The compiler is used when a request contains factual, comparative, multi-part or structured obligations that should be observable before evidence collection and publication.
Question Contract Compiler
Turn a question into observable answer obligations.
Extracts provider-neutral facets, relationships, value shapes and cardinality before evidence collection or answer publication.
A model-independent question contract describing the requested facets, relationships, cardinality and coverage obligations, with no expected answer values embedded inside it.
Why it exists
The problem this module is designed to solve.
A response can sound complete while silently omitting half of the request, collapsing two entities into one or returning the wrong number of values. Without an explicit contract, downstream systems cannot distinguish a complete answer from a persuasive fragment.
02 / How it works
A bounded path from need to accountable result.
The public model below describes responsibilities and decisions, not sensitive implementation details, provider secrets or customer data.
- 01
Separate the obligations
Break the request into bounded answer facets without guessing the answer or calling a provider.
- 02
Preserve relationships
Represent which entity, period, source or comparison each requested value belongs to so later evidence cannot be mixed.
- 03
Declare shape and cardinality
Record whether the answer needs one value, several values, a sequence, a comparison or an explicit absence.
- 04
Expose coverage gaps
Give retrieval and assurance a precise description of what is still missing instead of asking them to repeat the whole task.
03 / Customer and operator value
Reduces false completeness and lets downstream modules repair the missing facet instead of repeating the whole task.
Observable coverage before publication
Targeted repair instead of blind retries
No model, network or corpus dependency
Multilingual and typo-aware contract design
04 / Where it creates value
Concrete situations, not generic feature claims.
These are representative product situations. Every deployment still requires its own policy, data boundary and acceptance criteria.
Multi-part executive questions
Keep every requested company, period, measure and comparison visible through the full answer path.
Regulated decision and control reviews
Turn long questions into checkable obligations so missing evidence is reported rather than smoothed over.
Multilingual enterprise search
Preserve requested meaning and answer shape across language variation, typos and differently phrased facets.
05 / Role in the cognitive system
A clear responsibility creates a trustworthy boundary.
No module is allowed to become an invisible monolith. It owns a narrow contract, composes with named capabilities and refuses responsibilities that belong elsewhere.
What it owns
- Requested answer facets
- Cardinality
- Coverage draft
- Gap description
What it composes with
Cognitive EngineAnswer AssuranceRetrievalResearch
What it refuses to own
Boundary before convenience.
It never stores expected answers, retrieves documents, runs models or decides publication policy.
06 / Vision and mission
Accountable intent
The mission begins before evidence or inference: this module makes the user's real request inspectable, so the system can be held accountable for what it answered and what it did not.
AI systems that remain useful, inspectable and sovereign across models, providers and deployment boundaries.
Build the cognitive layer that chooses the smallest sufficient path and turns evidence into accountable action.
Capture the value of advanced AI without surrendering data control, architectural freedom or intellectual honesty.
07 / Evidence and maturity
What the current label means — and what remains open.
ENGINEERING MATURITYM4Mentaview internal laboratory assessment
OBJECTIVE REVIEW
Passed with moderate confidence
The question contract is compiled in the supported kernel path and has deterministic multilingual and metamorphic local evidence; the generated blind human review is not complete.
OPEN LIMITS
- The 40-case blind human review remains pending.
- No broad prospective customer-domain holdout.
- No organizationally independent custody or assurance.
What exists today
The provider-neutral v1 path remains available for simple untyped requests. The v2 compiler and answer contract are integrated into the kernel publication path; a sealed 40-case multilingual Wave 11 machine holdout passed 40/40 with zero additional model calls, and produced a blind-review packet whose 40 human judgments remain pending.
Implemented
A software or contract boundary exists in the current development project. This does not by itself mean general availability or independent certification.
What must earn promotion
Complete the generated 40-case blind human review, then run broader prospective customer-domain holdouts covering ambiguity, passive constructions and meaning-sensitive cases without over-decomposing simple questions.
Public truth boundary: M4− is a non-standard Mentaview engineering label for internal laboratory validation. It is not an official TRL decision and does not assert production qualification, customer acceptance, independent assurance, certification or universal performance. Inspect the complete assessment record.