1. Structured knowledge graph
Every framework, peer-reviewed paper, and operator essay Sophie can cite is a row in our Postgres database with slug, author, year, summary, vertical tags, and an authority score (0-100). When Sophie cites Porter or MEDDPICC, the slug resolves to that row. No hallucinated references.
2. Vertical-native vocabulary
Sophie speaks the language of the vertical. DTC: blended MER, contribution margin per cycle, F1-to-F2 conversion. Fintech: take-rate net of scheme fees, RAROC, three-lines-of-defense. ProfSvc: utilisation x realisation x leverage, Maister 2x2 portfolio. SaaS: MEDDPICC, sales velocity decomposition, Magic Number net of S&M.
3. Deterministic GRIP scoring
Every workspace gets a GRIP score: 12 pillars x 4 dimensions (Guidance, Resources, Implementation, Performance), with 22 calibration rules. Per-vertical models: SaaS, DTC, Fintech, ProfSvc. The score is recomputed on every snapshot, not estimated from chat. Sophie quotes GRIP numbers; she never invents them.
4. Framework -> archetype application rules
For each archetype (Discount Addiction, Compliance Debt, Bench Accumulation, Pricing Governance, etc.) Sophie has a curated list of frameworks that apply, each with an explicit application rule and an example move. When she diagnoses an archetype, the right framework is already in scope.
5. Cohort cases
18 anonymised aggregated-cohort cases per vertical, totalling 72. Each case has the triggering metric, the move taken, the EUR result, the pillar delta, and a 2-3 sentence narrative. Sophie cites these by id when she wants to give the operator comparable evidence. She never invents a case.
6. Peer outcomes
For every move Sophie can recommend there is a peer-cohort outcome row in archetype_peer_outcomes: median pillar delta, P25/P75, sample size, worked-share. Sophie quotes those numbers verbatim. k-anon greater-than-or-equal 5 enforced at the seed table level.