Methodology
Show the working.
One number, two halves, and every constant that goes into it printed on this page — including the parts that are unfinished. If a figure on your dashboard looks flattering, this is where you check it.
Readiness rubric v7.1.0
Contents
Documents
Every constant on this page is read off the code that computes it, and each section names the file it comes from. Where a number is unsettled, the section says so instead of rounding the doubt away.
01 The index
Readiness is your catalog: what you control, measurable today, moving the moment a fix publishes. Visibility is what the engines actually answered: not directly under your control, and slower. The weighting puts more of the index on the answers than on the preparation, because the answers are the thing a shopper sees.
Both halves stay visible on the dashboard. A single number that jumped the instant a fix was applied would be telling you something that had not happened yet.
index = (Readiness × 0.40) + (Visibility × 0.60)
- Clamped to 0–100.
- With no visibility measurement yet, the composite is readiness alone — an unmeasured half is not scored as a zero.
- Confidence intervals propagate into the composite rather than being dropped at the join.
02 Readiness — the eight components
Weights sum to 1.00. The ceiling is the score Burnish can take a component to, and it is the cap used when projecting lift — not a target it pretends every field can reach. Every one of the eight is closable by work inside your Shopify catalog.
| Component | Weight | Burnish ceiling |
|---|---|---|
| Description quality | 0.22 | 95 |
| FAQ presence | 0.14 | 100 |
| Structured data | 0.12 | 100 |
| Title optimization | 0.12 | 95 |
| Alt-text coverage | 0.10 | 95 |
| Meta description | 0.10 | 90 |
| Meta title | 0.10 | 90 |
| Manifesto | 0.10 | 100 |
The AI-ready bar is 80. Meta title and meta description cap at 90 and the description, title and alt components at 95, so a projection built on those ceilings never quietly promises the last few points of a field Burnish cannot take further.
03 Projected lift
Lift is measured against each component’s ceiling, not against 100. A component already
at its ceiling contributes nothing, and a component that can only reach 90 never
promises the ten points above it. The max(0, …) means a
field scoring above its ceiling contributes zero rather than a negative.
This is the same arithmetic the next scan runs. Publish the fixes, re-scan, and the readiness you get is the readiness that was projected — which is the only reason it is worth printing a projection at all.
Σ max(0, (ceiling − score) × weight)
- Ceilings are the table above; a scorer may supply a lower per-product ceiling, never a higher one.
- Readiness + projected lift = the readiness a full publish realises.
- It is a score projection. It is not a forecast of traffic, revenue or placement.
04 Visibility — two measurements
Visibility is two measurements, held apart on purpose: whether an engine knows you and is right about you, and whether you turn up when buyers ask.
4.1 Representation — does an engine know you, and is it right?
Computed from the fixed twelve-question battery as
breadth × [15 + Quality × 85], where
Quality = 0.6 × sentiment + 0.4 × accuracy. Breadth is
how many engines recognise you at all. A brand no engine recognises scores zero — not a
small number, zero — because there is nothing yet to be right or wrong about.
4.2 Presence — do you turn up when buyers ask?
Computed from your discovery board as
max(named score, proximity floor). Proximity alone —
your category discussed, your store not named — is capped at 45, so a question you are
absent from can never read as leading.
The 50/50 split between them is a placeholder, and the code says so in as many words. It is the least settled number on this page. When it changes, scores from either side of the change will not be directly comparable, and that will be said here too.
05 The instrument
Twelve reputation questions, identical for every store on Burnish. Only the brand name and the category are substituted, and the domain is never shown to the model — it has to recognise you the way a customer would, by name.
Holding the battery fixed is the whole point. If the questions moved with the score, a change would prove nothing: you could not tell a better store from an easier question. Fixed, a change in the answer means something changed — in your catalog, or in the engine. Both are worth knowing, and Burnish records the engine and the date against every answer so the two can be told apart.
Sample size is prompts × engines × cycles over 30 days. Under ten samples, Burnish shows “not enough data yet” instead of a confidence band. A thin sample presented as a precise score is the most common way a measurement product lies.
- The battery
- Twelve questions, identical for every store on Burnish
- Substitution
- Brand name and category only — the domain is never shown to the model
- On install
- All twelve fire
- Steady cadence
- One question every 6 hours — a full cycle every 72 hours
- Firing order
- Round-robin: the stalest question goes next, never a clump
- Discovery board
- Up to 50 questions on Presence and Authority, up to 100 on Omnipresence
- Board tenure
- A question sits at least 14 days before it can rotate out
- Scoring window
- Rolling 21 days, for both Representation and Presence
- Before a band is drawn
- Ten samples. Below that it says “not enough data yet”
06 Attribution
What the revenue page is allowed to say, and on what evidence.
| Canonical source | Your Shopify orders — totals, currency, timestamps, refunds, the test-mode flag. The pinned query reads no customer name, email, phone or address. |
| First-party signal | Burnish’s own web pixel, subscribed to exactly three events: page viewed, product added to cart, checkout completed. It stores no personal data. |
| Cross-check | GA4, optional. It corroborates the orders read; it is not what the number is built on. |
| Confidence | High — orders and GA4 agree within ±15%. Medium — orders only. Low — GA4 only. |
| Cohort window | 60 days: a 30-day baseline and a 30-day post period. |
| Before it reports anything | Seven days and three measurement points. Until then it says it has insufficient data and how many days are left, rather than showing an early number. |
| Units | Per-action lift is reported in index points. Dollar figures stay empty until a revenue source is connected — they are never modelled. |
The published position, unchanged: correlation-based, not a randomized controlled trial. Burnish shows the work you approved next to what happened afterwards, with the dates intact.
07 What this method cannot tell you
A measurement product is only worth the caveats it publishes. These are ours, in the order they are most likely to matter to you.
7.1 A rubric bump breaks comparability
The readiness rubric is versioned. A major bump redistributes weights, and scores computed before one are not comparable to scores computed after it. Burnish will not draw a long trend line across a version boundary, and neither should you.
7.2 The visibility split is a placeholder
Representation and Presence are currently combined 50/50. That number is marked in the code as a placeholder, not a designed weighting — it is the half of the methodology least settled. Saying so here is cheaper than you discovering it on a chart later.
7.3 Attribution is correlational
Burnish observes AI-referred sessions and orders alongside the changes you approved, over a 60-day cohort. It does not run a randomized controlled trial, does not hold out a control group, and does not claim your revenue moved because of the fix.
7.4 The engines are not deterministic
The same question asked twice can return two answers. That is handled by sampling repeatedly, scoring over 21 days and refusing to report a confidence band under ten samples — not by presenting a single answer as settled.
7.5 Readiness moves before visibility
A published fix lands in readiness at the next scan. Whether an engine changes what it says is a separate, later, measured question — which is exactly why the index keeps the two halves visible instead of blending them into one movement.
7.6 Monthly answer figures are allowances
The “AI answers analysed” number on each plan is the allowance that plan includes. It is not a measured count of how many answers your store has generated.
7.7 Rankings are not measured
Burnish V1 measures and fixes for answer engines. It does not track Google positions, is not connected to Search Console, and makes no claim about search rankings in either direction.