Home / System / Publication QA
Publication QA Rubric v2.0

Two gates, one verdict.

No article, landing page or station page is approved on judgment alone. Every piece passes through the same two-gate audit and comes out with one grade, one list of defects, and a yes or no. Here is the mechanism.

2
Gates, in fixed order
9
Binary blocking checks
5
Scored categories
100
Points, one number
01 The mechanism

Facts are checked before quality is judged.

The two gates ask different kinds of question. Gate 1 asks nine questions with factual answers — is there one H1, do the schema and the visible text match, is the pricing sourced. Gate 2 asks how good this is, and answers with a number. Facts come first, because a beautifully written page with invented statistics is not a near-miss. It is a liability.

THE TWO-GATE AUDIT · FIXED ORDER DRAFT SUBMITTED GATE 1 · BINARY 9 FACTUAL CHECKS PASS GATE 2 · SCORED 5 CATEGORIES / 100 READY SCORE 75+ ANY ONE CHECK FAILS · SCORE NEVER CONSULTED NOT READY AT ANY GRADE THE WINE PATH IS THE ONE WORTH REMEMBERING · A SINGLE FAILED FACTUAL CHECK SENDS CONTENT STRAIGHT TO NOT READY, AND THE QUALITY SCORE IS NEVER CONSULTED
02 Gate 1 · binary Pass / fail · nothing weighted

Nine checks with no room to negotiate.

These are facts, not opinions, so they are recorded as pass or fail with the exact location of the defect and what fixing it requires. Nothing here is scored, weighted, or traded off against anything else.

01 Factual accuracy No invented statistics, regulations or pricing. Unknowns are flagged, never guessed.
02 Terminology The market's primary terms, consistently applied, in the correct regional English.
03 Heading structure One H1, a logical H2/H3 hierarchy, no skipped levels, descriptive headings.
04 FAQ compliance A descriptive section heading, three to five questions, a bold standalone first answer.
05 Schema parity Structured data and visible copy match exactly. No schema-only wording.
06 Link format Internal links in the required form, real post IDs, no “click here” anchors.
07 Search intent match The page answers the query it targets. Wrong-intent content cannot be graded into readiness.
08 Technical compliance Alt text present, indexability and canonical concerns flagged, structure CMS-ready.
09 Brand compliance No sales-heavy language, no corporate jargon, no claims that misrepresent the offer.
Override rule

One blocker means NOT READY, at any grade, including A+. Quality cannot buy its way past a factual defect, so the two signals never get quietly averaged into “good enough”.

03 Gate 2 · weighted

Five categories, one hundred points.

Content that clears Gate 1 is scored on five categories at fixed weights. The weighting is the editorial position made explicit: whether the page answers the reader's actual question is worth twice as much as how it sounds.

RUBRIC WEIGHTS · 100 POINTS TOTAL INTENT SATISFACTION INFORMATION ARCHITECTURE SEO & AEO EXECUTION VOICE & TONE CONVERSION SUPPORT 30 20 20 15 15

Six review lenses feed these five numbers rather than being scored separately — SEO, answer-engine readiness, editorial quality, engagement, brand and conversion.

92 – 100 A+ Best-in-class. Publish and use as a reference example.
86 – 91 A Strong. Publish.
80 – 85 A− Solid. Publish, with the high-impact notes applied where cheap.
75 – 79 B+ Publishable floor. Worth a revision pass first.
below 75 REWORK Does not go live. Back to the writer with cited sections.

The weights shift with the dominant content type before scoring, and the audit states what it moved. Pages close to conversion are weighted for local relevance, booking information and commercial clarity. Informational pages are weighted for depth and answer-readiness. Inspiration-led travel content gets an explicit exception: it is not penalised for being less machine-extractable when the reader experience is stronger for it — destination writing does not become an operations manual to please a snippet.

04 The output

Every audit returns the same eight things.

The format is fixed, which is what makes results comparable across writers, content types and months.

Content type

Detected, with any weight adjustment named.

Publication status

Ready, or Not Ready.

Grade and score

One letter, one number out of 100.

Blockers

What, where exactly, and what fixing it takes.

Category scores

All five, at their weights.

Strengths

What the piece does well, so it gets repeated.

Improvements

High, medium and polish — each cited to a section.

Final verdict

Publish or revise, with the reasoning.

05 Deliberate limits

What the gate does not do.

It does not rewrite The audit reports; a human or a writing step fixes. Mixing the two would mean the reviewer grading its own work.
It does not invent standards The rubric lives in one shared source. If that source cannot be loaded, the audit stops and says so rather than improvising criteria that look authoritative.
It does not give generic feedback A note that could apply to any article is not a finding. Every recommendation names its section and its impact.
It does not publish The gate produces a verdict. Going live stays a human decision.
It does not smooth over disagreement The same rubric runs mid-draft and at the gate. When the two grades differ materially, that gap is reported as signal, not averaged away.
06 Scope and limits
What this describes

The audit mechanism and rubric structure I designed and operate. Weights, blockers, grade bands and output format are shown; the scoring guidance behind each category is not published.

Limits

The rubric grades a draft against a defined standard. It does not measure published performance — that is what the case-study evidence is for. Rubric v2.0, described as operated in August 2026.

See the full workflow library ↗

07 — Contact

Discuss an in-house role.

meililwyss@gmail.com ↗ Back to the system ↗