An open-source audit pack by ufodia

Test the promise.
Recheck the decision.

Check whether a software promise still holds after the product changes. Compare the evidence for product fixes, buying decisions and campaign or support copy.

Python local runner · Open-source MIT · No API key for the demo

ANATOMY OF A RECHECK01—03
01 / OBSERVED CHANGE

Lowest plan with CSV export

$9$49/mo

CSV export moves to Pro.
The old promise stays behind.

Fictional SaaS example
02 / DEPENDENT DECISIONSREVIEW
ProductReproduce the plan gate
BuyerReassess the $20 limit
CampaignCorrect the export claim
Evidence first. Conclusions in context.
Built for the moment
“that recommendation is out of date.”

Connect the claim. Inspect the task.
Show what changed—and what did not.

01 / The evidence workbench

One product change.
Three different decisions.

This fictional example follows a CSV export feature through a plan change and a documentation repair. Explore the captured records, then look through each audience’s lens.

Fictional fixture, not customer research. This page displays recorded example bundles. It does not browse a live product, run an AI model or predict conversions.

  1. 1

    Open “Plan changed”. The offer and help text still promise the old feature.

  2. 2

    Switch the audience. One change means different work for each reader.

  3. 3

    Try “Copy repaired”. Correct marketing does not fix the buyer’s budget.

DECISION REPORTLoading fixture

Loading the recorded example. The source repository includes the model-free local demo.

Recorded example · timestamps in bundle
THE SOURCE TRAIL

Open the evidence.

Follow a finding to its captured source. Open a record below, or download all four fictional bundles with their artifacts.

Download complete example bundles ZIP · all four snapshots + captured artifacts ↓

Evidence records will appear when the fixture loads.

02 / A bounded workflow

A review you can
follow, question, repeat.

Four optional AI review roles in the local runner prioritize recorded review points. They cannot rewrite a report’s decision; unknowns and required checks stay visible.

  1. 01

    Set the brief

    Specify the goal, plan, budget, sources and allowed actions.

  2. 02

    Inspect the task

    Capture scoped sources and isolated browser observations.

  3. 03

    Review the evidence

    Keep exact outcomes. Verify behavior before correcting a documented fact.

  4. 04

    Recheck the decision

    Revisit current sources and relevant tasks when the evidence changes.

“Current” does not mean “true.” An unchanged source can still be wrong. An old task result remains historical until the task is tested again.

Read the architecture

03 / Make it reproducible

Your next review
starts locally.

Start with the fictional example. Inspect the JSON, compare the reports and reproduce the results before connecting a real application.

For your own application, configure permitted sources and browser checks in the local runner. Its optional AI review keeps recorded outcomes and follow-up conditions intact. See how decisions stay consistent .

Read the quick start
MODEL-FREE EXAMPLE
git clone https://github.com/ufodia/decision-recheck.git
cd decision-recheck
uv sync --locked
uv run decision-recheck demo --output runs/demo

Python 3.12 · See the README for tested environments.
Then open runs/demo/index.html in your browser.
No API key or browser installation needed. Use a new output directory for another run.

Independent software. Shared evidence.

Help keep the
examples working.

Support maintains reproducible fixtures, browser adapters and clear documentation. You can also contribute a small test case, a source extractor or a useful bug report.

SUPPORT THE MAINTAINERBuy ufodia a coffee