Coming soon
The sample result holds 300 synthetic claims and 7,402 events. One rule runs across all of them: whoever assesses a claim must not approve it.
01 / Finding
How separation-of-duties checks work
The rule is separation of duties: the person who assesses a claim must not approve it. The Engine checks it on every claim that has both steps and a person on record.
109 of 224
That is 49% of the claims the rule could check. The other 76 could not be checked: a step or a person is missing from the records.
02 / Proof
Each flagged claim identifies the assessment, the approval and their source rows. Follow that trail through the result file, then run an independent SQL recount. The guide takes you through all five steps.
Run the five steps on the sampleHypothetical planning exercise
The finding gives you a count of claims to inspect. Add a review-time allowance and an hourly cost to plan the work. This hypothetical exercise combines the published sample with two planning assumptions. It is not an Observatory price, a measured customer outcome or an Engine calculation.
| Input or result | Value | Basis |
|---|---|---|
| Claims to inspect | 109 | Rule hits in the published synthetic sample |
| Review time per claim | 10 minutes | Planning assumption |
| Reviewer cost | £40 per hour | Planning assumption, including employment costs |
| Planned reviewer time | 109 × 10 = 1,090 minutes | 18 hours 10 minutes of review capacity |
| Planned review budget | (1,090 ÷ 60) × £40 = £726.67 | Budget derived from the count and assumptions |
Use your own review-time allowance and cost rate for a real investigation. This estimates the review budget. A claim about financial losses or savings needs separate evidence.
Discuss a pilot using your exports03 / Refusals
Rejected rows are excluded from calculations and listed with their source file, row number and reason.
The sample records each rejection explicitly. Here is what failed and where. Row IDs count data records, excluding the CSV header.
In claims_ops.csv record 2236, case_id and activity are empty. In step_export.csv record 321, start_time is 2026/02/30 25:61:00.000, an invalid date and time. Record 322 has end_time 2026/03/24 07:00:00.042, before its start_time of 2026/03/24 07:06:18.501.
04 / Next
The worked example preserves the earlier Reader preview of this result. The Reader tour shows a separate development demo. Explore the current views or see how a YAML contract maps your own exports.
Tour the Reader What it reads See the installation stepsWork with the source
Start with a precise ID, declared order and trustworthy timestamps.
Open the event log guideThe published claims result is a sample, with its own Engine version. Keep that scope when checking its SQL; use the input guide when moving to a real export.
