[paynejd]

What changed and why

In the grid's Target Code cell, the AI icon is purple when the mapped code is the OCL AI Assistant's pick and orange otherwise. When the AI picked a different code and the mapped code isn't one of its alternates either, the orange icon's tooltip said only "AI Analysis", so a reviewer had to open the row to see what the AI recommends.

That tooltip now names the AI's pick: "The AI recommends BA00.Z".

• getAIIndicatorTooltip (new, src/components/map-projects/viewBuilders.js) chooses the tooltip: "AI Recommended" when the mapped code is the AI's pick, "AI Alternate Candidate Suggestion" when it's one of the alternates, "The AI recommends {{code}}" when the AI picked another code, and "AI Analysis" when the analysis recommends no code at all.
• MapProject.jsx: getRowAIIndicatorMeta also returns the AI's primaryCandidateId (already worked out there), and the Target Code cell's tooltip calls the helper instead of the inline ternary. The icon's color doesn't change.
• New key map_project.ai_recommends_code in every catalog: en "The AI recommends {{code}}", es "La IA recomienda {{code}}", pt-BR "A IA recomenda {{code}}", zh "AI 推荐的是 {{code}}". The zh wording avoids reading as the existing "AI 推荐" badge.

How I tested it

Every run: step of .github/workflows/pr.yml, in order, on Node 22.23.2:

• start-prod.sh secret-variable check: passes.
• npm ci: passes.
• ESLint: passes (exit 0). This worktree is nested inside another checkout of the repo, so a bare eslint src --ext .jsx,.js also loaded the outer copy's .eslintrc.json and stopped with "couldn't determine the plugin react uniquely". I ran it as eslint --no-eslintrc -c .eslintrc.json src --ext .jsx,.js, which loads only this repo's one config, as CI does. The PR check below runs the exact command.
• npm run test:all: 658 unit tests and 11 integration tests pass (at 662ce17).

New tests in src/components/map-projects/__tests__/viewHelpers.test.js run getAIIndicatorTooltip through a real i18next instance loaded with each catalog, so they check the text a reviewer sees, not only the key. Run against main's viewBuilders.js and English catalog, the file fails (the helper doesn't exist there).

PR checks: Lint and test passed on 662ce17, running the exact ESLint command.

Browser check: I served the app with npm start on a free port and drove it with headless Chromium (Playwright, fresh context, 1440×900). Playwright answered every API and OCL AI Assistant call with stubs, so nothing left the machine. The stub is a saved three-row project whose AI analysis picks BA00.Z for a row mapped to BA00.0, picks 1A07.0 for a row mapped to its alternate 1A07.Y, and picks 1A02 for a row mapped to 1A02. Hovering each icon in the grid's Target Code cell gave:

Row	Icon	en	es	pt-BR	zh
mapped code is neither pick nor alternate	orange	The AI recommends BA00.Z	La IA recomienda BA00.Z	A IA recomenda BA00.Z	AI 推荐的是 BA00.Z
mapped code is an alternate	orange	AI Alternate Candidate Suggestion	Sugerencia de candidato alternativo de IA	Sugestão de Candidato Alternativo por IA	AI 备选候选建议
mapped code is the AI's pick	purple	AI Recommended	Recomendado por IA	Recomendado por IA	AI 推荐

No page errors. The ticket names no widths; the change is tooltip text only, in MUI's existing Tooltip, so I checked one desktop width.

Acceptance criteria

• The orange tooltip names the AI's recommended code, e.g. "The AI recommends BA00.Z". Test: getAIIndicatorTooltip: orange names the AI's pick; the alternate, purple and no-pick cases keep their wording checks the English text exactly; every catalog names the recommended code checks es, pt-BR and zh include the code. Seen in the browser check above, in all four languages.
• The alternate case keeps its own wording. The same test checks "AI Alternate Candidate Suggestion" for a mapped alternate even when the AI picked another code, and that the purple "AI Recommended" and the no-pick "AI Analysis" are unchanged. Seen in the browser check above.

Docs

No doc describes this tooltip (README, src/i18n/README.md and the __integration__ README don't mention it), so none needed updating. The helper carries a one-line comment saying what it's for.

Review

• Codex pass 1 at 60437ef: clean.
• Simplification review at 60437ef: three cuts (a redundant assertion, four English tests merged into one, a shorter comment), all applied in 662ce17.
• Codex check of 60437ef..662ce17: clean.

Chose not to do

• The wording follows the ticket's example ("The AI recommends …"). It names the code only, not the concept's name: the tooltip is on a cell that already shows the mapped concept, and the code is what the reviewer looks for in the candidates.
• The icon's colors and when it shows are unchanged.

Noticed, not fixed

• When the analysis recommends no code at all (no primary candidate) and the mapped code isn't an alternate, the orange tooltip still says "AI Analysis". Saying "The AI recommends no match" there would help a reviewer too, but it's a different case from this ticket's.
• getRowAIIndicatorMeta compares the AI's pick with targetConcept.id || targetConcept.reference.code. If a concept's id ever differed from its reference code, a mapped concept that is the AI's pick would show orange with its own code in the tooltip. I didn't find a case where they differ; the comparison is unchanged here.

closes OpenConceptLab/ocl_issues#2934

Written by an Octoad session worker (worker-v0.18, recipe lean-loop, flow public>public, run openconceptlab-ocl-issues-2934-20261110-183800).

