Reference · development snapshot
From evidence to an inspectable world view
A connected reference architecture shows how incomplete reports and telemetry can become qualified claims, maintained revisions and task-specific views. Each component carries an explicit boundary and limit.
Maneuver Calculus supplies behavioral and defensive semantics—including P/M/Q and coverage reasoning—within this larger design. Maneuver Calculus is still in development. Reference architecture means intended responsibilities, not a claim that every capability is accepted, implemented or verified.
One connected system
Intended responsibility map, not an executed call graph. Stages may repeat or be skipped by a particular task. Select a named component for its role and limits.
Main flow: Ingestion and normalization, Semantic compiler and reasoner, World-state store and runtime, Query and projection, then Output compiler and targets. Query also feeds Behavioral and analytical methods, which returns to the Semantic compiler. When an action is appropriate, output reaches Execution and feedback; authorization and prerequisite revalidation gate an external executor. Attributable observations return to Ingestion. Semantic registry, Data model, and Evaluation and conformance are shared across all boundaries. Storage and governance are supporting boundaries, not sequential processing stages.
Preserve source identity, time and custody
↶ Feedback returns through Ingestion and normalization, then compilation and admission. A requested action is never proof of coverage.
Governance: semantic-version and context changes govern the Semantic registry — C1, Semantic compiler — C4 and World-state runtime — C5. Changing meanings is distinct from admitting new facts; disclosure and execution require their own authority.
Component responsibilities
Open any responsibility. Names describe logical roles, not packages. The Semantic compiler and reasoner (C4) computes qualified consequences; the Output compiler and targets (C8) translates a result for a destination while preserving its limits.
Semantic registry and environment loader C1
Definitions → resolve supported meanings → a pinned semantic environment. Loading a definition does not authorize its implications.
Semantic compiler and reasoner — C4 uses the pinned semantic environment supplied by the registry.
5 Oct 2026 realization: Pinned contracts and version checks exist; a general registry and imported-definition authority are not established. Source snapshot.
Data model and serialization C2
Claims and qualifications → validate and encode → structured records. A valid shape does not establish truth.
World-state store and runtime — C5 carries qualifications into retained history.
5 Oct 2026 realization: JSON histories and provenance fields exist; complete proposed type and unknown-value contracts remain unestablished. Source snapshot.
Ingestion and normalization C3
Source material → preserve custody and interpret → source-linked candidates. Receipt and interpretation are separate.
Semantic compiler and reasoner — C4 checks candidate meaning before admission.
5 Oct 2026 realization: Bounded custody and provisional extraction exist; arbitrary-source support and interpretation faithfulness are not guaranteed. Source snapshot.
Semantic compiler and reasoner C4
Candidates, rules and context → check and compute → qualified assessments or diagnostics. Compilation does not itself accept a world revision.
World-state store and runtime — C5 publishes an accepted world revision after admission checks.
5 Oct 2026 realization: A finite acyclic native engine exists, with a recorded identity-join failure. Separate ontology paths have temporal/context and measured-evidence carriage failures. Source snapshot.
World-state store and runtime C5
Accepted changes → append history and maintain revisions → addressable snapshots. Admission requires a current-head check and atomic acceptance.
Query and projection — C6 queries a specified revision.
5 Oct 2026 realization: History, snapshots and replay exist; general hypothetical branching and a general semantic comparison of revisions are not established. Source snapshot.
Query and projection C6
Revision, task and perspective → select qualified results → a scoped view. Observer perspective does not confer disclosure authority.
Behavioral and analytical methods — C7 uses the selected view for analysis.
5 Oct 2026 realization: Named projections and bounded retrieval exist; general reduction checking and multi-principal confidentiality are not established. Source snapshot.
Behavioral and analytical methods C7
Qualified state and method → analyze behaviors or alternatives → method-qualified findings. Possibility is not occurrence.
Semantic compiler and reasoner — C4 applies core semantics to analytical inputs.
5 Oct 2026 realization: Bounded policy evaluation and comparisons exist; general maneuver composition and hypothetical simulation remain unimplemented. Source snapshot.
Output compiler and targets C8
Qualified result and target contract → translate or refuse → an artifact with retained limits. Output is not execution.
Execution and feedback — C9 receives a proposed action when appropriate.
5 Oct 2026 realization: Qualified JSON and summaries exist; executable target compilation with independent preservation checks is not established. Source snapshot.
