feat(data): Data screen per-row tabs, Results nodes, the Indicators move and Reprocess#1470

Closes #1150

Summary

The Data screen's per-row strip gains Extracted Text (the .txt sidecar, highlighted live from the selected Keyword Hit), Headers & TLS (from Capture.headers and tlsCertChain, labelled corroboration-only) and Manifest Ledger (the entries naming the Exhibit), each absent rather than empty when the row's kind has no data, and the Results nodes get their content: Keyword Hits with one child per Selector that filters the table to matched Exhibits with no snippet (X39), Integrity Exceptions with the three X37 buckets over Exhibits and Derived Files plus a sequential Verify all with progress, and the Manifest Ledger view with every entry typed, the chain verdict and the signer fingerprints read from manifest:snapshot and never computed in the renderer (X36, X25). Indicators keeps the IOC browser with Reprocess in its header (Q12). The MHTML Parts tab is omitted by maintainer decision on 2026-09-14; #991 owns it. Stacked on #1468 (#1149); the verification block below scores the whole stack against main.

Changes

• src/renderer/components/data/ledgerModel.ts: rows, naming, verdict and signer wording as pure functions of the snapshot
• src/renderer/components/data/ExtractedTextTab.tsx, HeadersTlsTab.tsx, ManifestLedger.tsx, IntegrityStrip.tsx: the tabs, the ledger view, the buckets strip
• src/renderer/components/data/dataTreeModel.ts, dataTableModel.ts: Keyword Hits children, keyword filtering, bucketForRow and integrityCounts over persisted and session verify results
• src/renderer/components/dashboard/cases/DataExplorer.tsx: tabs per kind, Results views, Verify all wiring
• src/renderer/lib/api/exhibits.ts, selectors.ts: useVerifyAll, selectorMatchingCapturesQueryOptions
• src/shared/types.ts: ExhibitVerification.derived and DerivedFileVerification, byte-identical to #1469 so the stack merges clean
• tests/components/DataExplorerResults.test.tsx (new), tests/renderer/components/ledgerModel.test.ts (new), model and screen test updates
• docs/agents/architecture.md: component list

Evidence-affecting

Yes

Evidence impact

• Evidentiary result or interpretation that could change: the screen now renders verification buckets, the chain verdict, signer fingerprints and per-entry hashes. An Exhibit's bucket comes from this session's exhibits:verify result when there is one and from the Capture's persisted state otherwise; a Derived File's comes only from its own outcome in the parent's session result, which the main process computed against the derivation entry, and without one it is unverified, with nothing inferred from the parent in the renderer (X36). A verify that throws is recorded as not completed and the row reads unverified for the session rather than its stale persisted state. legacy and verifier-too-old are filed as unverified, never as exceptions. "Chain intact through seq N" is shown only when the verdict says so, and a row's Manifest Ledger tab exists only when an entry this build can read names it.
• What verification proves and does not prove after this change: nothing about verification itself changes; Verify all is exhibits:verify per Exhibit in sequence. Session results for non-Capture Exhibits are not persisted and read as unverified on the next open. The derived list this screen reads is produced by #1469; on this branch alone every verify result has none and Derived Files read unverified.
• Backward verification preserved (existing evidence packages still verify): yes; no main-process code changes.
• Known-answer test added or extended, or why none applies: ledgerModel.test.ts pins the target wording of all nine entry types plus an unreadable line, the naming rule, the four verdict outcomes including too-old, and the signer wording; dataTableModel.test.ts pins the bucket rules including no inference for a Derived File; DataExplorerResults.test.tsx pins tab absence per kind (zero-byte text, unreadable-only ledger), the keyword filter with no snippet, that the second verify is not requested until the first resolves and the progress counter reads 0 / 2 while it waits, that a verify which throws re-buckets as unverified, and the too-old verdict rendered as its own outcome.

Verification

pnpm preflight at 0a239446a0e6ce061f316efbf0588cd4b9234e23 on Node v20.20.2; diff scored against origin/main (merge base c1ab6fb16). Result: pass.

• pnpm lint - pass (exit 0)
• pnpm lint:boundaries - pass (exit 0)
• pnpm lint:agents-md - pass (exit 0) - CLAUDE.md and AGENTS.md match
• pnpm typecheck - pass (exit 0)
• pnpm build - pass (exit 0)
• pnpm build:extension - skipped - no extension/ changes
• pnpm test:coverage - pass (exit 0) - 4173 passed, 10 skipped (4183), thresholds met
• pnpm coverage:diff - pass (exit 0) - 98.35% of 486 changed lines covered (floor 90%); 1 changed source file not instrumented

Pull request description generated by Claude Code