Vigency Debt — an executive brief

Your agents are not hallucinating. They are executing judgment that is no longer in force.

Canonics · NXLV Research · companion to Position paper (preprint) · July 2026

The failure that looks like success

When a company changes how it operates — dissolves a team, signs an enterprise contract, pivots a product from a single-page app to a server-rendered one — none of the old knowledge becomes false. The hiring plans were real. The old architecture constraints were real. They remain true as history.

But every long-lived agent reading that memory keeps applying them as guidance. It quotes a price built on a team that no longer exists. It flags a bundle-size regression that no longer matters. It defends a rate limit that was removed months ago. The output is fluent, well-sourced, and confidently wrong — which is exactly why nobody catches it. The agent is not making things up. It is citing obsolete judgment.

The missing object is not freshness but force: whether an interpretation is still in force under the way the business actually runs today. We call that property canonical vigency, and its slow, silent loss is the real cost center.

Vigency debt: the balance nobody is carrying

Every interpretation an agent has derived under an old operating model — and not yet re-derived under the new one — is a liability. The accumulated stock of those is vigency debt: the interpretive analogue of technical debt.

It behaves like debt, too. It jumps every time the operating model changes, and it is paid down only when interpretations are actually re-derived. Left alone, it compounds silently until it surfaces in a wrong decision.

(See Figure 2 — Vigency debt accruing.)

The dangerous part is that it is invisible on every current dashboard. Memory systems today can tell you whether a fact is still valid. None can tell you whether the paradigm a conclusion was built on is still active. So the debt accrues off the books.

What Canonics adds

Canonics makes the operating regime a first-class, versioned, human-declared object:

• A canon event records a regime change once, in one place — what assumptions retire, what begin, over which domains — and every downstream interpretation reacts automatically.
• Staleness is measured in canon distance (κ): how many in-scope regime changes an interpretation has crossed, not how many days have passed.
• Nothing is deleted. Old interpretations are kept as history; they simply stop steering the next decision.
• Detection is not automated away. A monthly Canon Review structures — and measures — the encounter between a machine that scans for regime change and a human who judges.

The one-sprint wedge (what it costs to start)

You do not adopt the whole protocol to get value, and you should not try. The minimal slice that pays for itself in a single sprint is two moves:

1. Keep a current-regime block in your agents' always-on instructions.
2. Attach a canon_distance tag to each retrieved derived artifact — use κ = 0 artifacts normally; on κ > 0, show a stale-regime banner or re-derive.

No graph database, no ritual, no benchmark required to begin. (See Figure 3 — the one-sprint wedge.)

Why this matters for a memory platform

Agent memory is becoming a platform primitive. As it does, whoever ships regime-aware retrieval implements a mechanism — and the governance layer above it (the vocabulary of vigency, the four validity states, the Canon Review) is what remains defensibly separate from any single product. The buyer's language for this is not "second brain"; it is AI governance, auditability, compliance, and operational risk: who may declare authoritative meaning for a shared agent memory, how a sensitive regime change propagates without leaking, and which audit trail lets a team reconstruct why an agent gave a given piece of advice under a prior regime.

Long-lived agent memory that guides decisions needs this layer before it can be trusted. Canonics names it.

Full paper, reference implementation, and the VRS (Vigency Resolution Score) benchmark spec: research.nxlv.ai/canonics. Contact: Lucio Amorim, NXLV Research — lucio.amorim@gmail.com.