Catalogue matching, coverage buckets, and the CI Score
Understand confirmed product identity, unresolved variants, recommendation coverage, and why the versioned CI Score differs from a single coverage percentage.
On this page
Identity is not availabilityThe reviewed confirmation ruleFive review statesCoverage buckets must add upThe CI Score is a composite indexDo not substitute headline coverage into an older scoreHistorical compatibilityResolve mapping before procurementIdentity is not availability
A confirmed match links an observed product to an active catalogue entry. It does not certify inventory, fulfilment, an unchanged price, or that every observed variant is offered. A product imported last week can be correctly identified and commercially stale today. Catalogue freshness and product identity are separate checks.
The reviewed confirmation rule
The shared matching rule requires a linked active catalogue product. A manually confirmed state qualifies; an automated matched state qualifies only with a score from 0.95 through 1 inclusive. The match score is an internal matching threshold, not a calibrated 95% guarantee. A record labelled confirmed without a usable active linkage should not be treated as a current catalogue hit.
Simplified confirmation rule
active linked catalogue product
AND (
status is confirmed
OR (status is matched AND 0.95 <= match score <= 1)
)
Do not replace this with 'the names look similar'.Five review states
The review state explains why a product does or does not contribute to confirmed coverage.
| State | Meaning | Next check |
|---|---|---|
| Confirmed match | Active product identity meets the confirmation rule | Verify the offered variant and current source date |
| No catalogue match | The product is marked not carried in the comparison | Review identity and the catalogue before asserting a real absence |
| Possible mapping | A candidate mapping exists but does not qualify as confirmed | Resolve model, variant, and linkage |
| Identity unresolved | No defensible catalogue identity is established | Review the original product evidence |
| Catalogue not connected | A comparison basis is unavailable | Connect the intended catalogue |
Coverage buckets must add up
The recommendation coverage distribution partitions the selected product recommendation units into confirmed coverage, units without a catalogue hit, and open mappings. These buckets should be mutually exclusive and sum to the total. A product-card count is not necessarily the same unit as recommendation coverage: one product can contribute units in several answers. Check whether a table is counting distinct products or answer–product pairs.
The CI Score is a composite index
The reviewed ci-score-v2 formula combines relevant-question coverage at 40%, relative recommendation position at 30%, a legacy assortment-fit component at 15%, and competitive position at 15%. Competitive position uses 70% relative position plus 30% provider agreement, with a documented baseline of 0.3 when that agreement is unavailable. Values are clamped to the applicable range, and the available component weights are renormalised. A connected catalogue is required. This differs from AEO V2's missing-component policy.
Do not substitute headline coverage into an older score
In the reviewed CI overview implementation, the score's assortment-fit component is based on frequently mentioned products with carried status. The newer headline catalogue-coverage metric uses confirmed recommendation units. They are different populations and confirmation rules. The score's own component breakdown and formula version are authoritative for reconstructing it; inserting the headline coverage percentage into a legacy formula can produce a different result.
Composite CI score, not the coverage card
score = round(100 × sum(component value × weight) / sum(available weights))
Weights: question coverage .40; position .30; assortment fit .15; competition .15
Requires a catalogue; missing components and small samples remain visible.
A 54 CI Score does not mean 54% of recommended products are carried.Historical compatibility
A saved score can retain older matching or aggregation behaviour. Changing a mapping can improve future coverage without retroactively proving an old answer had a confirmed identity at collection time. Compare snapshots with compatible formula, panel, source period, and catalogue interpretation. Do not label a change caused solely by revised matching logic as increased external AI visibility.
Resolve mapping before procurement
When an opportunity is popular but open, first verify the exact product and catalogue record. When it is genuinely absent, assess merchant constraints such as category strategy, supplier access, margin, regional rights, returns, and stock risk outside the recommendation metric. ShopFlare evidence helps prioritise review; it does not place an order or establish profitability.
Still need help?
Tell us what you expected and what happened.