feat(lineage): add compact validation summaries

[gcko]

PR checklist

• Ensure you have added or ran the appropriate tests for your PR.
• DCO signed

What type of PR is this?

Feature

What this PR does / why we need it:

• Adds compact lineage-node summaries for value, profile, top-k, and histogram validation results.
• Selects the latest valid evidence deterministically and handles both persisted mappings and live Pydantic results.
• Exposes counts and per-column run IDs without embedding raw frames, rows, values, labels, or bins.
• Refreshes lineage aggregates after all supported terminal validation runs.
• Preserves existing row-count payloads and state-loader interfaces.

Which issue(s) this PR fixes:

Closes DRC-4260

Special notes for your reviewer:

• Stack position: 1 of 2 in the lineage stack.
• Base: main
• Child: feat(lineage): add conditional validation summary chip #1540
• GitHub stack: https://github.com/DataRecce/recce/stacks/1541
• Final focused backend/model suites: 65 passed.
• Full Python suite: 1,824 passed, 5 skipped.
• Full frontend suite: 4,344 passed, 5 skipped.
• Frontend lint/type checks, production build, repository flake8, targeted formatting, diff checks, and hooks passed.

Does this PR introduce a user-facing change?:

Yes.

Lineage run aggregates now include compact summaries for value, profile, top-k, and histogram validations without returning raw result payloads.

[codecov]

Codecov Report

❌ Patch coverage is 93.44262% with 12 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
recce/apis/run_func.py	88.78%	12 Missing ⚠️

Files with missing lines	Coverage Δ
tests/apis/test_run_func.py	98.46% <100.00%> (+0.36%)	⬆️
recce/apis/run_func.py	88.33% <88.78%> (+4.60%)	⬆️

... and 3 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[wcchang1115]

Code Review: PR #1539

SHA 8151e17 · Verdict GO

Blockers

None.

Issues

None.

Notes

1. recce/apis/run_func.py:513,514 — validation_summary publishes two sibling counters that count different units.

   result_count counts runs. Each per-column top_k_diff or histogram_diff run adds one (:508). difference_count counts columns inside the latest value_diff result (:478). The names give no hint of that difference. One value_diff run with 5 mismatched columns gives {'result_count': 1, 'difference_count': 5}, so difference_count exceeds result_count.

   difference_count also has no defined value for a node with no value_diff run. Line 472 leaves it at 0, which reads the same as "a value_diff ran and found nothing".

   No consumer depends on the field yet. PR feat(lineage): add conditional validation summary chip #1540 reads summary.types.value_diff.difference_count instead (GraphNodeOss.tsx:133). So this PR is the last point where a change costs nothing.

2. recce/apis/run_func.py:355 — The tie-break for recency is str(run.run_id), which is random.

   run_at has one-second resolution (recce/models/types.py:54). Two runs on the same model and column in the same second tie on time. The winner is then the higher UUID4. So latest_run_id can name the older run.

   tests/apis/test_run_func.py:154 and :182 lock this in. Both profile_diff runs carry run_at="2026-09-04T11:00:00Z", and line 243 asserts that the higher UUID wins.

   Millisecond resolution on run_at makes the tie-break real.

3. recce/apis/run_func.py:390 — _value_diff_difference_count reads matched_p at fixed position row[2].

   The payload carries the column names in data["columns"]. The code never uses them to find the index. A producer with a different column order gives a wrong count and no error.

[wcchang1115]

Reviewed at 8151e17. GO — no blockers, no issues. Three notes in #1539 (comment)