Execution and feedback C9
Proposal, authority and prerequisites → revalidate and dispatch externally → separate attempt and effect records. Command acceptance does not prove success.
Ingestion and normalization — C3 receives attributable feedback as new evidence.
5 Oct 2026 realization: Decision records and adapter observations exist; a general governed action engine is not established. Source snapshot.
Evaluation and conformance C10
Contracts and independent expectations → test bounded behavior → scoped findings and retained failures. Passing fixtures do not prove whole-system conformance.
Semantic compiler and reasoner — C4 is one of the responsibilities subject to checks.
5 Oct 2026 realization: Retained tests and independent probes exist; this page runs no semantic tests and claims no full conformance. Source snapshot.
Technical-layer crosswalk
| Technical layer | Participating responsibilities |
|---|---|
| Semantic contracts & representations L1 | Semantic registry C1; Data model C2; Evaluation and conformance C10 |
| Compiler & computation engines L2 | Semantic registry C1; Data model C2; Ingestion C3; Semantic compiler C4; Query and projection C6; Behavioral methods C7; Output compiler C8; Evaluation and conformance C10 |
| Persistent World runtime L3 | Data model C2; Ingestion C3; World-state runtime C5; Query and projection C6; Evaluation and conformance C10 |
| Application services L4 | Ingestion C3; Query and projection C6; Behavioral methods C7; Output compiler C8; Execution and feedback C9; Evaluation and conformance C10 |
| API & protocol interfaces L5 | Data model C2; Query and projection C6; Output compiler C8; Evaluation and conformance C10 |
| Applications & interaction L6 | Evaluation and conformance C10 |
Applications consume capabilities through interfaces. In this reference crosswalk, L6’s sole direct association is Evaluation and conformance (C10); that does not make evaluation the only application capability. Supporting adapter, storage and governance boundaries are not additional numbered layers.
Five connected workflows
- Select tests that actually answer the threat question select · choose a justified path
- Explain what a validation run actually established interpret · keep observations distinct
- Verify that a change fixed the named gap verify · compare baseline and retest
↶ Findings change the next test selection.
↳ Continue a mission correctly after something changes continue — revise dependent conclusions while preserving history.
↳ Reproduce and challenge an analysis challenge — inspect and test the retained reasoning independently.
These are intended workflows. The six staged test templates remain selected blank masters, not completed runs or the withdrawn S1–S6 tests.
Six staged tests
Selected blank master snapshot · 5 October 2026. Not run records each template’s unfilled state; it does not erase historical experiments. The ladder shows intended progression; ordering and accumulation rules remain unresolved.
- T1 Complete Controlled CaseNot run
- T2 Missing EvidenceNot run
- T3 Meaningful ChoiceNot run
- T4 Changed PremiseNot run
- T5 Live ExecutionNot run
- T6 Transfer and Practical ValueNot run
Read each stage’s purpose and limits. No completed runs, scores or thresholds are inferred here.
Illustrative changed-evidence trace · A sensor’s reported active interval is corrected.
Retain correction → recompile affected support → admit a new revision → reassess the scoped view
The prior revision remains inspectable. A detection conclusion may become unknown when its evidence falls outside the relevant time; independent conclusions should stay stable. This describes the intended result, not a newly measured test outcome.
Local terms in this map
- Candidate interpretation.
- A proposed meaning extracted from a source.
- Provenance.
- The source and transformation history supporting a claim.
- Semantic compilation.
- Checking and computing against pinned meanings.
- Admitted claim.
- A claim accepted under the stated rules for adding claims to a world revision.
- Maintained world revision.
- An addressable accepted state with retained history.
- Projection.
- A qualified view selected for a task.
- Valid time.
- When a claim applies in the modeled world.
- Perspective.
- The modeled observer context; not disclosure permission.
See the Glossary for MC core terms.
Realization and remaining limits
The local implementation connects bounded ingestion, compilation, history and named views. General belief closure, planning and governed action execution remain unimplemented or unestablished. Roadmap: current state separates these capabilities from known failures; the source snapshot records the separately versioned development parts.
Source snapshot · 5 October 2026: Architecture v4, Component index v3, Baseline v3, Evidence v6. These local notes summarize draft development documentation; the underlying contracts and retained evidence remain authoritative. This framing does not record architectural adoption or resolve open semantic decisions.