Did the meaning survive?

1 · Why now

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.

2 · Provenance is not continuity

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.

3 · Lifecycle

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™

4 · Drift codes

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.

5 · Schemas (Apache 2.0)

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 boundaries
• meaning-drift-event.json — one record per detected drift
• meaning-continuity-decision.json — runtime ruling and obligations

Mirrored as a Hugging Face dataset: huggingface.co/datasets/KYE-Protocol/schemas.

6 · Apps and engines

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.

7 · Try it

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.

8 · Open / paid boundary

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.

Ready to see your AI agents flagged?

Start in shadow mode. We’ll deliver your first Evidence Pack™ in 4–8 weeks.

Canonical KYE™ surfaces referenced on this page: Evidence Pack™ · KYE Protocol™.