Neon-Desk/E2E_UX_REPORT_2026-03-14.md

E2E and UX Report

Date: 2026-03-14 Repo: fiscal clone Command run: bun run test:e2e

Executive Summary

The full Playwright suite did not pass. Result: 4 passed, 6 failed, 1 did not run.

The most important product issues surfaced by the run and manual pass were:

1. The overview workflow can land in a non-recovering loading state with no actionable guidance.
2. Auth/session reliability appears unstable under realistic navigation and parallel activity.
3. Filing sync creates noisy, duplicated notifications that follow the user across pages.
4. The filings workflow communicates one filter state while showing data from multiple tickers.

Automated E2E Outcome

Artifacts are under output/playwright/.

Failed specs:

• e2e/analysis.spec.ts: shows the overview skeleton while analysis is loading

  • Failure: timed out waiting for Microsoft Corporation.
  • Observed state: app chrome rendered, but the page remained on Loading company overview.
  • Artifacts:
    • output/playwright/test-results/analysis-shows-the-overvie-6b9b1-n-while-analysis-is-loading/error-context.md
    • output/playwright/test-results/analysis-shows-the-overvie-6b9b1-n-while-analysis-is-loading/test-failed-1.png

• e2e/financials.spec.ts: renders the standardized operating expense tree and inspector details

  • Failure: timed out waiting for Normalization Summary.
  • Observed state: page stayed in a loading shell in the failing run.
  • Artifacts:
    • output/playwright/test-results/financials-renders-the-sta-7a51f--tree-and-inspector-details/error-context.md
    • output/playwright/test-results/financials-renders-the-sta-7a51f--tree-and-inspector-details/test-failed-1.png

• e2e/financials.spec.ts: shows not meaningful expense breakdown rows for bank pack filings

  • Failure: sign-up flow never reached Command Center.
  • Observed state: browser ended on Secure Sign In.
  • Inference: auth/session creation or post-signup redirect is intermittently failing.
  • Artifacts:
    • output/playwright/test-results/financials-shows-not-meani-c9d8d--rows-for-bank-pack-filings/error-context.md
    • output/playwright/test-results/financials-shows-not-meani-c9d8d--rows-for-bank-pack-filings/test-failed-1.png

• e2e/graphing.spec.ts: supports graphing compare controls and partial failures

  • Failure: sign-up flow never reached Command Center.
  • Observed state: browser ended on Secure Sign In.

• e2e/graphing.spec.ts: distinguishes not meaningful metrics from missing data in the latest values table

  • Failure: sign-up flow never reached Command Center.
  • Observed state: browser ended on Secure Sign In.

• e2e/research-mvp.spec.ts: supports the core coverage-to-research workflow

  • Failure: strict-mode locator conflict on NVDA status.
  • Observed state: the page exposed two matching controls for the same accessible name.
  • Inference: duplicated DOM or duplicated accessible controls in the coverage row.
  • Artifacts:
    • output/playwright/test-results/research-mvp-supports-the-core-coverage-to-research-workflow/error-context.md
    • output/playwright/test-results/research-mvp-supports-the-core-coverage-to-research-workflow/test-failed-1.png

Runtime notes from the run:

• sqlite-vec fell back to table-backed storage repeatedly.
• workflow-local reported concurrency limit (1) reached.
• Live server logs repeatedly failed background filing workflows with:
  • SQLiteError: table filing_taxonomy_snapshot has no column named parser_engine
  • Inference: the local/e2e database schema is behind the code path used by filing taxonomy persistence.
• The suite completed in about 1.2 minutes before failing.

Manual Workflow Findings

Critical

• Overview page does not distinguish loading vs failure vs empty data.

  • Visiting /analysis rendered the full page shell but stayed on Loading company overview.
  • The user gets no timeout, error explanation, retry guidance, or indication that background filing tasks failed.
  • The local server logs showed repeated filing workflow failures caused by a schema mismatch on filing_taxonomy_snapshot.parser_engine, which likely contributes directly to this broken state.
  • This is a broken primary workflow, not just a cosmetic state issue.

• Auth confidence is too low.

  • Three failed specs fell back to Secure Sign In immediately after supposed account creation.
  • A user-facing version of this would feel like silent account creation failure or dropped session state.

• Filings filter messaging is misleading.

  • The page header said 100 records loaded for NVDA, but the ledger visibly contained both NVDA and MSFT rows.
  • This breaks user trust because the UI claims a scoped result set while showing cross-ticker data.

High

• Notification spam degrades every workflow.

  • Coverage creation and navigation produced stacked Filing sync toasts.
  • Multiple toasts showed nearly identical states and remained visible across pages.
  • Because failed background jobs keep retrying or re-reporting progress, the user sees noise instead of one durable status surface.
  • The notifications panel becomes ambient noise instead of actionable feedback.

• Coverage actions can trigger expensive side effects without clear consent.

  • Adding one coverage record immediately started sync-related background activity.
  • There is no up-front explanation of what will be fetched, how long it may take, or whether it can be deferred.

• Research workspace has high cognitive load on first entry.

  • Filters, note capture, uploads, memo editing, and packet review all appear at once.
  • The page is powerful, but the first-time path is unclear and there is no obvious recommended sequence.

Medium

• Empty states on Financials and Graphing are under-instructive.

  • Financials: No rows available, No trend data, and zeroed normalization metrics do not tell the user whether they need to sync, change ticker, or wait.
  • Graphing: No chart data available appears after opening a prefilled compare set, which makes the default state feel broken.

• Upload interaction in Research is confusing.

  • The page exposes two controls with the label Upload file, one active and one disabled.
  • That is confusing visually and poor for keyboard/screen-reader users.

• Navigation context is inconsistent.

  • Some links carry ticker context and others drop back to untickered routes such as /analysis.
  • The app often compensates by defaulting to MSFT, but that is implicit and can feel arbitrary.

• Filings table density is too high by default.

  • Loading 100 records into a single ledger creates a long scan path before the user can find the next action.
  • Pagination, grouped sections, or stronger filter summaries would reduce cognitive load.

Low

• favicon.ico returns 404.
• The boot splash and auth handoff add a short delay before the real screen appears, but do not communicate whether the app is loading, redirecting, or checking session state.

Suggested Fix Order

1. Stabilize post-signup session creation and redirect behavior.
2. Make /analysis fail loudly and recoverably instead of hanging in a loader.
3. Fix filings filter truthfulness so the page never claims NVDA while rendering mixed results.
4. Deduplicate or batch filing sync notifications.
5. Add next-step empty states to Financials and Graphing.
6. Reduce first-run complexity in Research with a guided progression or collapsed sections.
7. Remove duplicate accessible controls and labels in Coverage and Research upload flows.

Notes

The manual pass used the repo's e2e server entrypoint: bun run e2e:webserver.