Methodology · version 1.0 · 28 Jul 2026

Evidence before excitement.

Cosradar is a decision-support method, not a prediction guarantee. It records what was observed, when and where it was observed, what was inferred, and which evidence is still missing before a beauty manufacturer, packaging supplier or brand should act.

Current public-data status: latest public daily pulse verified 18 Jul 2026; latest regional boards verified 13 Jul 2026. The site shows historical snapshots until the next run passes source and quality checks.
01 · signal model

Four layers that must not be collapsed.

L0 · UPSTREAM

R&D and ingredient movement

Patent, supplier, formulation and trade-show signals. Useful for early watchlists; not consumer demand proof.

L1 · ATTENTION

Search and content heat

Queries, content velocity and discussion. Directional interest; brand queries, seasonality and relative indices are labelled.

L2 · TRANSACTION

Commerce behaviour

Observed sales proxies, rank movement, listings and price events. Promotions and stock changes can confound the signal.

L3 · BRAND ENTRY

Commercial adoption

Verified launches, funding, hiring or assortment moves. Context for timing; never presented as a buyer's procurement intent.

02 · trendscore

A prioritisation score, not a truth machine.

MOMENTUM
30

Change, persistence and cross-source support.

CAPABILITY
25

Can the required product or pack actually be made?

B2B ENTRY
20

Named-brand or commercial adoption signals.

COMPLIANCE
15

Market and claim feasibility.

MARGIN
10

Indicative economics, subject to a real quote.

Decision rule

TrendScore ≥ 70 enters human review. It does not automatically ship. capability = 0 or compliance = 0 vetoes progression regardless of total score.

03 · evidence object

What a publishable claim must carry.

FieldQuestion answeredPublic rule
Subject + marketWhich entity, product, component and country?No scope transfer between suppliers, factories, SKUs or markets.
Source + dateWho issued it, when was it observed and accessed?Undated numbers are not treated as current facts.
Method + unitWhat exactly was measured?Relative rising indices are not labelled absolute volume; rank movement is not labelled sales.
StatusObserved, normalised, inferred or recommended?Inference stays visibly separate from source facts.
LimitationsWhat could make the conclusion wrong?Seasonality, promotion, stock, sampling, access failures and missing runs are disclosed.
Commercial gateCan we quote, sample or claim it now?MOQ, lead time, capacity, price and certification require dated supplier/site/SKU confirmation.
04 · operating loop

How the radar becomes commercial work.

1 · SELECT

Cosradar chooses the question

Which market problem and product format deserve verification now?

2 · PUBLISH

Evidence pages make it answerable

Stable HTML explains the entity, capability, source, date, method and boundary.

3 · MEASURE

COSGEO checks AI visibility

Freeze the question cohort and record mentions, owned-domain citations, entity accuracy and evidence fidelity. A professional baseline uses 60 questions × 6 platforms × 3 rounds = 1,080 observations.

COSGEO AIGEO guide →

4 · LEARN

Content and sales return outcomes

美妆老高聊AI explains the method; outreach tests named accounts; qualified inquiries, samples, quotes and orders feed the next scoring cycle.

05 · update policy

Freshness is a visible product feature.

  • Every public snapshot shows its verification date and source scope.
  • More than seven days old is labelled historical until rechecked.
  • Failed, partial or missing runs are not silently carried forward as current.
  • Changed pages receive truthful sitemap lastmod dates.
  • Corrections preserve the old source date and state what changed.