Did the meaning survive?
Most systems prove what happened. KYE™ also asks whether the meaning survived. The Continuity Module™ checks for drift across intent, interpretation, memory, context, incentives, timing, and handoff.
Authority → execution is incomplete.
In agentic systems, an action's meaning can drift before it runs. The EU AI Act Art. 13 and NIST AI RMF Measure-2.5 both want you to show meaning was preserved end-to-end, not just that authority was granted.
Most AI governance products model:
authority → execution
KYE™ models the agentic reality:
intent → interpretation → memory → context → incentives → timing → handoff
→ state → proposal → admit → authority → commit → evidence
The Continuity Module™ catches drift before it becomes a commit.
Five questions. Five answers.
Provenance, attribution, authority, admissibility, and continuity are five distinct guarantees. KYE™ binds all five into one replayable record. Most systems stop after authority.
Attribution proves who acted. The Continuity Module™ checks whether the action still means what it was supposed to mean.
Signal to evidence — meaning checked at every step.
You see each step in the audit chain. Admission decides if the proposal may enter. Authority decides if it may proceed. Commit decides if it becomes real. Continuity checks meaning across the whole arc.
Signal → Intent → Interpretation → Memory + Context → Continuity Module™ → Proposed Action → Admissibility → Authority → Formal Rules → Runtime Decision → Commit Boundary™ → Execution → Evidence Pack™
Eleven drift types. Each one signed.
When meaning drifts, KYE™ records a drift event, hash-chains it into the audit ledger, and emits a signal on the KYE Signal Bus™. You wire alerts on the signal; auditors map the codes to ISO 42001 controls.
- constraint_loss — a material constraint lost during interpretation
- context_loss — context referenced at intent dropped before action
- memory_conflict — memory used contradicts current state
- timing_drift — delay between intent and action changed meaning
- handoff_drift — meaning changed across human / agent / tool boundary
- incentive_drift — commercial or affiliate conflict surfaced
- state_transition_drift — state changed since intent invalidates action
- interpretation_drift — agent re-read the goal materially
- assumption_injection — agent added assumptions not in the intent
- meaning_compression — interpretation lost detail
- meaning_expansion — interpretation widened scope
Decision outputs: meaning_preserved, meaning_degraded, meaning_broken, require_clarification, require_reconfirmation, require_human_review, pause_admission, quarantine_proposed_action, route_to_admissibility, route_to_authority_check.
Four normative JSON objects. Validated by ajv in CI.
Each schema is JSON Schema 2020-12 with a stable $id. You can mirror them locally or pull them from the public Hugging Face dataset.
meaning-continuity-context.json— the per-action context (original meaning, interpretation, memory, incentive, timing, handoff)handoff-trace.json— the chain of meaning transitions across boundariesmeaning-drift-event.json— one record per detected driftmeaning-continuity-decision.json— runtime ruling and obligations
Mirrored as a Hugging Face dataset: huggingface.co/datasets/KYE-Protocol/schemas.
Five engines. Contracts open. Engines paid.
Each engine binds to the same JSON schemas above. You can run a local validator on day one; the paid engines add the detection, scoring, and recon flow needed under DORA Art. 11.
KYE Continuity Lab™ on Hugging Face.
Pick a scenario. Edit the original and interpreted intent. Toggle the drift signals. The Lab returns a continuity score, the drift codes, and the recommended verdict — plus JSON you can download.
Contracts are open. Engines are paid.
The schemas, reason codes, and validators ship Apache 2.0. The managed Continuity Engine and Drift Monitor are the paid surface, billed by decision volume for buyers under ISO 42001 and EU AI Act audit.
Open source
- Continuity v1.1 schema update
- Meaning-continuity-context schema
- Handoff-trace schema
- Meaning-drift-event schema
- Meaning-continuity-decision schema
- Reason-code dictionary + drift taxonomy
- Webhook event names
- Sample payloads and handoff traces
- Conformance test vectors and validators
- Static demos and Lab fixtures
Proprietary (commercial licence)
- KYE Continuity Engine™ runtime
- KYE Drift Monitor™ (managed)
- Constraint-loss and context-loss detection
- Memory-conflict and incentive-drift analysis
- Timing-drift and handoff risk scoring
- Continuity scoring engine
- Reconfirmation orchestration
- Regulated-sector meaning packs
- KYE Handoff Trace Viewer™ Pro
- KYE Gateway™ runtime integration
Ready to see your AI agents flagged?
Start in shadow mode. We’ll deliver your first Evidence Pack™ in 4–8 weeks.