evidence-gates engine · beta

First prove people want your idea. Then build.

We do not judge how your idea sounds. We judge real evidence from real customer interviews: what a person actually did and what they committed to. Every conclusion comes with a quote and a timecode, not a guess.

How it works
  1. Idea
  2. Interview
  3. Evidence
  4. Green light

green means evidence collected, not a promise of PMF

The Evidence Judge

We judge behavior, not prose

Most AI validators read your text and tell you how polished it sounds. RightCode does something different: it takes a real customer interview, breaks it down into pains, jobs-to-be-done and commitments, and rules against a success criterion you locked before collecting the evidence. There are two Judges, deliberately separated.

The core rule: every conclusion comes from a real quote in the recording, nothing else. No quote, no conclusion. You get an Evidence artifact instead of a gut-feel score: what is proven, by what exactly, and what is still missing on the way to green.

Interview Judgewas the interview honest: no leading questions, behavior instead of opinions
Conclusion Judgedoes the collected evidence actually support your conclusion
Evidence is weakrisk not closed

“Last month I went over budget and actually started tracking expenses in my notes”

04:12
What next

Needs a second respondent and proof of a stable tracking habit, not a one-off reaction to a shock.

Not a gut-feel score: the conclusion rests on a real quote with a timecode.

Six gates

An evidence bar that rises with you

Validation is not one question, it is a ladder. Each next gate demands stronger evidence: first that the problem is real, then that you understand the customer's job, then the market, observable demand, retention. The bar rises from words to behavior and external data.

We start with the foundation: gates built on real interviews. The remaining steps open as the product matures.

  1. Problemavailable
  2. Customer / JTBDavailablebeta
  3. Hypothesisvision · soon
  4. Marketexperiment · beta
  5. Demandexperiment · beta
  6. Retentionvision · soon

Today the interview gates are live in the product. Customer / JTBD is beta: the loop works, Judge calibration against human labels is ahead. Market and Demand are experiments, the methodology is still maturing. The remaining steps are a roadmap.

Reward for evidence

Evidence earns the reward, it is not given on credit

The artifact and the path to green are forward value, and they are not handed out on credit. A weak input yields an empty frame: the shape is there, but there is nothing to fill it with. A strong input - a real quote with a timecode - fills the same artifact: what is proven, by what exactly, what comes next. You do not buy the reward upfront, you earn it with evidence.

Weak evidence
  • nothing yet for the conclusion to rest on
  • no quote
  • path to green is empty
frame is empty - reward not earned
Strong evidence

“I moved the deadline and refunded the deposit just to stop dragging this process further”

07:41
  • conclusion rests on a quote
  • path to green is concrete
  • risk is closed by evidence
artifact is filled - reward earned

A weak input yields an empty frame, a strong one yields a finished artifact. Evidence earns the reward, not an advance.

What makes us different

They judge text. We judge evidence.

Most idea validators score how plausible a description sounds, AI co-founders write plans and pitches, PM copilots help you shape a PRD. Too often it stays text work, where polished text passes for validation. But a pretty description does not mean a customer actually reaches for the solution.

RightCode stands in the white space: real evidence from real customer conversations plus strict but advisory gating. The moat is three layers that are hard to copy: an Evidence Judge with a locked criterion, a real-interviews-only principle, and a proprietary corpus of interviews and outcomes that makes the judgment better calibrated over time.

They · judge text
  • idea validators - score how plausible the description sounds
  • AI co-founders - write the plan and the pitch
  • PM copilots - help shape a PRD
We · judge evidence
  • an Evidence Judge with a locked success criterion
  • real interviews only - no synthetic data
  • a proprietary corpus of interviews and outcomes
Privacy as an asset

Your interviews do not leak into someone else's LLM

Privacy here is part of the construction, not fine print. The rented frontier model runs through a commercial API that does not train on your data by default; only the minimal input of a single inference leaves the perimeter: the transcript and the rubric for one specific verdict. Raw transcripts never leak into analytics or third parties.

More than that, your evidence accumulates as your asset: the Evidence artifact and the Risk Ledger stay with you and never reset. Over time RightCode improves the Judge on anonymized patterns, with consent, inside its own perimeter.

Self-hosting and training our own Judge on the corpus are the direction of the moat, not today's feature. Today it is rented inference, no training.

Your perimeter
  • Interviews
  • Raw transcripts
  • Evidence artifacts
  • Risk Ledger

Raw transcripts never go to analytics or third parties.

only the transcript + rubric for 1 verdict
Rented frontier model · no training

the commercial API does not train on your data by default

self-host · ZDR · our own Judge - the moat's direction, not today
Pricing

Start with one project. Grow when you have something to prove.

Three tiers shaped by how you work, not by a feature list. The price tracks the volume of real work, the number of projects and interviews, not premium buttons.

Solofor a single foundersoon

a few projects, limits sized for an honest discovery loop - the entry point to the first gates on real interviews

Profor an active foundersoon

more projects in parallel and higher processing limits

Enterprisefor teamssoon

collaboration, roles (founder / reviewer), extended limits

The pricing grid is in progress - exact prices and limits are not final yet.

Building is cheap now. Building the wrong thing is not.

Check what is actually proven before you burn the round on guesses. Start with one real interview and see what it proves.

How it works