Add baseline versioning and comparison foundation

[godtroop-github]

Goal

Add a baseline management foundation so completed OCR consensus tasks can be saved as versioned baseline snapshots and later compared against new task results.

Scope

• Save completed task output as a baseline snapshot.
• Store baseline metadata, basic records, evidence summaries, and employment rows.
• Provide baseline list and detail APIs.
• Document the baseline/versioning/diff design for later implementation phases.

Out of scope for this first slice

• Full baseline diff UI.
• Excel export.
• Candidate/golden promotion flow.
• Manual review workflow for accepting or rejecting differences.

Privacy note

This issue intentionally excludes private sample data, OCR outputs, server paths, environment details, names, identifiers, and enterprise information.

[godtroop-github]

Progress update:

• Added the first UI test point for baseline management.
• The processing workbench now links to baseline management.
• Completed tasks can be saved as snapshot baselines from the task list.
• Added a baseline management page for listing baselines and inspecting baseline details.
• Added /baselines route.

Still out of scope for this slice: full diff UI, Excel export, and candidate/golden promotion flow.

[godtroop-github]

Progress update:

• Moved the baseline save action from the processing workbench to the generated result list page.
• The result list now supports saving the current task as a snapshot baseline after manual review.
• The save request includes browser-side manual review state so baseline records can persist passed/failed/unreviewed status.
• The processing workbench keeps navigation to baseline management only.

Next implementation slice remains baseline comparison/diff.

[godtroop-github]

Progress update:

• Added the first baseline comparison API: POST /api/compare.
• Supports comparing a completed task against a selected baseline.
• Current diff scope covers basic business fields: name, masked identifier, related count, employment count, and shareholding count.
• Capture date/time is handled as weak collection-time difference and does not automatically become a business diff.
• The response includes record-level summary, field-level diffs, severity, and review-required flags.

Still out of scope for this slice: employment-row diff UI, persisted compare jobs, Excel export, and candidate/golden promotion flow.

[godtroop-github]

Progress update:

• Added the first page-level baseline comparison test point on /baselines.
• Users can now select a baseline and a completed task, then run POST /api/compare from the UI.
• The page shows summary metrics and a focused difference table for basic fields.
• Capture time remains a weak collection-time difference and is displayed separately from business field changes.

Still out of scope: persisted compare jobs, employment-row diff, Excel export, and candidate/golden promotion.

[godtroop-github]

Progress update:

• Improved baseline comparison visualization on /baselines.
• Added change composition bar and risk distribution.
• Added quick slices for all, review-required, business changes, added records, missing records, older screenshots, and capture-time-only changes.
• Enhanced field diffs with side-by-side baseline/current value cards.

Still out of scope: persisted compare jobs, employment-row diff, Excel export, and candidate/golden promotion.

[godtroop-github]

Progress update:

• Added detailed filters to baseline comparison: search, business status, capture-time status, severity, and review-required only.
• Added per-record expandable baseline/current detail panels.
• Field-level changes remain shown as side-by-side baseline/current value cards.
• Empty filter results now show an explicit empty state instead of falling back to default rows.

Still out of scope: persisted compare jobs, employment-row diff, Excel export, and candidate/golden promotion.

[godtroop-github]

Baseline compare update: separated baseline detail from compare results, removed first-record preview, added baseline detail tables, and extended compare rules to field-level employment details. Current task-vs-baseline compare now covers basic fields plus company name, business status, role, and share ratio for employment rows. Employment changes are counted separately and included in review filtering.

[godtroop-github]

Baseline compare redesign update: changed the compare flow into a review workspace. The compare list now surfaces concrete basic-field and employment-detail diff summaries first. Selecting a row shows image previews plus the baseline/current consensus structured records only; raw OCR text, method confidence, and model-level evidence are intentionally excluded from this baseline audit flow. Spec updated to separate baseline version browsing from diff review.

[godtroop-github]

Baseline compare update: removed the baseline-detail section from the main page to prioritize the compare workspace. The selected-record pane now shows baseline-batch images and current-batch images with independent navigation. Compare output now includes image_added/image_missing diffs and a quick filter/summary for image changes, while still comparing only reviewed consensus structured data for basic fields and employment details.

[godtroop-github]

Baseline compare refinement: capture date/time-only changes now remain info-level and do not require review. The compare workspace has been changed from a wide table into a diff-card list plus selected-record detail pane. Employment comparison now aligns at the company level first, then expands each company to field-by-field baseline/current comparisons.

[godtroop-github]

Baseline compare bugfix update: source-task self-compare now short-circuits to compare the baseline snapshot with itself, so it should produce zero business/image/review diffs. Employment row attachment was hardened with fallback matching by person key, file/name, and source image. Image diffs are now conservative and normalized by decoded filename to reduce false added/missing reports. Baseline list selection and compare dropdown now share one selected baseline state.

[godtroop-github]

Baseline compare update: added swap support between baseline and compared task. Swap resolves the compared task to its saved baseline and uses the current baseline's source task as the new compared task. Employment detail loading was hardened for older task/baseline formats: rows can now be extracted from multiple payload keys and field values can fall back to evidence_summary aliases for company name, status, role, share ratio, and source image.

[godtroop-github]

Baseline compare rule fix: added/missing records no longer produce field-level basic/employment comparisons; they are treated as record-level changes with one-sided consensus details. Image added/missing diffs were removed from compare semantics and images are now preview evidence only. Employment rows now attach to all records sharing the same person key in multi-image cases, instead of trusting a single baseline_record_id, which fixes cases where employment data existed but appeared empty on another image record for the same person.

[godtroop-github]

Baseline audit refinement: investigated cases where employment details looked empty. The data existed in baseline storage; the failing path was exact company-key matching and image-level attachment in multi-image/person cases. Added conservative fuzzy company matching to avoid false missing+added rows when OCR truncates a company suffix. Added/missing records no longer require review. Image add/missing semantics remain removed; images are evidence preview only. Layout changed to a top diff list plus full-width selected-record review pane so image preview and field confirmation have more room.