fix(mentorWriteRecommendations): Pin CSV columns so the feedback column is never dropped

src/activities/tasks/mentorWriteRecommendations.test.ts

@@ -0,0 +1,147 @@
+/**
+ * Offline unit tests for the mentor recommendations CSV serializer. No live DB,
+ * OpenAI, or Slack access — only the pure buildRecommendationsCsv function is
+ * exercised, with csv-parse used to round-trip the output.
+ *
+ * Run with:
+ *   npx ts-node src/activities/tasks/mentorWriteRecommendations.test.ts
+ */
+import 'reflect-metadata';
+import { parse } from 'csv-parse/sync';
+import { buildRecommendationsCsv } from './mentorWriteRecommendations';
+
+let failures = 0;
+
+function assertEqual<T>(actual: T, expected: T, message: string): void {
+  const a = JSON.stringify(actual);
+  const e = JSON.stringify(expected);
+  if (a !== e) {
+    failures += 1;
+    console.error(`FAILED: ${message}\n  expected: ${e}\n  actual:   ${a}`);
+  } else {
+    console.log(`PASSED: ${message}`);
+  }
+}
+
+const EXPECTED_COLUMNS = [
+  'mentor', 'linkedIn', 'Recommendation For', 'Pronouns',
+  'Count of Projects Mentored', 'Students Mentored', 'Count of Students Mentored',
+  'Mentorship Dates', 'What Students Had to Say', 'prompt', 'result',
+];
+
+// Mirrors the per-mentor `mentorInformation` shape built by the task, including the
+// conditional `What Students Had to Say` key that creates heterogeneous records.
+type MentorRecord = Record<string, string | number>;
+
+function baseRecord(overrides: Partial<MentorRecord>): MentorRecord {
+  return {
+    mentor: 'Mentor A',
+    linkedIn: 'LA',
+    'Recommendation For': 'Mentor A',
+    Pronouns: 'they/them',
+    'Count of Projects Mentored': 1,
+    'Students Mentored': 'Alice',
+    'Count of Students Mentored': 1,
+    'Mentorship Dates': 'Jan 1, 2024-Feb 5, 2024',
+    prompt: 'prompt text',
+    result: 'recommendation text',
+    ...overrides,
+  };
+}
+
+function recordWithFeedback(feedback: string): MentorRecord {
+  return baseRecord({
+    mentor: 'Mentor B',
+    linkedIn: 'LB',
+    'Recommendation For': 'Mentor B',
+    'What Students Had to Say': `\n` + feedback,
+    prompt: `prompt text\nWhat Students Had to Say:\n` + feedback,
+    result: 'rec B',
+  });
+}
+
+function parseRows(csv: string): Record<string, string>[] {
+  return parse(csv, { columns: true, relax_quotes: true, relax_column_count: true }) as Record<string, string>[];
+}
+
+// The header is pinned by an explicit `columns` list, so it no longer depends on the
+// keys present in the first record. When the first-returned mentor has no qualifying
+// feedback (key absent), the `What Students Had to Say` column must still be emitted.
+(function testHeaderIsPinnedRegardlessOfFirstRecordShape() {
+  const records = [
+    baseRecord({}),
+    recordWithFeedback('- q: They were wonderful and very supportive throughout.'),
+  ];
+  const csv = buildRecommendationsCsv(records);
+  const headerLine = csv.split('\n', 1)[0];
+  assertEqual(headerLine.split(','), EXPECTED_COLUMNS, 'Header is the pinned column list even when the first record lacks What Students Had to Say');
+})();
+
+// The original bug: with the column dropped, a later mentor's feedback only survived
+// inside the multi-line `prompt` cell. The fix gives it a dedicated, header-addressable
+// cell while feedback-less mentors get an empty cell.
+(function testFeedbackCellPreservedWhenFirstRecordLacksKey() {
+  const feedback = '- q: They were wonderful and very supportive throughout.';
+  const records = [
+    baseRecord({}),
+    recordWithFeedback(feedback),
+  ];
+  const csv = buildRecommendationsCsv(records);
+  const rows = parseRows(csv);
+  assertEqual(rows[0]['What Students Had to Say'], '', 'Feedback-less first mentor gets an empty dedicated cell');
+  assertEqual(rows[1]['What Students Had to Say'], `\n` + feedback, "Later mentor's feedback lands in its own dedicated cell, not just embedded in prompt");
+})();
+
+// Real feedback blobs are multi-line and comma-laden; the dedicated cell must round-trip
+// intact through csv-parse so downstream consumers can read it by header name.
+(function testMultilineFeedbackRoundTripsInDedicatedCell() {
+  const feedback = [
+    '- q1: They were wonderful and very supportive throughout.',
+    '- q2: Patient, knowledgeable, and always available when we needed help.',
+    '- q3: One of the best mentors I have worked with, highly recommend.',
+  ].join('\n');
+  const records = [
+    baseRecord({}),
+    recordWithFeedback(feedback),
+  ];
+  const csv = buildRecommendationsCsv(records);
+  const rows = parseRows(csv);
+  assertEqual(rows[1]['What Students Had to Say'], `\n` + feedback, 'Multi-line feedback survives CSV round-trip in its dedicated, header-addressable cell');
+  assertEqual(rows[1].mentor, 'Mentor B', 'Non-feedback columns still parse correctly alongside the feedback cell');
+})();
+
+// Pinning columns must not drop or reorder the unconditional fields, including the
+// GPT `result` deliverable and comma-bearing `Students Mentored` values.
+(function testUnconditionalColumnsRemainCorrect() {
+  const records = [
+    baseRecord({
+      mentor: 'Mentor A',
+      linkedIn: 'https://linkedin.com/in/a',
+      'Recommendation For': 'Mentor A',
+      Pronouns: 'she/her',
+      'Count of Projects Mentored': 2,
+      'Students Mentored': 'Alice, Bob',
+      'Count of Students Mentored': 2,
+      'Mentorship Dates': 'Jan 1, 2024-Feb 5, 2024, Sep 1, 2024-Oct 5, 2024',
+      prompt: 'the prompt',
+      result: 'the recommendation',
+    }),
+  ];
+  const csv = buildRecommendationsCsv(records);
+  const rows = parseRows(csv);
+  assertEqual(rows[0].mentor, 'Mentor A', 'mentor column preserved');
+  assertEqual(rows[0].linkedIn, 'https://linkedin.com/in/a', 'linkedIn column preserved');
+  assertEqual(rows[0]['Count of Projects Mentored'], '2', 'Count of Projects Mentored preserved');
+  assertEqual(rows[0]['Students Mentored'], 'Alice, Bob', 'Students Mentored preserves embedded comma via CSV quoting');
+  assertEqual(rows[0].result, 'the recommendation', 'GPT result column preserved');
+})();
+
+async function main(): Promise<void> {
+  if (failures > 0) {
+    console.error(`\n${failures} test(s) failed.`);
+    process.exit(1);
+  }
+  console.log('\nAll tests passed.');
+}
+
+main();

