# Baserok

> Baserok is a customer-discovery platform that grounds Jobs-to-be-Done analysis in real public discussion. Type what you're building; the platform pulls evidence from Reddit, YouTube comments, App Store reviews, Bluesky, Hacker News, and Product Hunt; synthesises 5 cited personas with their jobs, pains, and workarounds; and ranks features by algorithmic confidence score. If the evidence doesn't clear the quality bar, the run is refunded.

For founders, indie hackers, and product builders who know customer discovery matters but won't schedule 40 Zoom calls. Built solo, open architecture at github.com/a1ex-hughes/baserok.

## Public pages

- [Landing — what Baserok is, who it's for, pricing](https://usebaserok.com/): Customer discovery without the interviews. £0 first run, no card.
- [Sign up — start your first research run](https://usebaserok.com/signup): Email + password. Onboarding IS the first run; no paywall before value.
- [Sign in](https://usebaserok.com/login): Returning users.
- [Reset password](https://usebaserok.com/reset-password): Self-service password reset.

## Legal

- [Terms of Service](https://usebaserok.com/terms): Plain-English service terms — refunds, liability, data ownership, account termination.
- [Privacy Policy](https://usebaserok.com/privacy): GDPR-compliant. UK ICO-registered. Data export and account deletion both self-service from /settings.

## How it works (high-level)

The pipeline runs in stages:

1. Source planning — an LLM router proposes which subreddits, YouTube queries, App Store searches, and Bluesky terms to gather, given the product description and target market.
2. Connectors gather raw citations from Reddit, YouTube comments, App Store reviews, Bluesky, Hacker News, Product Hunt, and Google Trends — typically 200-300 raw quotes per build.
3. A relevance classifier filters for on-topic content; a specificity classifier annotates lived-experience vs marketing-copy.
4. Each surviving citation gets a veracity composite score (trust × specificity × freshness).
5. Synthesis stage builds 5 grounded personas with jobs, pains, workarounds, and a day-in-life simulation. Every persona is evidence-indexed back to specific quotes.
6. Quality gate: if the corpus is too thin, too one-sided, or too low-credibility, the build refunds the credit and shows a directional-only banner. Hard failures (no corpus / synthesis collapse) refund and block the deliverable. Healthy corpora produce a full report.

## Pricing

- Free: 3 starter runs + 2 free runs/week thereafter (no card).
- Starter: £29/month — 5 runs/month.
- Pro: £79/month — 20 runs/month.
- Pay-as-you-go: £12/run.

If a run lands in `limited_research` (directional, evidence one-sided) or `failed_research` (corpus too thin), the credit is automatically refunded.

## Optional

- [GitHub repository — open architecture, ADRs, failure logs](https://github.com/a1ex-hughes/baserok): Solo build, ~6 weeks, 40+ ADRs, 30+ migrations, full pipeline source.