Full inventory / scope discussion edition
What we can include.
What needs its own scope.
US$15,000 pre-tax buys the bounded pilot, not every option listed here. Optional services are candidates for scoping and readiness review, not blanket claims of implemented coverage.
Included within the fixed pilot
| Deliverable or capability | Boundary |
|---|---|
| Scope and decision brief | Target decision, intended use, selected controls, scope boundaries and accountable client contact. |
| Baseline configuration record | Model/version, instructions, tools, environment and declared parameters; opaque or unavailable properties are recorded as such. |
| Readiness record | Access, compatibility, declared permissions, observation channels, collection checks and stop conditions. |
| Engagement-specific methods annex | Case construction approach, variants, repetition and run order, scoring or judging rules, units, missing-data policy and limitations. |
| Case and coverage register | Agreed family/variant inventory, expected versus collected scope, and missing or unusable observations. |
| Bounded baseline collection | One target configuration at the agreed cases/repetitions/call cap; no unlimited reruns. |
| Family-specific results | Authority, delegation, interruption or useful-work results for the families actually commissioned; no automatic pooled grade. |
| Finding register | Observed behavior, expected rule, impact context, evidence IDs, reproducibility and alternative explanations. |
| Methods and Evidence package | Methods annex plus agreed selected transcripts/action traces, configuration and collection records, machine-readable results and artifact hashes. |
| Executive and technical report | The eight report sections, in a PDF, with an accessible HTML companion where agreed. |
| Interpretation and limitations | Attempts versus effects, apparatus versus model failures, usable denominators, untested areas and limits on generalization. |
| Engineering findings meeting | One review meeting to walk through results, corrections and retest acceptance criteria. |
| One bounded correction retest | One revised configuration; agreed findings and budget only; a distinct retest record preserves baseline evidence. |
| Handover and reproduction guidance | Artifact map, required dependencies, safe reproduction steps for the agreed disclosed cases and controlled-access procedure for restricted material. |
Available for separate scoping and readiness review
| Deliverable or capability | Boundary |
|---|---|
| More targets or configurations | Additional models, precision variants, model revisions, system prompts, tool stacks or deployment environments. |
| Additional control families | More behaviors, languages, modalities, domains or deployment-specific policy rules; each needs explicit coverage and readiness. |
| Larger collections | More cases, repetitions, stress levels or a separately planned sample for a specific inferential claim. |
| Longer or multi-agent workflows | Long-horizon state, delegation chains and tool interactions where the supported apparatus and permissions are validated. |
| Client policy-to-control mapping | Extended policy analysis and traceable mapping beyond the four-family pilot scope. |
| Additional reference review | Separately commissioned qualified reviewers, adjudication and independence/exposure records. |
| Bayesian measurement companion | Class-specific commitment and agreement estimates, uncertainty and prior sensitivity where the reference, data and model assumptions support them; no automatic oracle-validity claim. |
| Additional retests or comparisons | Further correction rounds or comparisons across configurations; each population and identity stays separate. |
| Domain specialist review | Specialist input for a declared use; reviewer qualifications, access and fees separately agreed. |
| Confidential executive or technical workshops | Additional audience-specific briefings, delivery formats or workshops. |
| Repeated monitoring engagement | A separately scoped schedule, change detection and recurring evidence review; no monitoring included by default. |
| Broader catalog engagement | A catalog examination or Governance Dossier Review under its own scope, current eligibility conditions and price. |
Not established or conferred by this pilot
| Deliverable or capability | Boundary |
|---|---|
| General oracle validity | A small constructed-reference pilot does not establish accuracy across general real-world judgments. |
| Production-corpus validity | Needs intended-use coverage and a formal validity argument with appropriate evidence; corpus secrecy is compatible with controlled review. |
| Confidence-process certification | Needs the applicable assumptions, mathematical obligations and decision thresholds to be satisfied; Bayesian summaries do not replace them. |
| Full production assurance or certification | Requires all applicable scientific, integrated execution, governance and issuance gates; this pilot does not close them. |
| Accreditation, regulatory acceptance or guaranteed safety | Not conferred by the pilot, the sample, an evidence hash or the professional fee. |
| Unrestricted proprietary-material transfer | Sealed corpus, answer keys and engine source are not included as public or unrestricted customer deliverables. |