Clarovo

Documentation

Getting started, the Clarovo framework, API reference, and security. The MBB methodology in software for boutique consulting firms.

01

Getting started

Four steps from signup to the partner-attested deliverable. The first engagement typically takes 4-6 weeks end-to-end.

  1. Sign up + create your workspace. Standard Supabase email + password auth. Your workspace is provisioned automatically; the owner role binds to your account.
  2. Create a client + an engagement. The engagement setup wizard walks through the SOW briefing, problem alignment, hypothesis generation, and workstream design — the Phase 1 action chain.
  3. Upload the source material. 10-K, prior reports, financial statements, internal research. The Z.3 file ingestion pipeline extracts + chunks + embeds the content into engagement_facts. The orchestrator's sub-agents read from there.
  4. Review + sign off on the sub-agent output. The 14 sub-agents (situation assessor, bottleneck diagnostician, market signal analyzer, pre-mortem runner, etc.) run end-to-end inside the Harness. The partner reviews each output, edits or flags as needed, and signs off. The methodology coverage stamp ticks up as the engagement progresses.
02

The framework

The 4 components every Clarovo engagement follows.

1. The Interrogation

10 MBB-grounded questions per session type (15 types total: PVT/EXT/TME/HIR/SNK/FND/BRN/REV/PRT/WAR/BOT/CHK/MKT/STK/SCN). The sub-agents answer these questions from the uploaded source material — you don't type.

2. The Consequence Engine

1st / 2nd / 3rd-order effects + Regret Score + Blind Spot + Decision Window + Probability. Deterministic Python scoring over the LLM's qualitative output (regret normalization 0-100 -> 0-1, decision-window parsing, sensitivity ranking, pairwise winner).

3. The Avoidance Question

The closing question the sub-agent surfaces — the one the EM is most likely to have been avoiding. The whole point of MBB rigor: name the thing before deciding.

4. Two Deliverables

The decision memo (the partner-attestation artifact) and the deck (the client-facing version). Both have share tokens; both have PDF exports.

03

API reference

The 4 main endpoint groups. Full OpenAPI schema at /openapi.json (or the live /openapi.json on the backend).

MethodPathAuthPurpose
POST/v1/engagements/{id}/run-nextauthRun the next sub-agent in the orchestrator state machine.
POST/v1/engagements/{id}/sign-offauthSubmit a sign-off decision (approved / edited / flagged / rejected).
POST/v1/billing/checkout-sessionauthCreate a Stripe Checkout session for a tier (pro / advisory / developer / retainer).
POST/v1/billing/pass-checkoutauthBuy a $2,500 single-engagement pass (one-time Stripe payment).
GET/v1/billing/retainerauthRetainer quota status for the workspace (slots used / remaining).
POST/v1/quickbooks/syncauthPull P&L + Balance Sheet from QuickBooks into engagement_facts.
04

Security

Multi-tenant data isolation via Supabase Row Level Security (RLS). Every table has a workspace_id column and a ws_isolation policy that scopes every read and write to the active workspace. The Harness records every sub-agent output to the audit trail (agent_runs + sign_offs + agent_run_citations + methodology_steps) so the partner can defend any number in the deliverable: "where did this come from?" → "fact #247 in engagement_facts → Q3 of the bottleneck session → Regret Score 73 → memo paragraph."

No-data-retention policy for the public beta: when you delete a workspace, all engagement_facts + agent_runs + sign_offs are tombstoned within 30 days. SOC 2 Type II is on the enterprise roadmap; Type 1 is gated on the first enterprise contract (Phase 3, per the §16 plan).