Global 500 Enterprise AI-Affinity: Evaluation Methodology

1. Cohort and ranking boundary

The cohort is the complete 2026 Fortune Global 500. Fortune rank is revenue-based and is retained only as a comparison field; it never enters the AI-affinity score.

• All 500 companies received a baseline screen of official-site entry points, public pages, and bounded endpoints.
• 175 evidence-rich candidates advanced to full five-dimension review.
• Every company that clears the benchmark is published: 108 this edition, with a 60-point cutoff.

2. The 60-point benchmark

A company must show first-party evidence of deployed AI, score at least 15/25 for operating adoption and 7/10 for evidence quality, and reach 60/100 overall.

• 90–100: frontier platform, with strong model, agent/interface, operating, and research evidence.
• 75–89: leader, with multiple production capabilities and at-scale adoption evidence.
• 60–74: meets the AI-affine benchmark, with deployed and reviewable evidence but less breadth in openness or research.

3. AI-agent evaluation pipeline

• Cohort lock: read the official Fortune 2026 list and retain company name, revenue rank, sector, country/territory, and Fortune company page.
• Evidence discovery: search only official corporate, developer, research, newsroom, and open-source properties.
• Bounded probing: inspect the home page, robots.txt, sitemaps, llms.txt, common MCP/OAuth discovery paths, and public AI/agent pages.
• Dimension scoring: apply one rubric across five dimensions; industry reputation cannot substitute for missing public evidence.
• Review and freeze: validate dimension caps, totals, source links, benchmark gates, ordering, and tie-breaks before fixing the version and audit date.

4. Five scoring dimensions

5. Evidence rules and exclusions

• Prefer 2025–2026 first-party material; older evidence is retained only when it still represents a current capability.
• Third-party media, vendor case studies, job posts, and unconfirmed reports may aid discovery but cannot earn points by themselves.
• Buying a model, owning data, or operating in technology does not automatically earn a high score; identifiable product, interface, operating, or research evidence is required.
• Failure to find a public MCP or agent interface affects only the openness dimension and does not prove that private systems are absent.

6. Ordering, ties, and rank lift

Entries are ordered by total score. Ties retain the frozen audit order for this version; no secondary criterion is claimed unless it is actually computed. Fortune rank never breaks a tie. Rank lift equals Fortune rank minus AI-affinity rank and exists only to explain the difference between the two orderings.

7. Known limitations

• Public evidence can understate confidential, regulated, or primarily internal AI capability.
• Multilingual sites, regional properties, dynamic rendering, access controls, and bot defenses affect discoverability.
• Scores are an audit-date snapshot and do not predict future adoption, business outcomes, safety, or investment returns.
• This index is not affiliated with Fortune or any ranked company and does not imply endorsement, partnership, customer status, or investment advice.

8. Reproducibility and provenance

The process ran in two stages and they are disclosed separately. The baseline stage was performed by Codex CLI with gpt-5.6-sol (azure provider): Cohort baseline screen and the initial 100-company edition. This is a publisher statement that predates the verifiable artifacts. The review stage was performed by DeepSeek Harness (DSH) with deepseek/deepseek-v4.1-flash (commandcode provider): Rebuilt the candidate pool, re-reviewed every first-party evidence link, re-scored all entries against one deterministic rubric, and now publishes every benchmark-qualified company. That stage is fully reproducible from the committed artifacts.

Tooling is disclosed because it shaped the process, not the numbers: the published score of a company comes from its evidence profile and the rubric in this document, and both are published. Model attribution is taken from the session records of the runs themselves rather than from machine configuration, which changes over time: the Codex sessions that produced the initial edition record gpt-5.6-sol, and the review-stage default is deepseek/deepseek-v4.1-flash. Stage 2 is fully reproducible from the committed artifacts; stage 1 predates them.

To reproduce: clone this repository and run the commands below from plugins/longarena-investor. Pool construction and scoring are deterministic; authoring evidence profiles is judgement work, so an exactly identical result requires using the same pool/profiles/ directory.

• Exactly reproducible:
• · Cohort fields: rank, name, sector, country/region, Fortune page slug and official site, read from the Fortune Global 500 payload.
• · Bounded baseline screen: fixed path list, 12s timeout, no crawling; every probe recorded with status and outcome.
• · Agent-discovery verification: a bare HTTP 200 is not accepted; JSON content-type or an OAuth/MCP marker is required.
• · Candidate-pool inclusion: three documented routes, applied by script, with the reason recorded per company.
• · Scoring: five dimensions computed from an evidence profile by fixed anchors; the evidence-quality dimension is measured, not supplied by a reviewer.
• · Publication gates and the published set: score >= 60, operating adoption >= 15, evidence quality >= 7, no slot cap.
• · Every acceptance gate: structure invariants, reproducible tie-breaks, cohort cross-check, per-link reachability and content quality.
• Judgement-dependent (reviewable, but not guaranteed identical):
• · Evidence discovery and the content of each evidence profile (which first-party page best represents a company, and how many named products or quantified metrics it supports).
• · Frontier-model tier, granted centrally from a reviewed list rather than by each profile.
• · Registered exceptions where a cited page blocks automated clients or cannot be reached from the reviewer network.
• · Exclusions: a company whose own properties cannot be read is removed and disclosed rather than scored at zero.
• Expect these differences in any reproduction:
• · Live pages change: a reproduction run on a later date may find dead links, or different evidence, and therefore different scores.
• · Network egress matters: bot walls (403/406/412), regional blocks and DNS interception change what a reviewer can read; the committed run records each such case.
• · Client-rendered corporate sites return little static text; rendering is required before concluding a page is empty.
• · Screen recall is limited (measured against known-good companies it re-found only 47 of 100), which is why the candidate pool is a union of three routes rather than the screen alone.
• · Brand and ownership traps exist: similarly named but unrelated companies, and AI that belongs to a subsidiary rather than the parent, both distort scoring if missed.
• Commands:
• $ npm --prefix plugins/longarena-investor run refresh:scan
• $ npm --prefix plugins/longarena-investor run refresh:queue
• $ # review agents write pool/profiles/<rank>-<slug>.json, one per candidate
• $ npm --prefix plugins/longarena-investor run refresh:finalize
• $ npm --prefix plugins/longarena-investor run refresh:gate
• Verification: a reproduction should match the funnel fields in ranking.json (screenedCompanies / candidatePoolSize / fullReviewCompanies / publishedCompanies) and the gates (60 / 15 / 7), then pass the eval gate over the whole result.