CNS 7.1 / GCTS: Grounded Chiral Tensor Synthesis

This is the current research hub for CNS 7.1 / GCTS: Grounded Chiral Tensor Synthesis.

GCTS changes the center of gravity from narrative synthesis to likely-truth ranking under limited, contradictory, and access-controlled evidence. It ranks claims through structured possible worlds, evidence atoms, record-access states, proof traces, contradiction residuals, and calibrated uncertainty.

The core move:

CNS should build a distribution over structured possible worlds, quantify the mismatch between language, logic, evidence, and access states, and emit ranked, confidence-calibrated likely-truth hypotheses with explicit evidence, record dependencies, proof support, and uncertainty.

Current Research Boundary

The ingredients are crowded. Fact verification, source-trust scoring, provenance, probabilistic logic, possible-world semantics, legal evidence models, and missing-data theory all have substantial prior art.

The GCTS research boundary is the integration: typed record-access states, generation-duty-aware missingness, contradiction-preserving claim graphs, strict-proof separation, likely-truth posterior ranking, and runtime oracle-boundary controls in one evidence-first architecture.

The most important distinction is the record layer. A missing record is not collapsed into generic uncertainty. GCTS asks who would control the record, whether ordinary procedure would generate it, whether the event should have been observable, how the record was requested, what production response occurred, and how strongly that access state should affect claim ranking.

What Changed From CNS 2.0

The earlier CNS 2.0 work introduced Structured Narrative Objects, chirality, evidential entanglement, critic pipelines, and generative synthesis. GCTS keeps the useful intuition that productive disagreement has structure, but replaces several loose parts with stricter machinery:

Design Commitments

1. Likely truth is the target. Claims are ranked by calibrated posterior mass across possible worlds, not by direct LLM confidence.
2. Strict proof is separate. A strict claim requires resolvable evidence, zero-temperature closure, and a proof trace.
3. Access states are first-class. Available, inaccessible, sealed, withheld, destroyed, not-generated, unknown, partial, contradicted, and unavailable-at-time records are distinct epistemic states.
4. Absence has prerequisites. Absence affects ranking only through generation duty, expected observability, access path, control, production state, and source or institutional incentives.
5. No runtime oracle. Labels and expert judgments may calibrate the system offline, but runtime ranking must come from evidence, access states, rules, worlds, and calibrated model parameters.
6. Multiverse views are first-class. The output is a ranked set of possible worlds before any single answer is selected.
7. Every claim gets a status. The report distinguishes proven, probable, plausible, record_contingent, conflicted, unsupported, rejected, and insufficient_evidence.

Reading Path

1. Theory for the formal objects, chirality, worlds, score split, and confidence decomposition.
2. Architecture for the runtime pipeline and how it differs from standard fact verification.
3. Oracle Boundary for which inputs are allowed to influence runtime truth ranking.
4. Prior-Art Boundary for the closest neighboring systems and the conservative novelty posture.
5. Record-Access Ontology for the typed access-state model.
6. Adversarial Evidence for missing-record discipline, source control, and selective production.
7. Experiments for the falsifiable test plan.
8. MVP Build for the implementation path.
9. Worked Example for a synthetic case showing how record contingencies affect ranking.
10. References for primary papers, standards, and adjacent systems.
11. Glossary for canonical terms.

Source Status

Source note: adapted from the CNS 7.1 / GCTS research docset generated and revised in May 2026. It presents a buildable research proposal before a completed implementation. Current public framing should remain conservative: GCTS proposes an architecture-level integration. Automated fact verification, provenance, probabilistic logic, and missing-data theory all have substantial prior art.