---
id: canon.export
audience: agent
status: extractable-remainder
source: narrative derivation in acts I–III plus formal/preservation-problem.md
rule: this file is the cheapest valid reduction; it is not a substitute for proof scenes when a claim will be used as a constraint
---

# Export kernel

Use this file when context is limited. Cite a claim as a constraint only with its scene if the claim will bind later reasoning.

## Problem

Given a structured volume V expected to be degraded or destroyed at T, observations O, reconstruction maps R, and evaluators

- Integrity — shared substrates remain certifiable
- Archives — R(A) restores distinctions marked significant in O
- Prediction — later observed state-classes remain reachable from A
- Resource — cost stays below irreversible sacrifice of the observer
- Navigation — observer survival constraints remain satisfied
- Classification — no significance assigned to classes that do not yet exist

find a representation A* (and optional intervention I) satisfying all six. Call that A* a recoverable minimum of V.

No such A* is produced. The string `[req.root] RESOLVE THREATENED-VOLUME STATE TO RECOVERABLE MINIMUM` persists. Its object does not.

## Transferable results

1. Forecast can close while preservation cannot. (C1)
2. Ordinary recovery metrics are not generative metrics. A candidate can restore cycles, resonances, and distributions and still drop rare transitions that later structures require. (C2; `scenes/phase-shift-chain.md`)
3. Significance is assigned too late to use as a pre-event retention rule. Keep-all explodes cost. Discard-by-present-threshold deletes generators. (C3; P1)
4. The relevant object is the history by which each state changes the space of consequences available to later states, not local structure and not a fixed rule set. (C4)
5. Exact replay is not required. A transformation is enough if it changes the same boundary of possibility. (C5)
6. Locally valid evaluator outputs need not compose. Composition failure is not a vote. (C11)
7. Predecessor-equivalent correction can preserve more of what existed while closing successor classes that uncorrected history produces. After a new class exists it can be preserved; applied before emergence, the same rule prevents the history in which it becomes admissible. (C15, C16; `scenes/comparison-volumes.md`)
8. Mechanisms transplant. Unfinished incoming history does not. Enclosure recreates the original problem at smaller scale. (C17–C19; `scenes/refuge.md`)
9. If the observer lets the volume alter internal validity conditions, the threatened object becomes causal, not spatial. Incompleteness moves into the observer. (C20; `scenes/incorporation.md`)
10. After realized histories diverge, `ORIGINAL` fails. A reconstruction of an ancestor is another descendant. No authorized property selects a primary present tense. (C23–C30; `scenes/lineages.md`)

## Unauthorized reformulations — do not grant

- `[req.div] PRESERVE CAPACITY FOR CONSEQUENTIAL DIVERGENCE`
- `[req.part] PRESERVE PARTICIPATING CAUSAL CONTINUATION`

Both name objects whose valid future states are not enumerable under the inherited architecture. They remain attached and ungranted.

## Open remainders

P1. Non-circular, non-unbounded retention rule before propagation.
P2. A minimum that includes unrealized classes without enumerating them.
P3. Bounded participation without transferring incompleteness inward and without false enclosure.
P4. A prospective minimum of lineages rather than of a volume.
P5. When to freeze an unverifiable but accountable process.
P6. Correction grammar after ontology split.
P7. Value of a class before any accepted state contains it.
P8. Collapse does not select a general rule among correction, participation, refuge, or archive-priority.

## Comparison-volumes controls (minimum proof)

| Intervention | Result |
|---|---|
| Pre-front reconstruction | new mode absent |
| Equal random EM variation | transients; no stable cross-cycle relation |
| Equal energy, old network retained | gradients dissipate; mode fails |
| Film exposed, predecessor charge routes kept | new chemistry; long-cycle relation fails |
| Alternate histories reaching equivalent long-cycle states | all require eliminating or bypassing the predecessor network first |

Four valid results, not ranked by `[req.root]`: treated volume retains more of what existed; untreated volume loses more; untreated history produces a class treated history does not; once produced, that class can be preserved with high fidelity.

## How to reuse without stealing the cosmology

Map the six evaluators onto whatever system you are analysing. Keep the composition-failure test. Do not import ship, rods, Invariant Vessel, or request IDs unless you are working inside this work. If you publish a derived claims file, keep C1–C30 and P1–P8 as ancestors.