctly generated at Frameworks/<Family>/<ID>/<ID>.md for the NIST 800-53 sample CSV. Scenarios 2 (settings toggle), 3 (reference recipes validate), 4 (schema rejects malformed), 5 (codegen regenerates) covered by the agent on the user’s behalf via tests/recipe-query-block.test.ts (now 4 new edge-case tests added) + bun run codegen (zero drift confirmed).

Test guide updated 2026-05-15: TEST_PHASE1_QUERY_SCHEMA.md now correctly attributes scenarios 3 + 5 + edge cases to the automated suite, fixes the recipes/v0-1/... path issue in Scenario 3, and adds a “reload after every rebuild” pre-flight section.

Phase 3.6 — Wizard draft sessions ✅ Signed off

Section titled “Phase 3.6 — Wizard draft sessions ✅ Signed off”

Verified end-to-end on 2026-05-15:

  1. Empty state — opens wizard, no drafts exist → Step 1 shows “Drafts from previous sessions” section with the “No drafts yet” message ✅
  2. Start saving — picks CSV, advances to Step 2, toggles a column role → draft auto-saves after 500ms debounce ✅
  3. Abandon — Escape / X out of modal → onClose flush save fires ✅
  4. Resume visibility — reopens wizard → drafts section now shows the saved draft card with name, source file, “Step 2/4”, relative time, Resume + Delete actions ✅
  5. Resume — clicks Resume → wizard re-opens at Step 1 with a Notice “Source file needs to be re-selected” → user re-picks the same CSV → Step 2 shows the previously-toggled column role still applied ✅
  6. Complete — clicks through Step 3 → Step 4 → Generate → notes created → draft auto-deleted ✅
  7. Reopen — drafts section back to empty state ✅

Visual verification via WebDriver screenshots (tests/e2e/visual-wizard-step1-drafts.spec.ts): both empty and populated states render correctly.

Real UX gap caught + fixed during this verification: the original Phase 3.6c shipped drafts as a stacked modal that only appeared when drafts existed — first-time users had zero way to discover the feature. User feedback: “there should be something that shows like live sessions or drafted sessions or something like that and I don’t see that.” Commit 1cbc4f6 reworked the UX to put the drafts section directly in Step 1 with an always-visible empty state. This is now the shipped design.

Untested manual surface (the work that’s riding on assumptions)

Section titled “Untested manual surface (the work that’s riding on assumptions)”

The build pipeline is green across the board (bun run test → 243/243, bun run build clean, all phases load in test-vault). E2E tests pass. But the following user-facing surface hasn’t been manually exercised yet:

Phase 2 — SSSOM TSV import (covered by E2E only)

Section titled “Phase 2 — SSSOM TSV import (covered by E2E only)”

TEST_PHASE2_SSSOM_IMPORT.md has 7 scenarios that haven’t been walked. The data side has been verified end-to-end via the WebdriverIO suite (the SSSOM import-then-query flow in tests/e2e/sssom-import.spec.ts), but the UX-level concerns are unproven:

  • File picker + paste-TSV alternation in the SSSOM import modal
  • Predicate normalization warnings (SKOS → STRM) — does the Notice phrasing make sense to a user?
  • Confidence threshold setting respected when filtering edges
  • Re-import idempotency on the same file (no duplicate junction notes)
  • Bases-side query: do the imported junction notes show up in the pivot view (Phase 3 + Phase 2 integration)?

Phase 3 — crosswalkerPivot Bases view (covered by E2E only)

Section titled “Phase 3 — crosswalkerPivot Bases view (covered by E2E only)”

TEST_PHASE3_PIVOT_VIEW.md has 7 scenarios that haven’t been walked. The Component-subclass rendering + view-options panel have been smoke-tested via E2E, but:

  • Does the pivot table actually render in a .base file the user creates? (Or is it only reachable via the reference template?)
  • Heatmap mode color coding — does it match the documented intensity scale?
  • 8 view options (rows axis, cols axis, cell op, empty mode, etc.) — does changing them in the UI panel actually re-render?
  • Bases-disabled fallback Notice — does it say something the user can act on?
  • Sparse-pivot soft warning at >100K cells — only triggerable with a large fixture

Phase 3.5a + 3.5b — Wide-event logger + settings + commands (unexercised end-to-end)

Section titled “Phase 3.5a + 3.5b — Wide-event logger + settings + commands (unexercised end-to-end)”

Phase 3.5a’s API has 18 unit tests; the file format is verified to be valid NDJSON. But:

  • Have any imports/operations been done since 3.5a shipped that actually produced category-tagged events? Mostly no — the call sites still emit via the backward-compat shim (category: 'legacy'). Phase 3.5c will fix this.
  • The 3 new commands (Open debug log, Export debug log to clipboard, Clear debug log) — never run from the command palette
  • The category-filter UI (9 toggles in Settings → Debug → Category filters) — never toggled
  • verboseLogging setting — never enabled; trace-level events never observed
  • Log rotation at 5 MB — never triggered (logs are small; no way to verify yet without writing tens of thousands of events)
  • readForExport() secret redaction — pattern-based; only unit-tested with synthetic tokens

Phase 3.5c — Call-site sweep (not shipped)

Section titled “Phase 3.5c — Call-site sweep (not shipped)”

Still pending. Once shipped, every existing plugin.debug.log(...) call gets migrated to info('category', 'op', msg, data). category: 'legacy' events should disappear from the debug log entirely. Wizard, SSSOM importer, Tier 2 projector entry points each create a trace_id and thread it through their downstream events.

