How scoring works
Pass the Sift is an independent tool, not connected to the Civil Service, the NHS or any employer. It reads your statement the way a sift panel does, against the criteria the panel uses, and tells you what it finds. Here is exactly what that means, and what it can't do.
What it checks against
- Civil Service behaviours: the published Success Profiles behaviour descriptors for the grade you choose, on the 7-point scale departments publish.
- Personal and supporting statements: the criteria you paste from the advert or person specification, one by one.
| 1 | Not demonstrated | No positive evidence and/or substantial negative evidence |
| 2 | Minimal demonstration | Limited positive evidence and/or mainly negative evidence |
| 3 | Moderate demonstration | Moderate positive evidence but some negative evidence |
| 4 | Acceptable demonstration | Adequate positive evidence; any negative evidence would not cause concern |
| 5 | Good demonstration | Substantial positive evidence of the behaviour |
| 6 | Strong demonstration | Substantial positive evidence, including some evidence of exceeding expectations at this level |
| 7 | Outstanding demonstration | The evidence wholly exceeds expectations at this level |
How a check works
- Code counts what can be counted: words against the limit, “I” and “we”, figures, sentence length.
- An AI model (Anthropic's Claude) reads the statement against each criterion and decides whether it is evidenced, weak or missing, quoting your exact sentence. It finds the STAR parts, judges whether the evidence fits the grade, and estimates the score a fair panel would give, as a range.
- It ranks the three fixes that would raise the score most and phrases each as a question for you to answer. It is instructed never to write text for you, and we test that it doesn't.
How we test it
No employer publishes officially scored statements, so we can't yet say how close our scores are to real panels. What we do instead: we wrote a set of statements to target scores, with known strengths and known flaws (all “we”, no result, the wrong behaviour, over the word limit, evidence below the grade), and supporting statements where we know which criteria are evidenced. Every change to the scorer is tested on that set: does it rank them correctly, catch the flaws and map the criteria? On the latest run, 98% of scores were within one point of the target.
That shows consistency, not agreement with real panels. For that we need real results: if you apply, please tell us the score the panel gave (there is a box on every report). We will publish accuracy figures only when we have enough real scores to back them.
Limits
- Panels differ, and campaigns set different pass marks. That's why you get a range, not a single number.
- The panel may see things we don't: your CV, other answers, the number of applicants.
- We can only judge what you write. A modest example described well can outscore a big one described vaguely; that is true of real panels too.
- It is not careers, legal or immigration advice.
Why we refuse to write statements: employers' AI rules. Questions: hi@passthesift.co.uk.