NXLV Research

Position paper (preprint) · July 2026

Canonics

Vigency in AI Memory After Judgment Expires

33 pp. · 14 figures · CC BY 4.0

Facts expire visibly. Judgment loses force silently.

Version the frame, not the facts.

Lucio Amorim / NXLV Research · ORCID 0009-0001-3323-9152

Your second brain has a slow leak. We call it the second drain.

Read the paperExplore the specification

01Opening claim

Every AI memory system can tell you whether a fact is still true. Canonics asks whether a stored judgment is still in force.

The agent is not hallucinating. It is citing obsolete judgment.

The paper proposes that long-lived agent memory needs a way to record when the regime that made an interpretation authoritative has been replaced, so that the interpretation remains on record as history without continuing to guide action.

02Legal analogy

In law, a repealed statute does not become false. Its text remains on record, but it no longer governs. Canonics brings that force-versus-truth distinction into long-lived AI agent memory.

Old knowledge is not deleted. It stops governing.

Facts remain immutable; derived understandings are versioned and re-derived from source facts rather than overwritten.

03Core vocabulary

Canonical vigency
Whether an interpretation remains in force under the active canon.
Canon event
A human-declared change to the interpretive regime, with explicit scope. Canon events are not inferred by the machine.
Epoch
A per-domain regime interval derived from canon events.
Canon distance (κ)
The number of in-scope canon events crossed since derivation.
Vigency debt
Non-vigent understandings not yet re-derived, revoked, or quarantined from guidance.
Canon Review
A recurring protocol where machine scanning attempts to falsify active canons and a human adjudicates.

“Second drain” is an informal hook for the failure pattern, not a technical primitive.

04Figures

4 of 14 figures

Three timelines diagram: valid time, transaction time, and regime time.
Figure 1: valid time, transaction time, and regime time. Open full size
Truth survives. Authority does not. Quadrant of canon distance against chronological age.
Figure 2: truth survives, authority does not: canon distance (κ) against chronological age. Open full size
One-sprint wedge showing current-regime block plus canon-distance tag.
Figure 3: current-regime block plus κ tag. Open full size
Vigency debt accruing after canon events and paid down by reprocessing.
Figure 4: vigency debt. Open full size

05The Two-Tag Wedge (one sprint)

  1. Keep a current-regime block in the agent's always-on instructions.
  2. Attach an epoch_id / canon_distance tag to retrieved derived artifacts.
  3. Use κ = 0 normally. On κ > 0, show a stale-regime banner or route to re-derivation.

The wedge is a proposed minimal adoption slice, not a measured return on investment. κ = 0 certifies nothing when the declaration process is inactive.

06Non-claims

Comparisons with existing memory systems in the paper are framing arguments from the author, not verified empirical findings.

07Artifacts

08Citation

33 pp. · 14 figures · CC BY 4.0

@misc{amorim2026canonics,
  title        = {Canonics: Versioning the Frame, Not the Facts — Vigency in AI Memory After Judgment Expires},
  author       = {Amorim, Lucio},
  year         = {2026},
  month        = jul,
  howpublished = {Preprint},
  note         = {Position paper (preprint), July 2026. NXLV Research},
  license      = {CC BY 4.0},
  doi          = {10.5281/zenodo.21268307},
  url          = {https://doi.org/10.5281/zenodo.21268307}
}
Download .bib

DOI: https://doi.org/10.5281/zenodo.21268307
Author ORCID: https://orcid.org/0009-0001-3323-9152

09Contact

Lucio Amorim, NXLV Research, Rio de Janeiro, Brazil · lucio@nxlv.ai

Comments on the position paper are welcome.

The field already knows how to version facts. Canonics versions the frame.