feat(studies-ui): restore lost Trials + Convergence list columns

[SoundMindsAI]

Summary

Adds Trials + Convergence columns to the /studies list table — answering the operator request to surface trial count on the studies page.

This is a restoration of lost work, not a new feature. The backend already returns trial_count + convergence_verdict on StudySummary (shipped in feat_studies_convergence_visibility, PR #421, Story 1.1 / b90d5477). The matching frontend columns (Story 1.2, commit ed5ca276) were authored + reviewed but never reached main — they were dropped in the PR #421/#422 rebase that de-duplicated Epic 1 commits. So the two fields sat returned-but-unsurfaced; the /studies list ran with only its original 6 columns.

This PR cherry-picks ed5ca276 to restore the reviewed original (better than a fresh reimplementation: header tooltipKey pattern, hideable, satisfies Record<ConvergenceVerdict> exhaustiveness, the study.trial_count glossary key, a dedicated unit test, + an E2E spec):

• Trials column — trial_count (the non-baseline count, resolved server-side via the batched count_trials_for_studies aggregate — no N+1).
• Convergence column — convergence_verdict → Converged / Improving / Too few trials badge, em-dash when null (in-flight / <5 complete trials). Reuses CONVERGENCE_VERDICT_VALUES + the convergence_verdict glossary key.

Bug caught during restoration

The header tooltipKey renders an <InfoTooltip> in the column header, whose Radix Tooltip needs a TooltipProvider ancestor. The app shell (layout.tsx) provides one in prod, but page.test.tsx renders <StudiesPage> in isolation — so it now wraps in TooltipProvider, mirroring the layout. That ed5ca276 didn't already carry this fix confirms it never ran the full suite on main (consistent with being lost pre-merge).

Doc-drift corrected

state_history.md and the feat_studies_convergence_visibility implementation plan both marked Story 1.2 as shipped via PR #421. They didn't. Both now carry a CORRECTION (2026-06-03) annotation pointing here.

Guide regen

Guide 06's 01-studies-list.png was stale (old 6-column list). Regenerated against the populated stack — now shows the TRIALS column (50/200/200/15/…) + the CONVERGENCE column (CONVERGED / TOO FEW TRIALS badges). 03-create-study-modal + 04-study-detail regenerated in lockstep; walkthrough video promoted.

Test coverage

Layer	File	Cases
UI vitest	studies-table-convergence-column.test.tsx (restored)	7
UI vitest	studies-table-ceiling-badge.test.tsx (fixture extended)	5
UI vitest	studies/page.test.tsx (TooltipProvider + mock fields)	5
E2E	studies-convergence-columns.spec.ts (restored)	smoke-gated

1039 vitest green; tsc + build clean; freshness gate clean (no backend change). No migration.

Test plan

• pnpm --dir ui test → 1039/1039
• pnpm --dir ui typecheck + pnpm --dir ui build clean
• pnpm --dir ui lint → 0 errors
• Freshness gate: git status clean after canonical regen
• Guide 06 regenerated against populated stack (columns visible with real data)
• CI green

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request restores the frontend 'Trials' and 'Convergence' columns in the /studies list that were lost during a previous rebase. It adds the column configurations, glossary tooltips, unit tests, and an E2E test, while also wrapping the page in a TooltipProvider to prevent Radix tooltip errors. The review feedback suggests adding a defensive check in the column configuration to handle unmapped convergence verdicts gracefully during rolling deployments, and importing ReactNode explicitly in the new test file to avoid referencing the global React namespace.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

During rolling deployments or API updates, the backend might return a new or unexpected convergence_verdict value that is not yet mapped in VERDICT_BADGE. To prevent runtime crashes (e.g., TypeError: Cannot read properties of undefined (reading 'variant')), add a defensive check to handle cases where badge is undefined.

      const badge = VERDICT_BADGE[verdict];
      if (!badge) return <span className="text-muted-foreground">—</span>;
      return (
        <span
          className="inline-flex items-center"
          data-testid={`convergence-verdict-${row.original.id}`}
          data-verdict={verdict}
        >
          <Badge variant={badge.variant} className="text-[10px] uppercase tracking-wide">
            {badge.label}
          </Badge>
        </span>
      );

[gemini-code-assist]

Import ReactNode from 'react' to type-safe the cell renderer return type and avoid referencing the global React namespace.

