Research · Methodology
How the Q2 2026 Pulse is built.
The Pulse is an outside-in public-signal benchmark. Company scores are RPS (Raw Public Signal) visibility; market context uses CRP and CERS benchmarks. Adjusted Enterprise Readiness (AER) is not published this release. This page says what each measure is, and what the index does not claim.
From screening universe to published index
Absence from the index is not absence of activity. A candidate that cleared the gate but is not published carries evidence too thin to score, not a verdict on the organisation.
What each number means.
A 0 to 100 company visibility score from five public signals. Visibility, not verified readiness.
RPS with a hiring-mix correction. A visibility indicator, still not readiness.
A macro and state readiness prior per market. Never added to a company score.
A triangulated market signal that shrinks raw company visibility by evidence coverage. Market-level, never a company score averaged up.
Not published this release. The enterprise-adoption benchmark covers 12 of 23 markets; full coverage is not met.
The index does not claim to see inside an organisation. Strategy intent that has not shipped, and budget that has not produced visible output, are not measured. The Pulse is deterministic and identity-filtered: search results are matched to the named organisation before scoring. Companies with identical RPS scores share a score band; within a band, the tie-break is alphabetical by domain and does not distinguish readiness.
Counting companies
Four counts of the same market, and how they reconcile.
A market page, the index table, and the market dossier each report a company count, and the three differ. They are not estimates of one another. Each answers a different question, and the differences are exact rather than approximate.
| Count | Singapore | What it answers |
|---|---|---|
| Screened | 419 | Every candidate organisation assembled for the market. |
| Below the evidence gate | 154 | Too little public evidence to score. Not a claim that these organisations are unready. |
| Scored | 265 | Cleared the gate. The dossier’s working set. |
| Domestic signal entities | 245 | Scored and domestically headquartered. The only count that feeds CERS, so it is the one the market page leads with. |
| Listed rows | 257 | Scored, after groups scoring under several domains are collapsed to their strongest row. What a reader can count in a table. |
The arithmetic
- coverage265 scored + 154 below the gate = 419 screened.
- ownership245 domestic + 20 subsidiaries and delivery centres = 265 scored.
- display265 scored − 8 collapsed duplicate rows = 257 listed.
The collapsed rows are not dropped organisations. A group that scores under more than one domain is one organisation with several records, and it is shown once at its strongest record so a table of 257 rows is a table of 257 organisations. These three identities are asserted for all 22 scored markets at export; a market that stopped reconciling would fail the build rather than publish.
Companies and places
This methodology ranks companies. Places are measured differently.
The index on this page measures organisations and publishes company tiers. AIR APAC also measures how AI systems represent places, and that instrument never ranks one place against another: a place is compared against its own prior measurement and against its published method. The two policies are different because the units are different. How places are measured states that method in full.
Audit appendix
What was checked, and what is still missing.
Known methodology deviations
- coverage_quality (cov_q)Proxy cov_q = 0.5*share(companies with evidence_confidence A/B) + 0.5*min(1, n_effective/50). This stopgap silently drives lambda and therefore CERS; it is not validated. Replace when real coverage data exists.
- services_class R/O/C/PImplemented as NAME-ONLY word-boundary heuristic for precision. HIGH precision, LOW recall. Consequence: the company-level services_class_factor penalty (O->0.5, R->0.0) rarely fires, so the company-level BPO correction relies primarily on (a) the evidence-based company_hiring_intent_factor (official-careers vs third-party job-board counts, which still fires for all rows) and (b) the country posting-bias factor (macro CRP-derived). The offshore-hub hiring dampening is therefore partly macro, not fully company-level. Disclosed per Section 0.1.
- CERS precisionCERS published as integer plus a band spanning the k=75/150/300 sensitivity results. Precise sensitivity values are retained in the audit JSON.
- evidence_confidence A/B/C-DHeuristic from all per-dimension evidence URL host classification + dimension scores. Not validated.
Not yet implemented
full crp v02 three block coverage · innovation talent source block · digital infrastructure source block · local language queries · evidence recency exact dates · anchor ready ratio denominator · parent entity resolution.
Oxford 2024 and IMF 2023 are collinear (r = 0.913), so the CRP blend should be read as one macro signal, not two independent ones. Full source vintage, coverage, and licence notes are on the sources page.
Citation
How to cite this work.
Source: AIR APAC AI Readiness Pulse, Q2 2026. Center for AI Readiness in Asia Pacific (AIR APAC). airapac.org/research. The underlying data is available as a CSV download.
Cite RPS as a public-signal visibility score. Do not describe company rankings as validated readiness, and do not average company scores into a market readiness claim. CERS is published per market as that market’s own context; it is not a cross-market ranking, and no market-to-market comparison of CERS or mean RPS is published (see Technical: why markets aren't ranked against each other).