The trust framework for sign language AI
The SLAT Index (Sign Language Access Trust) is a published, versioned assessment framework: five domains, eleven criteria, and fifty-four indicators with defined evidence requirements, producing computed trust ratings that assessors cannot adjust. Built by a Deaf-led practice for the institutions that must answer for a deployment.
A framework, not an opinion
Sign language AI is being marketed faster than its evidence base has matured, and no public benchmark tells a buyer which claims are demonstrated and which are asserted. The SLAT Index exists to make that distinction reviewable.
It is built the way durable standards are built. The method is written down in a set of governed instruments: an evaluation standard, evidence requirements, a scoring rubric, an assessment procedure, and the supporting manuals, each versioned, each amended on the record. Ratings are computed from evidence under published rules; there is no departure route, and an assessor who disagrees with a computed rating records the disagreement rather than adjusting the result. Every amendment, correction, and decision behind the framework is documented in a public change record, so a reader can trace any rule to the reasoning that produced it.
Evidence before adoption. Trust before deployment.
The framework is Deaf-led by design, not as a slogan: determinations about what signed output actually does are reserved to qualified Deaf reviewers, and findings concerning signed output do not issue without their concurrence.
Five domains. Eleven criteria. Fifty-four indicators.
Each domain asks one question an institution must be able to answer before deployment. Each criterion beneath it is scored from defined indicators with stated evidence requirements: what counts as evidence, who may attest to it, and what class of evidence each conclusion demands.
Linguistic Trust
Whether the system's handling of sign language is disclosed, validated, and examined as language.
- LT-1 Linguistic Disclosure
What the vendor states about the language the system produces, and how precisely. - LT-2 Validation Evidence
Whether performance claims rest on independent validation or on assertion. - LT-3 Observed Output
What the signed output actually does when examined directly by qualified Deaf reviewers.
Transparency Trust
Whether the vendor discloses what a serious buyer needs to know, before being asked.
- TT-1 Transparency and Disclosure
How the system works, what it was trained on, and what its stated limits are. - TT-2 Procurement Readiness
Whether the vendor can answer the questions a public buyer is obligated to ask.
Governance Trust
Whether authority and data are governed accountably.
- GT-1 Deaf and Disabled Governance
Who holds decision-making authority, and whether Deaf and disabled people are in it. - GT-2 Data Governance and Privacy
How signer data, which is biometric and identifiable, is collected, stored, and used.
Accessibility Trust
Whether the system expands access in practice and keeps a human answerable.
- AT-1 Accessibility Impact
Whether deployment widens access or substitutes something weaker for it. - AT-2 Human Oversight
Who catches the error, with what authority, before it reaches the person relying on the output.
Deployment Trust
Whether this system belongs in this setting, with accountability when it fails.
- DT-1 Deployment Context
Whether the evidence supports use in the proposed setting, at its stakes. - DT-2 Institutional Accountability
Who is responsible when something goes wrong, through what mechanism, with what remedy.
The fifty-four indicators, their evidence requirements, scoring thresholds, and worked procedures are published in the licensed SLAT Index volumes. This page describes the architecture; the volumes carry the method.
Ratings that are computed, not negotiated
A SLAT trust rating states what the evidence supports, on a five-tier scale. The rules that assign it are published in the Scoring Rubric, executed arithmetically, and identical for every subject.
-
Tier I
Trusted. Validated claims supported by independent evidence, including direct examination of signed output.
-
Tier II
Conditional. A credible evidence position with identified conditions attached to reliance on it.
-
Tier III
Limited. Evidence too thin to support reliance; the gaps are identified and specific.
-
Tier IV
Defect located. The evidence establishes a material defect. This is a finding about the system.
-
Tier V
Not determinable. The rating could not be determined on the evidence available. This is a finding about the record, not the system.
Evidence cannot be gamed downward
The framework is engineered so that supplying less evidence can never produce a better published rating than supplying more. Disclosure is never punished relative to silence.
Defect and absence are kept distinct
A system with a located defect and a system with an undetermined record are different findings, and the ratings keep them from being confused.
Context sets the bar
A rating is read against the deployment's stakes. The same system may be recommended in one setting and not in another; the deployment classification states which, and why.
Two products, one method
Trust Snapshot
A public-evidence review. It reports what the public record does and does not establish, criterion by criterion, and produces an evidence availability profile rather than a rating: the absence of public evidence is a fact about the record, not the system.
Trust Assessment
The full engagement: formal evidence requests to the vendor and deploying institution, a Deaf Comprehension Review of observed output where scoped, and a computed trust rating with a deployment classification for the assessed setting.
What the process guarantees
- The subject is heard. Vendors receive the evidence requests, a defined window to respond, and a documented reconciliation path where they dispute an observation.
- Signed output is judged by Deaf reviewers. What the output did is determined by qualified Deaf reviewers; findings about signed output do not issue without their concurrence.
- Findings are checked before they publish. Second-assessor review or its published substitute controls, a self-review record, and a cooling period stand between a draft and a finding.
- Independence is structural. A paid assessment is not an endorsement and is not sold as one; commissioning and fee basis are disclosed with the finding.
Assessments scale with deployment stakes: the higher the stakes of the setting, the more the evidence must carry. The full procedure, including panel composition, evidence classes, and publication rules, is stated in the licensed volumes.
Published to be read. Licensed to be used.
Like established accreditation standards, the SLAT Index is published as a body of standards that organizations license in order to apply. The volumes are how the framework reaches you whole: the criteria, the evidence requirements, the scoring rules, and the worked procedures, in force and version-controlled.
Read it
The published volumes, including the complete instrument set and the assessor guide, are available through Publications. The evaluation standard itself is citable by DOI.
Use it
Applying the framework inside your organization, in procurement criteria, internal review, or vendor evaluation, is licensed use. Licensing keeps applications faithful to the published method and keeps every user on the version in force.
Engage it
Commissioning a Trust Snapshot or Trust Assessment requires no license: Novara Consulting Group conducts the engagement under the framework and delivers the finding.
Licensing inquiries, including institutional and multi-seat terms, go through Contact, topic “Publications.” The framework's name and ratings vocabulary are used under license; a rating may be cited, never self-awarded.
Evaluate before you deploy
The cost of a structured assessment is small. The cost of an access failure in a hospital, courtroom, classroom, or emergency is not.
Request an Assessment Request a Trust Snapshot →Request a SLAT assessment
Tell us the vendor or product, the intended use case, the deployment context, and the decision you are trying to make. We respond within one business day.
Scope of SLAT engagements. SLAT engagements are governance, procurement, and risk assessments. They are not legal, medical, technical, clinical, or linguistic advice, and not a government certification. The SLAT Index is an evaluation standard developed by Novara Consulting Group and published as the SLAT Index Evaluation Standard (v2) (DOI 10.5281/zenodo.21537271). Findings rest on evidence available at the time of review, and organizations remain responsible for their own procurement, legal, and compliance determinations.