Kill My Idea
One parallel Jev batch per startup idea: eight 0 to 4 scores, weighted average, KILL FIX SHIP verdict. No generative LLM on the hot path.
Overview
Kill My Idea (github.com/monteduro/killmyidea, killmyidea.stemonte.io) is a Vite playground that sends exactly one TypeSafe Jev request per idea. Ten parallel questions run together: eight rubric scores from 0 to 4 (real problem, money, competition, and similar indie fields), plus category and understandability nouls. src/lib/scoring.ts multiplies each score by 25, applies goal-specific weights (make money, open source, or just for fun), averages, and src/lib/verdict.ts maps below 50 to KILL, 50 to 64 to FIX, and 65+ to SHIP with a clarity gate. No generative LLM writes the verdict; the UI exposes raw Jev probabilities in a collapsible panel. api/evaluate.ts holds the TypeSafe key server-side; optional SQLite archives successful runs.
Problem: Founders waste weeks on ideas that fail basic indie-hacker sanity checks, but generic chat opinions are not reproducible or thresholdable.
Built for: Solo builders who want a blunt KILL, FIX, or SHIP label from structured scores, not a motivational essay from GPT-class models.
First indexed on Jev Directory: 2026-09-24
Creator and team
- Name
- Marco Monteduro
- Handle
- @monteduro
How Jev is used
- Role in the product flow
- Single batched Score and Noul pass that feeds deterministic weighted verdict math
- Primitives
- ScoreNoul
- State in
- User idea text plus selected goal mode that reweights Real problem, Money, Adoption, Fun, and related fields per README scoring section.
- Decision out
- Eight 0 to 4 scores, category and clarity signals, weighted 0 to 100 average, and KILL, FIX, or SHIP label with latency and token usage in the debug panel.
- Browser posts idea and goal to /api/evaluate
- Server calls TypeSafe Jev once with ten parallel questions
- scoring.ts applies WEIGHTS per goal and computes weighted average
- verdict.ts applies thresholds and clarity gate for the headline label
Jev is the entire judgment engine: parallel Score heads replace a jury of chat prompts. The product value is transparent math on top of typed outputs, so you can tune weights in git instead of re-prompting. anishfn-shapeshift fans out many intent questions for UI morphing; killmyidea fans out once for investor-style rubrics. Hosted TypeSafe keys stay off the client; TYPESAFE_MOCK=1 enables UI work without billing. This is a playground, not a replacement for legal or market research. Pair with classifier.dev when you need HTTP labels outside a Vite demo.
Sourced performance claims
- README states no generative LLM on the verdict path; one Jev request with ten parallel questions.Source: github.com/monteduro/killmyidea README
- Verdict thresholds documented as score below 50 KILL, 50 to 64 FIX, 65+ SHIP in src/lib/verdict.ts.Source: github.com/monteduro/killmyidea src/lib/verdict.ts
- Public GitHub repo monteduro/killmyidea had 191 stars and 24 forks when this listing was drafted.Source: GitHub API September 2026
Features and stack
Features
- KILL, FIX, SHIP headline with expandable raw Jev panel
- Goal modes reweight money versus adoption versus fun fields
- Server-side evaluate function; keys never shipped to the browser
- Optional SQLite analytics archive with per-request opt-out
- Live demo at killmyidea.stemonte.io plus Vercel deploy docs
Stack
- TypeScript
- Vite
- TypeSafe Jev API
- SQLite analytics
Pricing: Open source UI; each evaluation spends TypeSafe Jev tokens from your server key.
Links
FAQ
- Does a chat model write the KILL or SHIP text?
- No. README emphasizes Jev scores feed deterministic TypeScript verdict code. There is no separate generative model on the hot path.
- Can I run it without a TypeSafe key?
- Set TYPESAFE_MOCK=1 locally for deterministic MOCK DATA responses. Production needs TYPESAFE_API_KEY on the server.
- How is this different from asking ChatGPT to roast my idea?
- You get fixed rubric scores, published weights, and the same thresholds every time, which is closer to a gate than a prose opinion.
Related learn guides
Original Jev guidance that pairs with this product pattern.
- Jev use cases
The patterns builders actually search for: moderation, routing, triage, RAG verify, and agent gates.
Related products
Hand-picked neighbors with rich profiles or overlapping tags.