Suggested change

	import { render, screen } from '@testing-library/react';
	import { describe, expect, it } from 'vitest';
	import { render, screen } from '@testing-library/react';
	import { type ReactNode } from 'react';
	import { describe, expect, it } from 'vitest';

[gemini-code-assist]

Use the imported ReactNode type directly instead of referencing React.ReactNode without React in scope.

Suggested change

	const cell = column.cell as (ctx: { row: { original: StudySummary } }) => React.ReactNode;
	const cell = column.cell as (ctx: { row: { original: StudySummary } }) => ReactNode;

[SoundMindsAI]

Review adjudication (Gemini Code Assist)

Commit landing accepted fixes: 85eb14fb

Gemini Code Assist (3 findings — all accepted)

#	Sev	Location	Verdict	Notes
1	Medium	studies-table.column-config.tsx:172	Accepted	Forward-compat guard if (!badge) return <em-dash> for an unmapped convergence_verdict. Distinct from the declared_params or {} finding I rejected on PR #436: that was a NOT-NULL DB column → unreachable. This is REACHABLE — convergence_verdict is a backend-computed classification (backend/app/domain/study/convergence.py), not a fixed DB enum, so a newer backend could emit a verdict this frontend snapshot doesn't map during a rolling deploy, and badge.variant on the undefined lookup would crash the whole table render. Matches the codebase's existing forward-compat pattern (StatusBadge VARIANT_TABLE[...]?.[value] ?? 'secondary'; the best_metric column's rolling-deploy backward-compat comment in this same file). Added a regression test (cast an out-of-union verdict → asserts em-dash fallback, no crash).
2	Medium	studies-table-convergence-column.test.tsx:18	Accepted	Import type ReactNode from 'react' explicitly rather than referencing the ambient React.ReactNode namespace. Matches the rest of the codebase's import style.
3	Medium	studies-table-convergence-column.test.tsx:45	Accepted	Use the imported ReactNode directly (companion to #2).

Outcomes

• Applied fixes (3): forward-compat badge guard + regression test; explicit ReactNode import (×2).
• Rejected (0) · Deferred (0)

Note: this PR also corrects a doc-drift bug it uncovered — the studies-list columns were marked "shipped" in PR #421 but were actually lost in the #421/#422 rebase (see PR description + the CORRECTION annotations in state_history.md + the feature plan).

All 17 CI checks were green on bddf3570; the 85eb14fb fixes are frontend-only (re-running). Ready for human review + merge once green.

[SoundMindsAI]

Final cross-model review (GPT-5.5) — adjudication

GPT-5.5 returned 1 Low-severity finding, rejected:

#	Sev	Location	Verdict	Notes
1	Low	studies-table.column-config.tsx (badge lookup)	Rejected	"A verdict named __proto__/constructor/toString returns a truthy inherited method from VERDICT_BADGE[verdict], bypassing the if (!badge) guard." Unreachable in practice: convergence_verdict is a backend-computed classification with three fixed branches (backend/app/domain/study/convergence.py → converged/still_improving/too_few_trials) — it never emits a JS-prototype-method name. The failure mode is a cosmetic empty badge, not a crash. And the codebase's own StatusBadge uses the identical plain-index pattern (VARIANT_TABLE[...]?.[value] ?? 'secondary') with the same characteristic — hardening only here with Object.hasOwn would diverge from that precedent. If prototype-safe enum lookups are desired, that's a codebase-wide convention change (StatusBadge included), not a one-off in this PR. The accepted Gemini #1 guard already handles the reachable case (a real new verdict value → undefined → em-dash).

Outcomes (all reviews on this PR)

• Applied fixes (3): Gemini docs: reorganize into numbered IA, add pipeline skills and feature templates #1 (forward-compat badge guard + regression test) + Gemini docs: MVP1 architecture + 12 feature specs + GPT-5.5 review fixes #2/chore(pipeline): add project-wide priority-ordered status mode #3 (explicit ReactNode import).
• Rejected with reasoning (1): GPT-5.5 docs: reorganize into numbered IA, add pipeline skills and feature templates #1 (unreachable prototype-name verdict; cosmetic; codebase-consistent as-is).

Ready for human review + merge once CI is green on 85eb14fb.