Files
jobtrackingapp/docs/cv-integration-qa/END_TO_END_JOURNEYS.md
T

17 lines
2.7 KiB
Markdown

# Connected journeys — 2026-10-27
`real-browser-01.json` records six successful Chromium cases, zero retries: two retained import journeys and four new CV-QA journeys. The actual local pinned sidecar processed uploaded synthetic PDF/DOCX files; no extraction/model mocks or live generative inference were used. Each new journey checks the persisted parser identity, owned source SHA-256, `no-store`, foreign-owner source/review 404s, unchanged canonical data before approval and browser-refresh recovery. Stage attachments contain sampled real persisted states; stages completing between polls are not claimed as observed.
| Journey | Deliberate review and downstream proof |
|---|---|
| EN/light desktop, ambiguous employer-first DOCX | Keep uncertain original proposals, author a source-backed job, apply and independently read canonical/history; explicit existing destination and stale two-session comparison recovery; CV-only title override and PDF; submitted package/manual cover-letter snapshot survives later profile/CV changes. |
| Bokmål/dark 390px, title-first PDF | Same source-backed job authoring and canonical verification, technical CV customization/export; locale and responsive review. |
| EN/light 768px, ambiguous education DOCX | Author missing education from source, persist, change CV-only education title, preserve canonical qualification and export minimal PDF. |
| EN/dark 768px, repeated overlapping roles DOCX | Approve three resolved records with distinct source occurrences, preserve literal date precision/current state, verify same-file reimport/replay adds no records, customize one CV role and export technical PDF. |
The tests reuse actual Career Profile, editor, variant, revision, application association, letter and submitted-package contracts. No alternate profile/revision storage was introduced. Existing import cases separately exercise correction of extracted fields through profile/editor/PDF.
The original source and mapped review are captured in `mapped-review.png`; final editor previews in `tailored-preview.png`. Six actual exports were additionally read with pypdf and rendered with Poppler, with file hashes and text retained in `.artifacts/cv-integration-qa/imported-pdf-proof-01/results.json`. Browser assertions verify expected CV-only markers and preserved canonical values. These short imported documents complement, and do not replace, the original multi-page fidelity corpus.
Coverage limits: local automation does not establish physical-device behavior, real user correction burden, human language quality, native PDF viewer printing or real model semantics. Full shared-functionality and PDF preservation results are recorded in their separate reports.