ProveAI Origin
Try it

User journeys

Three named ways to use ProveAI Origin, depending on what brought you here. Most users hit them in order: acute first, then dev-loop, then CI-gate.

TL;DR
WhenJourneyOutput
Every production failure becomes a regression testFrom failure to gated regression testPaste a failing trace, get a cited cause in ~7s, save it as a snapshot, group snapshots into a suite, replay on every change, and gate PRs in CI.
Every production failure becomes a regression testFrom failure to gated regression testPaste a failing trace, get a cited cause in ~7s, save it as a snapshot, group snapshots into a suite, replay on every change, and gate PRs in CI.
Every production failure becomes a regression testFrom failure to gated regression testPaste a failing trace, get a cited cause in ~7s, save it as a snapshot, group snapshots into a suite, replay on every change, and gate PRs in CI.
01

From failure to gated regression test

Every production failure becomes a regression test

Problem

A production agent failure tells you something broke, but not what to do next: which span caused it, whether the fix actually fixes it, whether the next change regresses it, whether CI will catch it.

Outcome

Paste a failing trace, get a cited cause in ~7s, save it as a snapshot, group snapshots into a suite, replay on every change, and gate PRs in CI. The failure becomes the test that prevents the next one.

Flow
┌────────────────┐     ┌────────────────┐     ┌────────────────┐     ┌────────────────┐
│ Paste failing  │ ──▶ │ Review your a… │ ──▶ │ Save failure   │ ──▶ │ Group tests s… │
└────────────────┘     └────────────────┘     └────────────────┘     └────────────────┘
         │
         ▼
┌────────────────┐     ┌────────────────┐
│ Replay every   │ ──▶ │ Wire block re… │
└────────────────┘     └────────────────┘
Steps
  1. 1

    Paste a failing trace

    /app/new

    Paste a failing agent trace on the New attribution page. We auto-detect the format, pick the failure span, and return a rank-1 cited cause + L1/L4 verdict + quoted evidence in ~7 seconds.

  2. 2

    Review your attributions

    /app/inbox?source=attributions

    Every paste-and-go shows up in your Inbox Attributions feed. Open one to see the rank-1 cited span, L1/L4 verdict, confidence, and the substring the judge quoted from your trace as evidence.

  3. 3

    Save the failure as a test

    /app/snapshots

    Hit "Save as snapshot" on any attribution result page. The trace + cited cause + verdict become a regression test that lives on the per-agent Tests tab (and the flat Snapshots list). Auto-labeled, span-level evidence, no human curation needed.

    Save as snapshot
  4. 4

    Group tests into a suite

    /app/snapshot-suites

    Suites are the unit of governance. Group related tests together on the agent's Suites tab so the same baselines apply to every change — your fix, a PR from Cursor, a coding agent's mid-session commit.

  5. 5

    Replay on every change

    /app/snapshots

    From any snapshot, click "Replay" to re-run the test through the current pipeline. Each run shows up on the per-agent Runs tab and bucketizes results into PASSED / CHANGED-INTENTIONAL / CHANGED-UNEXPECTED — verify your fix, verify a PR, verify a coding agent's commit before it lands.

  6. 6

    Wire CI to block regressions

    /app/integrations

    Connect the GitHub App on the Integrations page, then drop the ProveAI Origin GitHub Action into your workflow. Every PR with a failing trace gets a rank-1 attribution + cited evidence as a PR comment; suite-wide regressions block merge.

02

From failure to gated regression test

Every production failure becomes a regression test

Problem

A production agent failure tells you something broke, but not what to do next: which span caused it, whether the fix actually fixes it, whether the next change regresses it, whether CI will catch it.

Outcome

Paste a failing trace, get a cited cause in ~7s, save it as a snapshot, group snapshots into a suite, replay on every change, and gate PRs in CI. The failure becomes the test that prevents the next one.

Flow
┌────────────────┐     ┌────────────────┐     ┌────────────────┐     ┌────────────────┐
│ Paste failing  │ ──▶ │ Review your a… │ ──▶ │ Save failure   │ ──▶ │ Group tests s… │
└────────────────┘     └────────────────┘     └────────────────┘     └────────────────┘
         │
         ▼
┌────────────────┐     ┌────────────────┐
│ Replay every   │ ──▶ │ Wire block re… │
└────────────────┘     └────────────────┘
Steps
  1. 1

    Paste a failing trace

    /app/new

    Paste a failing agent trace on the New attribution page. We auto-detect the format, pick the failure span, and return a rank-1 cited cause + L1/L4 verdict + quoted evidence in ~7 seconds.