src/activities/tasks/mentorWriteRecommendations.ts

@@ -58,6 +58,22 @@ Summarize positive feedback from their students, if anything.
 
 const DEBUG = makeDebug('activities:tasks:mentorWriteRecommendations');
 
+// Explicit, ordered column list for the recommendations CSV. csv-stringify derives
+// columns from the FIRST record only when `columns` is omitted; since the
+// `What Students Had to Say` key is conditionally present per mentor, omitting this
+// list would silently drop that column (and every later mentor's cell) whenever the
+// first-returned mentor had no qualifying feedback. Pinning the columns makes the
+// header order-independent and guarantees the feedback column is always emitted.
+const CSV_COLUMNS = [
+  'mentor', 'linkedIn', 'Recommendation For', 'Pronouns',
+  'Count of Projects Mentored', 'Students Mentored', 'Count of Students Mentored',
+  'Mentorship Dates', 'What Students Had to Say', 'prompt', 'result',
+];
+
+export function buildRecommendationsCsv(recommendations: object[]): string {
+  return stringify(recommendations, { header: true, columns: CSV_COLUMNS });
+}
+
 
 interface MentorWriteRecommendationsArgs {
   channel: string
@@ -183,7 +199,7 @@ export default async function mentorWriteRecommendations({ auth }: Context, args
     });
   }
 
-  const csv = stringify(recommendations, { header: true });
+  const csv = buildRecommendationsCsv(recommendations);
   const slack = getSlackClientForEvent(event);
 
   await slack.files.upload({