# P8-00 — Baseline and architecture

Completed 21 October 2026. Entry `ef94d579cbfec62d1ac38346f556f80276cee95a`, clean working tree, local and direct remote `codex/ui-redesign-phase4` equal. Requested Phase 8 branch did not exist locally or remotely; created `feature/phase8-dashboard-customisation` from that entry. Planning checkpoint `421f902d21c08a814de12d7c32bc6ab8d4c01e15` pushed normally; `git ls-remote` directly confirms it. Main remains outside this work.

Reviewed Phase 7 final acceptance and Phase 8 handoff, Phase 6 handoff, original implementation plan, dashboard/Phase 3 architecture, design system, responsive/accessibility, guardrails, information architecture, flows and master spec. The current implementation confirms only five local visibility flags; attention, stage-age, source-response and salary sections were outside them. The new nine-entry [catalogue and design](../README.md) preserves these sections and all existing data owners.

## Fresh starting tests

| Command | Result | Evidence |
|---|---|---|
| Pinned .NET `test JobTrackerApi.Tests/JobTrackerApi.Tests.csproj --configuration Release --no-restore` | 1,060 passed; 0 failed, skipped; 16s test duration | `.artifacts/phase8/baseline/p8-baseline.trx`, `baseline-backend.log` |
| UI `npm test -- --runInBand --json --outputFile=../.artifacts/phase8/baseline-frontend.json` | 899 passed / 101 suites; 0 failed, pending; 284.546s | `.artifacts/phase8/baseline-frontend.json`, `baseline-frontend.log` |

These executions precede product source changes. PowerShell's native stderr wrapping is retained in the raw log; Jest's result JSON records the actual outcomes. Historic Phase 7 browser351, separate23 and sidecar56 are inherited evidence, not fresh Phase 8 runs.

## Prerequisites and decisions

Account persistence is a necessary extension explicitly approved by the current brief, superseding Phase 3's proposed device-first option. Extend ApplicationUser with two additive fields and a narrow owner-only GET/PUT. Compare-and-update revision in one SQL statement; avoid whole-Identity updates for layout saves. Supported sizes are standard/full with a fixed full-width summary, one logical DOM order and single-column mobile. Reuse installed native drag helper and touch/keyboard move buttons; no dependency comparison/download needed.

The known failure investigation reviewed P7's deterministic held-reader reproduction against DELETE journaling, its WAL-before-bootstrap repair, and finite passing regression. Add actual layout reader/write and competing-save tests on that baseline. Do not infer permanent immunity to writer contention or identify the historical browser's exact competing connection.

No mandatory missing dependency blocks safe development. Retained conditions remain: real MariaDB runtime, human screen reader/physical devices/complete enlargement, other-browser/output and production release checks. Production NO-GO, P5-13 withheld. The [risk register](../README.md) carries these forward.