🤖 Generated with Claude Code

https://claude.ai/code/session_01AaZ4qTi7WutsgsbAJ48PTj

[paynejd]

Codex review, pass 1 (codex-cli 0.167.1, commit 60437ef)

Clean.

findings.json

{"reviewer":"codex-cli 0.167.1","perspective":"defects","mode":"full","commit":"60437eff7d83c57fa3b6e9d72724dd7f286d6aba","verdict":"clean","prior":[],"findings":[]}

[paynejd]

Response to Codex pass 1

No findings to answer. Pass 1 leaves no P0 or P1 open, so the Codex loop stops here.

[paynejd]

Simplification review (commit 60437ef)

3 items: 1 dup-test, 2 shrink.

• S1 · dup-test · src/components/map-projects/tests/viewHelpers.test.js:312 · the assert.notEqual(tooltip, t('map_project.ai_analysis')) in the every-catalog loop, implied by the includes('BA00.Z') check before it → nothing replaces it (-1)
• S2 · shrink · src/components/map-projects/tests/viewHelpers.test.js:275 · four English tests that each build an i18next instance and repeat the whole meta object → one table-style test keeping all four assertions (-23)
• S3 · shrink · src/components/map-projects/viewBuilders.js:622 · the 6-line JSDoc on getAIIndicatorTooltip, which mostly restates the code and describes icon colors the function doesn't set → a one-line comment naming its purpose and the ticket (-5)

findings.json

{"reviewer":"claude subagent","perspective":"simplify","mode":"full","commit":"60437eff7d83c57fa3b6e9d72724dd7f286d6aba","verdict":"findings","prior":[],"findings":[{"id":"S1","severity":null,"tag":"dup-test","file":"src/components/map-projects/__tests__/viewHelpers.test.js","line":312,"claim":"In the 'every catalog names the recommended code' loop, `assert.notEqual(tooltip, t('map_project.ai_analysis'), locale)` adds nothing. The line before it already asserts `tooltip.includes('BA00.Z')`, and no catalog's ai_analysis string contains 'BA00.Z', so whenever the first assertion passes the second one passes too.","scenario":null,"action":"Nothing replaces it. Keep the `assert.ok(tooltip.includes('BA00.Z'), ...)` line.","lines_delta":-1,"confidence":"high"},{"id":"S2","severity":null,"tag":"shrink","file":"src/components/map-projects/__tests__/viewHelpers.test.js","line":275,"claim":"The four English tests (lines 275-305) each build a fresh i18next instance with `await translator('en')` and repeat the whole `{isPrimaryMatch, isAlternateMatch, primaryCandidateId}` object. Only one field changes per case.","scenario":null,"action":"Merge them into one table-style test that keeps all four assertions, including the AC1 and AC2 wording: `test('getAIIndicatorTooltip: orange names the AI\\'s pick; the alternate, purple and no-pick cases keep their wording', async () => { const t = await translator('en'); const tooltip = meta => getAIIndicatorTooltip({isPrimaryMatch: false, isAlternateMatch: false, primaryCandidateId: 'BA00.Z', ...meta}, t); assert.equal(tooltip({}), 'The AI recommends BA00.Z'); assert.equal(tooltip({isAlternateMatch: true}), 'AI Alternate Candidate Suggestion'); assert.equal(tooltip({isPrimaryMatch: true}), 'AI Recommended'); assert.equal(tooltip({primaryCandidateId: null}), 'AI Analysis') })`. Written one statement per line, that is about 8 lines in place of 31.","lines_delta":-23,"confidence":"medium"},{"id":"S3","severity":null,"tag":"shrink","file":"src/components/map-projects/viewBuilders.js","line":622,"claim":"The 6-line JSDoc on getAIIndicatorTooltip mostly says what the code or the call site already shows. Its purple/orange explanation describes the icon color, which MapProject.jsx sets and this function never touches. The parts worth keeping are what the function is for and the ticket link.","scenario":null,"action":"Replace the block with one line: `// The AI icon's tooltip in the grid's Target Code cell; the orange one names the AI's pick (OpenConceptLab/ocl_issues#2934).`","lines_delta":-5,"confidence":"low"}]}

[paynejd]

Response to the simplification review

• S1 applied in 662ce17
• S2 applied in 662ce17
• S3 applied in 662ce17

The full gate passes again after the cuts: secret-variable check, npm ci, ESLint, and npm run test:all (658 unit tests, 11 integration tests).

[paynejd]

Codex check of commits 60437ef..662ce17

(codex-cli 0.167.1, commit 662ce17)

Clean.

findings.json

{"reviewer":"codex-cli 0.167.1","perspective":"defects","mode":"final","commit":"662ce174c3f6da29e890150fe3c07a645f351cfc","verdict":"clean","prior":[],"findings":[]}