Governance

This page says who runs the Standardized Token Cost Index, what rules apply, and where the gaps are. The full text is in the repository as document 024.

Administrator

STCI is operated by intentsolutions.io LLC.

Independence is a requirement the administrator holds itself to. It is not an established fact. No third party has reviewed it.

intentsolutions.io LLC does applied AI engineering work. It may buy inference from providers whose prices STCI records. Its purchases do not enter any index. The indices use listed prices only.

Rule changes

Every change to the rulebook is recorded, with its date, in the governance record and the repository change log. The current rules are on the methodology page.

Blend convention

STCI blends 3 parts output to 1 part input. This is an assumption, not a measurement.

Other products use other conventions. Artificial Analysis uses a 7:2:1 mix of cache, input and output. BenchLM is described as 3:1 input to output. Do not compare blended values across indices.

Data retention

Raw source responses, observations and the per-endpoint snapshot are private. Each daily value carries a SHA-256 verification hash. Observations are available on request so that you can recompute a value.

No retention period has been set yet.

IOSCO Principles for Financial Benchmarks

This is a self-assessment. STCI is not a regulated benchmark. No independent review has taken place. Principle names should be checked against the IOSCO text.

No.PrincipleStatusEvidence
1Overall Responsibility of the AdministratorPartlyAdministrator is named; no formal accountability framework.
2Oversight of Third PartiesNot yetSingle source, OpenRouter; no agreement or oversight.
3Conflicts of Interest for AdministratorsPartlyDisclosed here; no register, recusal procedure or independent check.
4Control Framework for AdministratorsPartlyDeterministic code, tests and hash; no written control framework.
5Internal OversightNot yetOne small team; no oversight committee.
6Benchmark DesignPartlyDocumented baskets and blend; blend is an assumption and the source is single.
7Data SufficiencyPartlyCoverage floor and endpoint screen; inputs are listed prices, not transactions.
8Hierarchy of Data InputsPartlySource tiers defined; only tier T1 in use.
9Transparency of Benchmark DeterminationPartlyMethodology and hash public; raw inputs on request only.
10Periodic ReviewPartlyMonthly review specified; no completed cycle under the current rules.
11Content of the MethodologyPartlyRules and limits documented; no expert-judgment or stress-condition policy.
12Changes to the MethodologyPartlyVersioned with a change log; notice period untested; no consultation process.
13Transition PolicyNot yetThe 2026 data gap is disclosed; no cessation policy.
14Submitter Code of ConductNot yetNo submitters yet; the reported-price panel is planned.
15Internal Controls over Data CollectionPartlyExclusions logged and rules re-applied; no check on source errors.
16Complaints ProceduresPartlyContact below; no response time or escalation path.
17AuditsNot yetNo internal or external audit.
18Audit TrailPartlyObservations and hashes kept on the host; no retention period; not held independently.
19Cooperation with Regulatory AuthoritiesNot yetNo procedure; no regulator involved.

Count: Met 0, Partly 13, Not yet 6.

Complaints and corrections

Write to support@intentsolutions.io. Give the date, the index or model, and why the value looks wrong. If a value was wrong, we publish a dated correction. No response time is promised yet.

What a grade is and is not

A grade is a view over published facets, such as serving precision. A content-hashed rule_id identifies the rule that produced it. The same rule over the same facets gives the same grade.

A grade does not test output quality. Precision labels come from the source. No audited grade exists. A grade is not an offer, a quote or financial advice.