NoFlattery

A small local web app that tells you whether an idea should be KILLed, FIXed or SHIPped before you spend time on it. You describe the idea in plain text; the app scores it, explains the verdict, and gives you a one-page report you can export.

Idea  →  Jev scores a rubric  →  rules decide the verdict  →  local LLM (Ollama) explains it  →  report

Step	Who does it	What it produces
1. Scoring	Jev (API)	A 0–4 score and a confidence for each rubric question
2. Decision	Plain code (no AI)	Verdict, strengths, weaknesses, low-confidence answers, path to SHIP
3. Explanation	Ollama (runs on your machine)	The reason, top 3 risks, advice, and a test to run this week

Strengths, weaknesses and the path to SHIP are computed from Jev's scores by code, so the small local model cannot invent or misquote them. It only writes the prose.

The server listens on 127.0.0.1 only. It is meant to run on your own computer and is not reachable from the network.

Privacy: the idea text is sent to the Jev API for scoring. Ollama runs fully locally. Don't paste anything confidential you aren't allowed to send to Jev.

---

Requirements

• Node.js 18 or newer (uses the built-in fetch; there are no npm dependencies, so no npm install)
• Ollama installed and running, with a model pulled (e.g. gemma3:4b)
• A Jev API key and endpoint from your TypeSafe account

Setup

1. Get the code and open a terminal in the project folder:

   git clone https://github.com/Prestical/NoFlattery.git
   cd NoFlattery

2. Install a local model with Ollama (once):

   ollama pull gemma3:4b

   gemma3:4b is fast and good enough because it only writes the explanation. A bigger model gives sharper wording. Any model you pull shows up in the app's model dropdown.

3. Create your .env from the template:

   cp .env.example .env

   Then fill in every value:

   Variable	Meaning
   JEV_KEY	Your Jev API key (secret)
   JEV_URL	The Jev endpoint URL from your account
   JEV_MODEL	The Jev model name to use
   OLLAMA_URL	Where Ollama is listening, normally http://127.0.0.1:11434
   OLLAMA_MODEL	Default model for explanations, e.g. gemma3:4b
   PORT	Port for this web app, e.g. 8070

   All six are required. The server refuses to start and tells you which one is missing. .env is git-ignored, so your key never gets committed.

   Optional: JEV_PRICE_PER_M_INPUT and JEV_PRICE_PER_M_OUTPUT (USD per million tokens) change how the Jev cost is estimated.

Run

Make sure Ollama is running (the desktop app, or ollama serve), then:

node server.js        # or: npm start

Open http://localhost:<PORT> in your browser. Stop it with Ctrl+C.

Run the unit tests for the scoring rules with:

npm test

Using the app

1. Pick the project type (top-left select). It decides which questions count (see below).
2. Pick the model if you have more than one in Ollama.
3. Write the idea in the text box. The more of these you answer, the better the score: what problem, for whom, why now · how you measure success · what you will and won't deliver · who must agree · resources, skills and time · the top risks · what you give up by doing it.
4. Click Categorize. It takes a few seconds to a minute, depending on your model.
5. The input moves to the top and the report appears underneath.

Reading the report

• Verdict + score (0–100) and whether the verdict is firm. If low-confidence answers could flip it, the report says what it could become.
• Scores: one tile per question with the 0–4 score, Jev's confidence (⚠ when low), and the rubric level it matched.
• Strengths / Weaknesses: the highest and lowest scores.
• Unsure answers: answers where Jev's confidence was below 50%.
• Top 3 risks: each with a mitigation and an early warning signal.
• Path to SHIP (or Next improvements for a SHIP verdict): the changes that gain the most points, shown as "now → next level" from the rubric. Gate questions that block SHIP come first.
• Advice and Test this week.

How the verdict is decided

Each question has a weight; the weighted average becomes the 0–100 score.

• KILL: score below 40, or the Problem score is under 0.5
• SHIP: score 70 or higher and every gate question scores at least 2.0
• FIX: everything in between

Gate questions can veto SHIP no matter how high the total is: Problem always, and Demand for startups.

Project types

Type	Questions	Notes
Startup / product	All 12	Demand is a gate; Money and Reach count
Team / internal tool	11 (no Reach)	Money and Difference weigh less
Personal project	10 (no Money, no Reach)	Demand is no longer a gate; the LLM is told not to suggest market work

Dropped questions are not sent to Jev at all.

History and re-scoring

Every run is saved as JSON in Reports/history/. The History button lists saved reports; sort by newest or best score, reopen one, or delete it. If you edit a saved idea and run it again, the report shows the score change per question and in total.

Exporting

• Print / PDF: opens the browser's print dialog. The report is measured and scaled automatically to fit one A4 page (shrinks long reports, enlarges short ones). In the dialog choose Save as PDF; leave margins on Default and scale at 100%.
• Markdown: downloads a .md file.
• HTML: downloads a standalone .html file with the styling included.

Project layout

server.js            Node server: Jev scoring, verdict rules, Ollama call, history API
public/index.html    Page structure
public/style.css     Screen layout and the one-page print layout
public/app.js        UI, report rendering, history, exports, print fitting
tests/logic.test.js  Unit tests for verdicts, profiles, facts and model-output cleanup
docs/screenshot.webp Screenshot used in this README
Reports/history/     Saved reports (git-ignored)
usage.json           Running Jev token/cost totals (git-ignored)
.env / .env.example  Your settings / the empty template

Troubleshooting

Problem	Fix
Missing in .env: … on start	Add the listed variables to .env (copy them from .env.example)
Cannot reach Ollama at …	Start Ollama and check OLLAMA_URL
Model dropdown is empty or wrong	Run ollama list; pull a model with ollama pull <name>
EADDRINUSE	The port is taken; change PORT in .env
Report shows the verdict but "Explanation unavailable"	The LLM failed; click Categorize again, or try a different model
Jev errors (HTTP 401/403)	Check JEV_KEY, JEV_URL and JEV_MODEL
PDF spills onto a second page	Set the print scale to 100% and paper to A4 portrait; margins on Default