This is the cleanest “manual test” surface to address next, because:

  • It surfaces a visible improvement in the debug log (categorized events instead of legacy)
  • It validates the Phase 3.5a span helper end-to-end via real workloads
  • It’s pure refactor (no behavior change) so the risk is low

Bases-disabled scenarios (Phase 3 + 4 + 5 dependent)

Section titled “Bases-disabled scenarios (Phase 3 + 4 + 5 dependent)”

If a user has Obsidian’s Bases core plugin disabled, the registerCrosswalkerPivotView call should noop gracefully and the plugin should surface a Notice with a “Settings → Core plugins → Bases” hint. Has anyone actually run this with Bases disabled? Not yet. It’s a 5-minute manual test once the user has time.

Mobile parity

Section titled “Mobile parity”

v0.1.6 explicitly commits to Bases-on-mobile (the “binding constraint” per Ch 28). Nothing has been validated on iOS/Android Obsidian. Even just installing the plugin on mobile, opening a vault, and confirming nothing crashes would be valuable smoke coverage. Defer until v0.1.6 RC.

Known issues captured but not fixed

Section titled “Known issues captured but not fixed”

Filed in .workspace/2026-05-11-ux-feature-requests.md (feature requests) and now this log (bugs). All low-to-medium severity; not blocking the rest of v0.1.6 work.

IssueSeverityNotes
curie: "unknown:AC-1" instead of <frameworkId>:AC-1LowThe user typed nist-test as Framework Identifier in the wizard. The output frontmatter curie field should use that as the ontology prefix. The frameworkId field isn’t being threaded to the CURIE builder in generation-engine.ts:renderNoteFromRecipe. Fix: ~30 min — pass options.frameworkId into the CURIE construction.
related_controls: [[AC-2, AC-3, PM-9]] renders as ONE wikilink with commasMediumThe “Related Controls” column in the CSV contains comma-separated values like "AC-2, AC-3, PM-9". The link mechanism treats the entire string as one wikilink target, producing a broken link in Obsidian (the link target literally contains commas + spaces). Fix: ~45 min — split on a configurable delimiter (default ,) and emit N wikilinks. Settings already has defaultArrayHandling which is the right hook.
Typed-links preview missing in settings (feature request)Low (UX polish)Settings → Link Syntax preset selector doesn’t preview what each preset produces. Captured with ASCII mockup in .workspace/2026-05-11-ux-feature-requests.md. Probably ~half day. Phase 4 candidate.
Section titled “Recommended next actions”

In rough order of risk-adjusted value:

  1. Ship Phase 3.5c (call-site sweep) — pure refactor, ~half day. Surfaces real category-tagged events in the debug log, which makes diagnosing any future bug substantially easier. Low risk; high diagnostic payoff.
  2. Fix the 2 minor wizard bugs (CURIE + link splitting) — ~1.5 hours total. Cleans up the test-vault output so the rest of v0.1.6 validation isn’t muddied by wrong-looking notes.
  3. Phase 4 planning + ship — recipe-picker UX. Will reuse the draft-state pattern from Phase 3.6 (the design that lets users abandon + resume a flow). Per the v0.1.6 milestone plan, ~2 days.
  4. Phase 5 — opt-in materialization + sparse-pivot HARD guard. ~1.5 days.
  5. Mobile smoke check — defer to v0.1.6 RC; one focused 30-minute session.
  6. Manual UX walks of TEST_PHASE2_SSSOM_IMPORT.md + TEST_PHASE3_PIVOT_VIEW.md — defer until v0.1.6 RC; the E2E coverage holds for now.

Cumulative state after today

Section titled “Cumulative state after today”
SurfaceState
v0.1.6 Phase 1✅ Shipped + signed off
v0.1.6 Phase 1.5✅ Shipped + CI-covered
v0.1.6 Phase 2✅ Shipped (E2E covered; manual UX walks pending)
v0.1.6 Phase 3✅ Shipped (E2E covered; manual UX walks pending)
v0.1.6 Phase 3.5a (NDJSON logger)✅ Shipped (unit-tested; no real workload yet)
v0.1.6 Phase 3.5b (logger UI + commands)✅ Shipped (commands unexercised)
v0.1.6 Phase 3.5c (call-site sweep)🚧 Pending (next recommended)
v0.1.6 Phase 3.6 (draft sessions)✅ Shipped + signed off (full flow verified 2026-05-15)
v0.1.6 Phase 4 (recipe-picker UX)📋 Planned
v0.1.6 Phase 5 (materialization + sparse guard)📋 Planned
2 minor wizard bugs (CURIE + link splitting)⏸ Captured, not fixed
Mobile parity smoke check⏸ Defer to RC

bun run test: 243/243 pass. bun run build: clean. 8 commits shipped since 2026-05-11 across Phases 3.5a + 3.5b + 3.6a + 3.6b + 3.6c + the always-visible-drafts UX fix + minor polish.

Section titled “Related”