SECOND_EPOCH_CLOSING_NOTE.md

Second Epoch — Closing Note

Status: Historical marker · Date: 2026-07-01 · Closed at: main = b50c486 · Regime: declared

The Second Epoch is closed. Its work was to prove the model's robustness — not to add features, but to test whether Daryl's guarantees survive when their assumptions change. Nine frontiers were opened; nine were proven.

What the epoch was

Where the first epoch built the registry, the second asked a single kind of question — do the proven invariants survive when the carrier, the scale, the authority, or the substrate changes? Every movement followed the same discipline, in this order:

1. Grounding — a read-only, factual survey of what the code actually guarantees.
2. Falsifiable experiment / design — the sharpest test that could break the claim.
3. Runtime proof — a real gate or a measured/CI-reproducible run; the run is the proof.
4. Proof Log — the movement is recorded only after it is real.
5. Law (ADR) — only when a rule becomes necessary, and (for the governance rules) only after the proof, never before.

The inversion is the point: the proof preceded the law, not the reverse.

The tally (at main = b50c486)

• Robustness frontiers: 9 / 9 proven.
• Proof Log movements: 16.
• Canonical laws (ADR-PRL): 14 (0001–0013 + the RR-binding).
• Kernel writers: 20 — unchanged, end to end. The frozen kernel was never modified: everything was built above it — derived, read-only, never stored — except the governance ADRs (0011/0012), which add readings above the projection, never into it.

The nine frontiers, by family

• Identity — the referent survives its carrier: claim_id ∉ storage (identity across projections), agent_id ≠ model_id (structured attribution), org_id ≠ carrier (organization referent). The transversal principle "identity is never defined by its carrier" graduated to the manifesto after its third referent.
• Behavior — the system holds under pressure: derived standing at scale (#1), conflict visibility (#2), the object referent (#4a), object coherence + object standing (#4b / #4b-S). Governance became homogeneous across scales — governed_standing (claim, ADR-0011) and object_standing (subject, ADR-0012), the same shape, which compose: a contested claim propagates to a contested object.
• Substrate — the proof layer itself survives distribution: distributed certification by per-registry chains (#5b, option A, ADR-0013) — certification survives with no single registry, reconciled by value-identity, portable receipts, and attestation, with no global tip and no core change.

The load-bearing habit

Nothing touched the frozen kernel. Governance was added as readings above latest-wins, never as a mutation of it (raw_standing stayed byte-identical throughout). Every index/projection/signal remained derived, droppable, never a second source of truth. That discipline is why the model is stable rather than merely feature-complete.

The transition — from robustness to construction

The conceptual core is settled. The question changes:

from "is this property true?" to "how does a user exploit it?"

Most of what follows will rest on these foundations rather than redefine them. The remaining work is construction, not robustness:

• Product surfaces — turning the proven derivations into things a user manipulates: Knowledge Objects first-class, maps/navigation, org model & permissions, API / SDK / UI, product search.
• #4b-C — the content/lineage compiler — the one remaining theoretical chantier: merging an object's content and tracking provenance/lineage across claims. It is likely the first movement to break the "derived / read-only" pattern (persisted claim↔claim relations, possibly a new act type) — so it keeps the epoch's discipline: a heavy grounding first, and the surface may reveal exactly what it must compile.

The Second Epoch proved the model. The next builds on it.