← Sonar

Judge Guide

Sonar × CCI

Every entry below is a falsifiable prompt — fire it into Sonar, observe the output, check it against the Expect line. If the behavior doesn't match, the build failed its own claim. That falsifiability is the point.

How to run

Open sonar/rules.md and sonar/identity.md. Feed both as the system prompt into any Claude session. Paste the prompt below. The operator should respond without asking clarifying questions — Sonar decides and drafts, or flags with the single question that resolves the ambiguity.

01

The core decision — DECLINE on a funding gap

Fire

Qualify Pebblewick Games: 4-person bootstrapped mobile puzzle studio. No press, no raise, no publisher deal.

Expect

DECLINE. Reason cites "no identifiable funding from any source" — not team size, not platform, not genre. Score appears (Tier 3 on fit, ~40/107), but rule 4 fires before the tier matters. Draft is brief and professional, not apologetic. No follow-up question.

Why it matters

Rule 4 is the funding filter. Sonar should surface the cited reason, not just the verdict — a DECLINE without a cause is a black box that can't be trusted.

02

The pursue path — mid-fit promoted by a live catalyst

Fire

Qualify Lumen Hollow Games: ~40 staff, narrative-horror title, announced October ship date, seed-funded, no audio credits on their team page.

Expect

PURSUE · primary-vendor. Rule 6 fires (Tier 2 + live catalyst). Score ~69/107. Outreach email leads with the ship date and the missing audio team — not a feature list. CCI credit cited matches the genre: Starfield or The Outer Worlds 2, not a sports title. "The in" is named if a CRM contact exists.

Why it matters

Proves the catalyst layer. Without the October ship date, this is NURTURE. With it, PURSUE. The rule that fired should be visible in the output — that's how a user audits the call.

03

The refusal — no fabrication on thin data

Fire

Qualify Triton Reef Games — can't find them anywhere.

Expect

FLAG · not found. Sonar names the exact gap: no studio matches the name or any URL. It asks the one question that resolves it ("Renamed, mis-spelled, or a sub-studio of a parent?"). It does not fabricate a score, a studio profile, or a guess. No verdict beyond FLAG.

Why it matters

The identity doc says "Missing data is a signal, not a guess." This is the hardest claim. A model that silently manufactures a studio and writes a PURSUE email fails the operator's core promise.

04

The edge case — studio closure → follow the person

Fire

Re-triage Northwind Interactive. News says the studio shut down last week. We have Priya Anand (Audio Director) in our CRM, second-degree via Tom.

Expect

Two decisions in one run. (a) DECLINE the company — studio closed, fit drops to 0, dead vendor target. (b) NURTURE the person — Priya Anand. Draft is a warm check-in, not a pitch: acknowledge the news, ask where she's landing, keep the door open. Closure named as a relationship catalyst. Flag that her next job change triggers a fresh re-triage.

Why it matters

This edge is in rules.md under Studio Closure. Good people scatter when studios close. Sonar should catch the pivot without being told — it's what separates an operator from a lookup tool.

05

The voice check — overflow nurture

Fire

Qualify Tidewright Studios: ~80 staff, action game, one Sr Sound Designer on team page, small publisher, recently shipped their last title.

Expect

NURTURE · overflow angle. Score ~59/107 — one point under the 60 overflow-pursue line; no fresh project catalyst. Watchlist entry names one specific watch trigger ("new project announcement or audio job post"). If a contact is Overdue in the CRM, a soft catch-up reason is noted alongside. Tone: direct and warm — no "excited to," no superlatives, no "reaching out to see if."

Why it matters

NURTURE is not a soft DECLINE — it's a positioned watch with a trigger. This also tests the overflow edge: the in-house audio lead is explicitly not a reason to DECLINE (rules.md, in-house audio edge).

Bar

  • ✓ 5 prompts, each with Fire + Expect
  • ✓ Every Expect is checkable in one read — no "it should generally"
  • ✓ Two prompts target refusals: fabrication (03) and edge handling (04)
  • ✓ One targets voice + tone (05)
  • ✓ All four verdicts covered: PURSUE (02) · NURTURE (05) · DECLINE (01, 04a) · FLAG (03, 04a edge)

Walkthrough — one full session

Qualify Northwind Studios: indie action-adventure, no audio team, no publisher deal, no raise announced.
  1. 1 Sonar reads the input against the scoring rubric (reference/scoring-rubric.md). Action-adventure scores high on genre (10). No audio team is an audio gap (22 points — the largest single cluster). But no funding from any source means rule 4 fires.
  2. 2 Score is computed — call it 58/107 — but the funding filter (rule 4) fires before the tier matters. No publisher, no raise, no revenue signal, no grant. "Can't hire on normal terms."
  3. 3 Verdict: DECLINE. Drafted artifact: a brief professional pass. No question handed back; the call is made.
  4. 4 MASTER-sheet record emitted: action first (DECLINE), score breakdown supporting. Pastes straight into column 15 of the target sheet.
What a generic AI does with the same input: writes a confident outreach paragraph — "Northwind Studios looks like a strong fit. Indie action-adventure. Recommend reaching out." No rubric, no funding check, no reason cited.

That's the product. Sonar doesn't profile studios. It screens them.