Hierarchical publish gate · The UltimateSEO Framework
The publish gate: hard gates first, then the quality score
The gate is not a ranking formula. Eight hard gates are decided first and are never scored: if any one fails, the page is rejected or fixed before anything else. Only when the gates pass does the 100-point quality score apply, and it is always reported together with its evidence coverage.
1. Hard gates (not scored)
If any of the following fails, the decision is reject or fix first, whatever the total score.
- No spam, deception, fake reviews or authors, cloaking or scaled abuse. G-01 G-02 G-03 G-04 G-05 G-06
- The target URL is reachable, has the correct canonical, renders and is indexable. T-01 T-02 T-03 T-05 T-09
- Critical facts are verified and an evidence chain exists. Q-05 Q-06
- Author, business and entity information is real. E-01 E-02 E-06 E-08
- Structured data does not contradict visible reality. R-14
- Local pages do not claim coverage outside the real service area. I-08
- If the topic is YMYL, the required expert and source standard has been applied. E-05
- Transactional facts such as price, stock and availability agree with the source of truth. E-09
The 21 controls that act as hard gates
2. Quality score, only when the hard gates pass
| Main purpose | Weight |
|---|---|
| Intent and user-task fit | 15 |
| Accuracy and evidence chain | 15 |
| Original, non-commodity value | 15 |
| Technical eligibility | 12 |
| Retrievability and UX | 10 |
| Entity and trust | 10 |
| Business usefulness and conversion readiness | 8 |
| Authority and distribution readiness | 5 |
| Freshness and maintenance plan | 5 |
| Measurement and experiment readiness | 5 |
| Total | 100 |
How UltimateSEO computes it
- Only controls in state PASS or FAIL count. UNKNOWN, NOT_APPLICABLE and BLOCKED controls are excluded, never assumed.
- Per bucket: score = pass ÷ (pass + fail) × weight. Buckets with no evidenced control are left out.
- The total is shown with its evidence coverage (evidenced controls ÷ 100) so a high score on thin evidence cannot pass for a verified one.
- If a hard gate fails the decision is gate failed: the score is still shown, labelled "fix gates first".
3. Decision bands
| Score | Decision |
|---|---|
| 90–100 | Publish |
| 80–89 | Publish with minor fixes |
| 70–79 | Rework |
| 0–69 | Reject |
| Any hard gate FAIL | Reject: fix the gate first |
4. Page-level dependency order
An agent or a reviewer assesses a page in this order. Each question is only worth asking once the one before it has a satisfactory answer.
- G Is there a policy or spam problem?
- T Is the URL technically eligible?
- I Is it the right user, intent and page type?
- Q Is the information accurate, sufficient and original?
- R Can people and retrieval systems understand it easily?
- E Are the author, brand and entity trustworthy?
- A Is there an authority or distribution opportunity?
- M How will the result be measured and kept up to date?
5. Change classes used in findings
Every recommended change in a report carries one of these classes so you know what can be applied safely and what needs evidence or approval first.
- AUTO_SAFE: Reversible changes that create no new business claim: an obvious broken link, accidental duplicate metadata, a verified canonical inconsistency, missing semantic HTML.
- AUTO_WITH_TEST: Changes that must be measured after release: title and meta, internal linking, layout order, image optimisation, schema clean-up, copy clarity.
- EVIDENCE_REQUIRED: Changes that need a source of truth before they are made: price, product features, 'tested' or 'best' claims, legal claims, service area, author bio, case study, review, availability.
- APPROVAL_RECOMMENDED: Changes with a wide blast radius that a named owner should approve: deleting or merging high-traffic URLs, sitewide navigation rewrite, brand positioning change, pricing or offer change, robots or crawler policy change, large redirect map.
- DO_NOT_DO: Never: spam, fake authority, fake reviews or personas, cloaking, hidden ranking copy, doorway pages, fabricated evidence, manipulative link schemes.