  2. 2

    Review your attributions

    /app/inbox?source=attributions

    Every paste-and-go shows up in your Inbox Attributions feed. Open one to see the rank-1 cited span, L1/L4 verdict, confidence, and the substring the judge quoted from your trace as evidence.

  3. 3

    Save the failure as a test

    /app/snapshots

    Hit "Save as snapshot" on any attribution result page. The trace + cited cause + verdict become a regression test that lives on the per-agent Tests tab (and the flat Snapshots list). Auto-labeled, span-level evidence, no human curation needed.

    Save as snapshot
  4. 4

    Group tests into a suite

    /app/snapshot-suites

    Suites are the unit of governance. Group related tests together on the agent's Suites tab so the same baselines apply to every change — your fix, a PR from Cursor, a coding agent's mid-session commit.

  5. 5

    Replay on every change

    /app/snapshots

    From any snapshot, click "Replay" to re-run the test through the current pipeline. Each run shows up on the per-agent Runs tab and bucketizes results into PASSED / CHANGED-INTENTIONAL / CHANGED-UNEXPECTED — verify your fix, verify a PR, verify a coding agent's commit before it lands.

  6. 6

    Wire CI to block regressions

    /app/integrations

    Connect the GitHub App on the Integrations page, then drop the ProveAI Origin GitHub Action into your workflow. Every PR with a failing trace gets a rank-1 attribution + cited evidence as a PR comment; suite-wide regressions block merge.

03

From failure to gated regression test

Every production failure becomes a regression test

Problem

A production agent failure tells you something broke, but not what to do next: which span caused it, whether the fix actually fixes it, whether the next change regresses it, whether CI will catch it.

Outcome

Paste a failing trace, get a cited cause in ~7s, save it as a snapshot, group snapshots into a suite, replay on every change, and gate PRs in CI. The failure becomes the test that prevents the next one.

Flow
┌────────────────┐     ┌────────────────┐     ┌────────────────┐     ┌────────────────┐
│ Paste failing  │ ──▶ │ Review your a… │ ──▶ │ Save failure   │ ──▶ │ Group tests s… │
└────────────────┘     └────────────────┘     └────────────────┘     └────────────────┘
         │
         ▼
┌────────────────┐     ┌────────────────┐
│ Replay every   │ ──▶ │ Wire block re… │
└────────────────┘     └────────────────┘
Steps
  1. 1

    Paste a failing trace

    /app/new

    Paste a failing agent trace on the New attribution page. We auto-detect the format, pick the failure span, and return a rank-1 cited cause + L1/L4 verdict + quoted evidence in ~7 seconds.

  2. 2

    Review your attributions

    /app/inbox?source=attributions

    Every paste-and-go shows up in your Inbox Attributions feed. Open one to see the rank-1 cited span, L1/L4 verdict, confidence, and the substring the judge quoted from your trace as evidence.

  3. 3

    Save the failure as a test

    /app/snapshots

    Hit "Save as snapshot" on any attribution result page. The trace + cited cause + verdict become a regression test that lives on the per-agent Tests tab (and the flat Snapshots list). Auto-labeled, span-level evidence, no human curation needed.

    Save as snapshot
  4. 4

    Group tests into a suite

    /app/snapshot-suites

    Suites are the unit of governance. Group related tests together on the agent's Suites tab so the same baselines apply to every change — your fix, a PR from Cursor, a coding agent's mid-session commit.

  5. 5

    Replay on every change

    /app/snapshots

    From any snapshot, click "Replay" to re-run the test through the current pipeline. Each run shows up on the per-agent Runs tab and bucketizes results into PASSED / CHANGED-INTENTIONAL / CHANGED-UNEXPECTED — verify your fix, verify a PR, verify a coding agent's commit before it lands.

  6. 6

    Wire CI to block regressions

    /app/integrations

    Connect the GitHub App on the Integrations page, then drop the ProveAI Origin GitHub Action into your workflow. Every PR with a failing trace gets a rank-1 attribution + cited evidence as a PR comment; suite-wide regressions block merge.

Why these three (and not four, five, ten)

We deliberately cut by trigger, not destination. Insights / Leaderboard / Patterns are surfaces you might land on in any journey — they aren't separate journeys. If a surface feels like its own journey but you can't name a distinct trigger that isn't a sub-step of the three above, it's supporting infra. Add it as supporting infra; don't fork a fourth journey.

Start the acute journeyRead how it works