How the readiness score works

DullDocs tells you how ready your evidence is for inspection. This page publishes exactly how that score is calculated — the weights, the thresholds, the hard rules, and the things the product is forbidden from saying. Every number below is read from the same code that scores your setting, and our automated tests fail if this page and the engine ever disagree.

DullDocs helps organise, monitor and prepare document evidence. It does not replace professional compliance, safeguarding, fire-safety, legal, Ofsted or CQC advice, and does not certify compliance.

1. What the score is

Every evidence requirement has a severity. Severities carry different weights — a missing critical item hurts far more than a missing low-priority one:

• critical — weight 5
• high — weight 3
• medium — weight 2
• low — weight 1

Each requirement then earns a fraction of its weight based on its current status:

• Up to date — 100% credit
• Overridden (a recorded human decision) — 100% credit
• Due soon — 70% credit
• Needs review — 50% credit
• Review due — 40% credit
• Expired — 0% credit
• Missing — 0% credit
• Not applicable — excluded from scoring entirely

The score is simply earned ÷ possible, as a percentage. No AI, no judgement calls, no black box — the same inputs always produce the same score.

2. Why an empty setting reads amber, not green

A setting with no applicable requirements yet (nothing provisioned, or everything marked not-applicable) is not assessed — the engine returns a score of 0 and a status of “amber”, never a green 100%. You cannot look inspection-ready by having nothing to inspect. (That result is computed live on this page by calling the engine with an empty setting: assessed = false.)

3. The hard rules no percentage can override

• Any critical requirement that is missing or expired forces the overall status to red — regardless of the score. Worked example, computed live: a setting scoring 86% with one critical item missing reads “red”.
• Any critical item that is due soon, stale or needs review caps the status at amber.
• Otherwise: below 60% is red, 60–84% is amber, and 85% or above is green.
• Marking a requirement not applicable requires a written reason, which is shown wherever the requirement appears — including on generated reports. Critical items additionally require explicit confirmation.
• A human override is the only way to bypass the engine — it is labelled as an override, carries its reason, and is audited.

4. What happens when you upload a blurry invoice instead of a gas certificate

Evidence must pass a deterministic quality gate before it can satisfy anything. A document below a confidence of 0.7, or carrying any of the following quality problems, is capped at “needs review” — it can never turn a requirement green:

• blurry_or_unreadable
• expired_version
• invoice_not_certificate
• missing_page
• missing_required_fields
• quote_not_certificate
• random_image_or_non_document
• unacceptable_evidence
• unknown
• wrong_company
• wrong_person
• wrong_site

Worked example, computed live: a clearly-classified invoice flagged blurry_or_unreadable is blocked by the gate (can satisfy: false).

5. AI never decides

AI in DullDocs reads and suggests; deterministic rules decide. Final requirement statuses are set only by the engine described on this page. Specifically:

• AI classification attaches automatically only at 90%+ confidence; 70–90% attaches marked “needs confirmation”; below 40% a document stays unclassified.
• An AI-extracted date that no deterministic read corroborates routes the requirement to needs review instead of marking it valid.
• Every AI suggestion is reviewable and correctable by a human, and corrections win.

6. Old policies can’t hide behind a future review date

For 6 key policies (safeguarding, complaints, medicine, allergy, mobile-phone and recruitment), a policy whose last review or issue date is before the framework threshold date (2024-09-01) is downgraded to needs review — a future “next review” date is deliberately not accepted as proof of currency. Worked example, computed live: a policy last reviewed May 2023 is flagged: true.

Transition note (EYFS 2025): our citations reference the EYFS 2025 edition (in force since 1 September 2025), but this vintage check deliberately still tests policies against the 2024 threshold above, so no setting’s statuses changed when the citations were refreshed. Policies should now be reviewed against EYFS 2025; the threshold will move to 1 September 2025 on a date announced at least 30 days in advance, after which policies not reviewed since then will show needs review.

7. Coverage that can’t be gamed by deleting a record

Some requirements are derived from live data rather than documents. Paediatric first-aid cover, for example, is valid only while at least one active staff member at the site holds in-date evidence — deactivating your only first-aider makes the site read missing, not greener. (Honest limitation: staff are currently modelled at a single site each, and DullDocs does not compute rotas or ratios.)

8. Why we don’t cite KCSiE for early years

Every nursery requirement in DullDocs cites its source in the EYFS 2025 (group and school-based providers) statutory framework (or the named statute where an item is not directly an EYFS requirement), and those citations are printed in every generated report’s source appendix. We deliberately do not cite Keeping Children Safe in Education: KCSiE is statutory guidance for schools and colleges, not early-years settings. Citing it would make a checklist look more thorough while being less accurate — and an automated test in our codebase fails if a KCSiE reference is ever introduced into the early-years sources.

9. What DullDocs will never tell you

DullDocs measures evidence-readiness. It does not certify compliance, and the following phrases are contractually banned from the product — they are enforced inside every AI prompt and pinned by automated tests:

• “you are compliant”
• “you will pass inspection”
• “this satisfies CQC”
• “this satisfies Ofsted”
• “you legally do not need this”
• “this is definitely valid”

The language the product is allowed to use instead:

• “evidence found”
• “evidence not found”
• “appears to match”
• “appears expired”
• “appears due for review”
• “needs review”
• “based on your checklist settings”
• “marked not applicable by user”

If a vendor tells you their score means you’ll pass an inspection, ask them to publish how it’s calculated. Ours is above.