Using the assessment
The assessment is a single web page that runs entirely in your browser. Your answers are stored only in that browser and nothing is sent anywhere. You can also download the file and use it offline, which some organizations will prefer for something this candid.
Open the assessment Download the file
It opens with example answers so you can see the results first. Choose Start fresh to enter your own. It takes about 15 minutes.
What you enter
| Input | What you enter |
|---|---|
| Goal | More output, lower cost, lower risk, or a mix |
| Stage timing | Elapsed days a typical change spends at each of seven stages, the hands-on part if you know it, and which stages AI has already sped up |
| Statements | 24 statements rated 0 to 4, one at a time or all at once |
| Baseline metrics | Optional: lead time, review pickup, merges with zero review, revert rate, incidents per 100 changes, CI queue time, time to roll back |
In one at a time mode, keys 0 to 4 answer and the arrow keys move. Each dimension has a "what counts as evidence" hint.
What you get
- Readiness level (0 to 4) and your weakest dimension. A strong dimension does not offset a failing one. See levels.
- Autonomy floor and tiers: which change types your evidence supports running with less review.
- First moves for your two weakest dimensions.
- A summary you can copy into a document.
Where a change waits
Each stage is a block sized by elapsed days, split into hands-on work (solid) and waiting (light). The constraint is the longest stage AI has not yet sped up. Two sliders change one stage at a time: work (up to 20 times faster) and waiting (cut by up to 90%). The headline number is the speedup for the whole path.
The lesson usually arrives in seconds: speeding up the work at one stage barely moves the total, while cutting waiting at the constraint does.
Compare with a colleague
Every result has a short code, such as ABSORB1-211222212221322322121221B. It holds only the 24 ratings. Two people score the same team on their own, then paste each other's code.
The comparison shows both sets of scores side by side, the average gap per dimension (the framework aims for 0.5 or less between two raters), and the statements where you differ by 2 points or more, under Talk about these first. Those disagreements are usually about evidence: one person has seen something the other has not.
Decision record
Write down the constraint you believe in, the one change you will make, and a prediction, for example lead time from 32 to 24 days by 31 January. Save a snapshot.
At the next assessment, re-enter the stage timings and the actual value. The record says whether the prediction was met, partly met or not met, and whether the constraint moved since the snapshot. Over a few rounds this shows how well your team reads its own system.
Privacy
- No accounts, no cookies, no analytics, and no network requests from the assessment page.
- Answers live in your browser's local storage on your device. Clearing site data removes them.
- Result codes contain only the 24 ratings. Share them the way you would share the scores themselves.
Versions
The page on this site is copied from a tagged release of the framework repository. Every release is archived on Zenodo under DOI 10.5281/zenodo.23895787.