What

Follow-on to the agentic-taxonomy PR , prepares the three pilot customers (Pirros, Aspire, Vooma) for extraction, and applies Peter's feedback on two questions. No DB writes.

Customer-vocab docs (the Step-1b deliverable)

One per customer in its audit folder, each vocab field with values + glosses + evidence, a triage (seed-now / auto-canonicalize-post-extraction / optional-ask), and a tight "outstanding" ask-list:

• docs/audits/aspire/aspire-customer-vocab.md , mined from Aspire's live DB seeds (6/9 vocabs already seeded)
• docs/audits/vooma/vooma-customer-vocab.md , from Tenzin's transcript + IPP (+ his separately-provided loss reasons & call types)
• docs/audits/pirros/pirros-customer-vocab.md , from Peter's resource files (incl. demo_story_type + story→state mapping)

Pipeline: vocab-extraction (Step 1b) upgraded to produce these

• .claude/skills/vocab-extraction/SKILL.md , expanded from 7 vocabs to the full field set; added the triage legend + multi-source priority (DB → transcript → profile docs) + the standard docs/audits/<org>/<org>-customer-vocab.md deliverable + required "outstanding" section.
• .claude/skills/data-audit/SKILL.md , Step-1b row updated to match.

Master question doc + cross-company HTML

• Master taxonomy-map: added Q-DISCO ("which discovery questions actually matter") , a data-first mine of the rep.question corpus (content embeddings + landing strength + pain/use-case elicitation + outcome), with discriminative semantic mining as the spine.
• docs/audits/taxonomy-question-master.html , the cross-company master (32 question cards + per-company vocab + internal field defs); the reference to hold next to an extraction run.

Peter's feedback (org-agnostic)

• P1 (feel questions) → single pure COUNT of process_sentiment per discovery call (dropped the %/share).
• P3 (customer stories) → "customer speak": broadened social_proof narrative_type in product.ts + discovery.ts to include anonymous/vernacular customer references (firm name not required). outcome_roi kept as the co-axis.

Tests

• defaults-equivalence.test.ts content-hash re-pinned for the social_proof broadening (field id/order unchanged , 29 fields; only the two description texts moved). Test passes 2/2 locally. This directly addresses the automated "default change lacks regression coverage" advisory , the hash pin is the regression guard, updated consciously with a dated note.

Raw customer transcript kept out of the PR.

🤖 Generated with Claude Code

https://claude.ai/code/session_018Tggg2sBmoV3cCZGMvpsdX

Greptile Summary

This PR expands the taxonomy vocabulary-extraction workflow, adds customer-specific vocabulary references for Aspire, Pirros, and Vooma, and broadens the social_proof extraction definition. It also updates the taxonomy question documents and re-pins the defaults-equivalence hash.

• Defines DB-first, transcript-second, profile-document-third vocabulary sourcing and triage.
• Adds three customer vocabulary deliverables and cross-company taxonomy reference material.
• Broadens anonymous customer language under social_proof for pitch and demo-feature extraction.
• Updates the defaults hash regression guard for the changed descriptions.

Confidence Score: 4/5

The PR should not merge until the contradictory transcript-only instruction is reconciled with the new required multi-source workflow.

The Step-1b skill currently gives mutually incompatible sourcing instructions, so an operator can follow the documented method and omit the seeded DB values and profile evidence that the new deliverable requires.

Files Needing Attention: .claude/skills/vocab-extraction/SKILL.md

Important Files Changed