The SEO score — transparent, versioned, Pro-owned

The Pro scan gives every page a 0-100 SEO score — the single number a RankMath or Yoast migrant looks for. Unlike a black-box grade, this one is fully auditable: every point it deducts traces to exactly one scan issue, and the same set of issues always produces the same number.

score = 100 − Σ penalty(issue) for each scored issue   (floored at 0)

One number, one owner

The numeric score is a Pro feature. It lives on the Pro seo_scan_results record, never back in core's seo_meta (the old seo_score column was removed in Core 3). The free core seo:audit reports per-page pass / warn / fail with no number — the score is the paid value-add.

The rubric

The score is computed from a published, versioned rubric — Rankbeam\Seo\Pro\Scanning\ScoreRubric. Two things define it: an explicit allowlist of issue codes that count, and a fixed penalty per severity.

Severity	Penalty	Meaning
critical	−40	High-impact finding in this rubric.
warning	−15	Finding to investigate soon.
notice	−5	A nice-to-have.

Each code's severity is read straight from the issue registry (the single source of truth) — the rubric never re-derives it. Severity is 1:1 per code precisely so the score stays deterministic.

What the score counts

These are deterministic checks selected by Rankbeam’s product rubric, including heuristics that need editorial interpretation. A critical costs 40, a warning 15, a notice 5; the score is not a prediction of search performance.

Code	Severity	Penalty
missing_title	critical	−40
missing_description	warning	−15
missing_og_image	notice	−5
duplicate_title	warning	−15
duplicate_description	warning	−15
title_too_long	warning	−15
title_too_short	notice	−5
description_too_long	warning	−15
description_too_short	notice	−5
robots_conflict_indexing	critical	−40
robots_conflict_following	warning	−15
noindex_warning	warning	−15
invalid_canonical	critical	−40
cross_domain_canonical	warning	−15
shared_canonical	notice	−5
insecure_canonical	warning	−15
http_error	critical	−40
empty_response	critical	−40
missing_canonical	notice	−5
missing_h1	notice	−5
missing_image_alt	warning	−15
thin_content	notice	−5
mixed_content	warning	−15
canonical_target_broken	critical	−40
canonical_target_redirect	warning	−15
canonical_target_noindex	warning	−15

Because the metadata codes are detected on a model scan and the rendered/network codes only on a URL scan (see the execution classes), a model target's score reflects the metadata checks and a URL target's score reflects the rendered page. A model scan scoring 100 means "no metadata defects", not "the rendered page is perfect" — scan the URL for that.

What the score deliberately does NOT count

These registry codes are excluded on purpose. The exclusions are part of the contract (a test asserts every registry code is either scored or listed here):

Code	Why it's excluded
missing_focus_keyword	Advisory. Gated behind the opt-in seo.keywords.enabled workflow — a page must not score lower for not adopting focus keywords, and the score must not depend on a config flag.
noindex_page	Informational. Being noindex is a deliberate state, not a meta-quality defect. The "noindex with a self-canonical" heuristic is scored via noindex_warning instead.
multiple_h1	Informational. Google tolerates multiple H1s — no multi-H1 penalty.
blocked_url	Absence of evidence. The SsrfGuard refused the fetch, so the page was never checked — not a defect of the page.
canonical_target_blocked	Absence of evidence. The canonical target could not be verified — not a defect of the page.
hreflang_invalid_code, hreflang_missing_self_reference, hreflang_duplicate_code, hreflang_missing_x_default	Advisory (for now). These Pro hreflang codes surface in the scan; the free audit has its own hreflang codes but does not yet move the score — adding it would require a VERSION bump.
html_lang_missing, html_lang_invalid, html_lang_mismatch	Advisory. Language checks are held out of this rubric.
hreflang_not_reciprocal	Advisory. Optional reciprocity check, not scored.
hreflang_target_unverified	Absence of evidence. Reciprocity could not be verified.
aeo_missing_author, aeo_article_missing_date	Advisory. Answer-readiness (AEO) signals — they flag an article missing an author entity or a publish date in the scan + free audit, but do not move the score (would require a VERSION bump).

Keyword density, power words, and the rest of the on-page checklist never enter the score at all — they are advisory checks (a separate pass/warn/fail list), not registry codes.

Versioning — historical scores never silently change

Every persisted score is stamped with the ScoreRubric::VERSION that produced it (rubric_version). Two consequences:

• A new issue code scores nothing until it is deliberately added to the allowlist — so shipping a new check can never retroactively change a stored score. (An allowlist or weight change is itself a rubric change and bumps the version.)
• The score is stored, not recomputed on read. The number you saw last week is the number you see today, alongside the rubric that explains it.

Where it's stored

Each scan upserts one row per target into seo_scan_results:

Column	What it holds
scannable_type / scannable_id	The scored model (null for URL targets).
url	The scored URL.
score	The 0-100 number.
rubric_version	The rubric that produced it.
penalty_total	Raw penalty sum before the 0-floor.
scored_issues	How many issues moved the number.
breakdown	[{code, severity, penalty}, …] — the full trace.
keywords_enabled	The seo.keywords.enabled state at scan time (recorded for transparency; the score does not depend on it).
scan_run_id	The run that scored it (set null, not deleted, when a run is pruned — scores are current state, not run history).
scored_at	When it was scored.

Reading the score

Headless — the latest score for a model:

use Rankbeam\Seo\Pro\Facades\SeoPro;

$result = SeoPro::resultFor($post);

$result?->score;     // e.g. 85
$result?->grade();   // 'A'..'F'
$result?->breakdown; // [['code' => 'cross_domain_canonical', 'severity' => 'warning', 'penalty' => 15]]

php artisan seo-pro:scan-status prints the average site score in its summary. The Filament dashboard shows it as the headline "Avg. SEO score" stat, coloured by grade.

Grade bands

A presentational letter grade derived from the number (the number is the contract):

Score	Grade
90-100	A
75-89	B
50-74	C
25-49	D
0-24	F

The shipping signal (noindex_warning)

noindex_warning fires when a page combines noindex with a self-canonical (a canonical naming its own URL). Rankbeam treats this as a shipping signal to review. A self-canonical does not prove that indexing is intended: the combination can be deliberate. A cross-domain canonical does not trigger this heuristic. The issue carries context.shipping_signal (for example self_canonical), plus the compared canonical and page_url.

Both scanners apply the check. The model scan (PageScanner) compares the stored canonical to the model URL. The rendered URL scan (UrlScanner) escalates a self-canonical noindex page from informational noindex_page to scored noindex_warning. That is why noindex_page itself is excluded: the potential conflict is handled by noindex_warning on either path. Review the page's actual intent before changing its indexing directive.

Configuration

// config/seo-pro.php → 'scan'
'score' => [
    'enabled' => true, // turn the scoring pass + its persistence on/off
],

The allowlist and the weights are not configurable: the score must be deterministic for a given rubric_version across every install, so changing how it is computed is a code-level rubric change, not a setting.