Skip to content
Kiragu MainaSystems workspace
Systems onlineSend a brief
All tools

Analytics and Experiment Repair Kit

Trace the recorded steps toward a purchase, separate repeats and tests, and see whether the data supports your next decision.

Founders deciding what to change when visitor, chat, and purchase reports tell different stories.

Visual examples. Plain language. Go at your own pace.

Analytics and Experiment Repair Kit

Package preview. Illustrations use sample data.

Numbers you can trustExample

Sample click records

Click ACopy of click ATest click
3 recordsNot necessarily 3 real clicks
A small example cannot prove which page sells more.
The challengeRepeat clicks mixed with tests
Illustration only. These are sample screens and values.

Slide 1 / 6

The redesign looks like a winner

Imagine launching a new product page and seeing more activity. Then you remember testing checkout yourself and importing the same records twice.

  • Recorded actions are not necessarily different people.
  • A bigger count is a question to investigate before calling the change a success.
1 of 6

What this means for your project

Make sense of your website numbers before deciding what to change.

The redesign that looked like a winner.

You change your homepage and the numbers jump. Was the new message better, or did yesterday’s testing get counted? Before you spend another weekend rebuilding, trace the counts, separate repeats and tests, and see which questions the data can actually answer.

How would I use this with AI?

A skill is a set of instructions and examples for your AI coding assistant. You bring your idea and your project. The skill gives the assistant a process to follow.

  1. Tell your AI what you want to build and share the package instructions.
  2. Work through the setup together. You may need accounts or tools listed below.
  3. Review the changes and test them before putting them on your live site.

This is a package you use while building, not a finished website or a service that runs it for you.

Not sure where to start? Ask about a project like yours.

Explore the example

Made-up example
  1. Read records
  2. Remove repeats
  3. Separate visit groups
  4. Show report

Starting point

Same records imported twice: 18 records received

What changes

Still 9 distinct activities across the two data sources

Try including visits marked as tests and see the counts rise. Older records stay separate. A paid order with a missing checkout step still counts toward payment totals.

Explore an imagined launch report with repeated records and test activity. These are not customer results.

See how it works

Step 1 of 6

The redesign looks like a winner

Imagine launching a new product page and seeing more activity. Then you remember testing checkout yourself and importing the same records twice.

Starting situationSite activity and payment records come from several sources
Where it breaksRepeated records and test visits can make demand look higher than it is
Made-up example. Try the interactive preview to see what changes.
  • Recorded actions are not necessarily different people.
  • A bigger count is a question to investigate before calling the change a success.
About this example: Inspect the sample

Step 2 of 6

See what is left after the test runs

Separate marked test activity and repeated records. Follow the recorded steps toward a purchase, with gaps visible beside the counts.

BeforeRepeated records, test visits, and payments in different currencies
AfterEach activity counted once, visits grouped by type, and payment totals kept separate by currency
Made-up example. Try the interactive preview to see what changes.
  • Keep currencies separate and do not guess missing data.
  • Keep older visitor identities separate where they cannot be reliably matched.
About this example: Inspect the sample

Step 3 of 6

Trace a visit through the recorded steps

Check each record, remove duplicate IDs, and keep actions in order. The result is a funnel: the recorded path through stages such as visit, checkout, and payment.

  1. Check the records
  2. Remove repeated records
  3. Separate visit groups
  4. Show gaps in the data
Steps shown in the example.
  • Importing the same records again must not inflate totals.
  • Only a later action counts as progress to the next step.
  • Compare groups that meet the rules of your experiment.
About this example: Inspect the sample

Step 4 of 6

Put your own test visits back in

Toggle test activity and repeated imports in the sample. See which totals change and whether there is enough evidence to compare the pages.

What to tryTry importing the same records again or including visits marked as tests
Expected resultImporting twice leaves activity counts unchanged. Including test visits raises the totals
Made-up example. Try the interactive preview to see what changes.
  • The invented sample has known expected results.
  • Too little suitable data leaves the result undecided.
Try the example
About this example: Inspect the sample

Step 5 of 6

Give your AI the records behind the chart

Use the source list, import tools, and reporting example to investigate why your tools disagree. Keep the explanation alongside the number.

SKILL.md
Instructions for listing tracking sources and investigating disagreements
scripts/
Tools to align records, count steps, and create reports
examples/
Invented activity records without real visitor details
tests/
Checks for repeated imports, action order, and uncertain results
What the package contains. Check the release details for the exact file list.
  • Notes on missing data, excluded records, and uncertainty
  • A report in JSON, structured text your code can read
  • Sample records to test the checks; a hosted dashboard is not included
About this example: Inspect the sample

Step 6 of 6

For the decision you need to make next

This fits a founder who can get the tracking records and wants to judge a change with better evidence. Sometimes the useful answer is that you need more data.

What you need
  • Tracking records with private details removed, or access to your own site’s activity collection code
  • A stable ID for each recorded action and documented times, so repeated records and action order can be checked
  • For an experiment, a chosen result to measure, such as completed purchases, and rules for which visitors belong in the comparison
  • Node.js 20 or later to run the tools locally; the included examples need no account or external service
Limits to knowMissing data stays unknown. This small sample cannot show which page version performs better
Check what you need and what the product can do before choosing it.
  • It cannot recreate activity that was never recorded.
  • An unknown source stays unknown; clicks alone do not prove what caused a purchase.
  • A fair experiment and enough suitable data are needed before declaring a winner.
About this example: Inspect the sample

What you receive

  • A list of tracking sources and tools to put their records into a consistent format
  • Imports that avoid counting the same record twice and checks on the order of visitor actions
  • An example report showing data gaps and comparisons between versions of your site
  • Invented sample data with known results for checking the tools
See the files in the package
SKILL.md
Instructions for listing tracking sources and investigating disagreements
scripts/
Tools to align records, count steps, and create reports
examples/
Invented activity records without real visitor details
tests/
Checks for repeated imports, action order, and uncertain results

What you need to get started

  • Tracking records with private details removed, or access to your own site’s activity collection code
  • A stable ID for each recorded action and documented times, so repeated records and action order can be checked
  • For an experiment, a chosen result to measure, such as completed purchases, and rules for which visitors belong in the comparison
  • Node.js 20 or later to run the tools locally; the included examples need no account or external service

What you can check here

Package version 1.0.0 / page version 3

  • Inspect the sample

    The interactive preview uses invented data to explain the steps. It does not show customer results or evidence of increased sales.

Will it work with your tools?

Codex and Node.jsExample setup only
Paid package version 1.0.0. Check the package compatibility record for the exact setups tested. Other tools, hosting services, and live provider connections need validation in your own setup.

Before you choose

Read the paid skill terms for permitted use, what your purchase includes, help, and refunds.

  • One purchase licenses one individual or organisation to use and modify this version in unlimited own and client projects, including commercial work. Do not resell the standalone package, sublicense it as a competing package, redistribute it publicly, or share it with unrelated purchasers. Later releases may be sold separately; no lifetime updates are promised.
  • It cannot recover past activity that was never recorded.
  • Too few visitors, or test activity mixed with real visits, can leave a comparison without a reliable winner.
Can it recover missing historical traffic?

No. It reports the gap and helps you make future tracking consistent.

Can it select a winner from a few clicks?

Not reliably. Too few visitors, or test activity mixed with real visits, can leave the comparison undecided.