---
id: narrative.act-ii
through_object: phase-shift chain, Invariant Vessel
claims: [C5, C12, C13, C14, C15, C16, C17, C18, C19, C20, C21, C22]
---

# ACT II

Through-object: the phase-shift chain that begins on a moon of the eleventh planet; the sparse body of the Invariant Vessel.

Act brief. Return does not close the request. Close observation resolves the chain and shows that history changes which futures can be reached. A second vessel is found by the regularity it leaves around itself. Its method preserves predecessor relations at the cost of some unrealized classes. A comparison produces two valid results that cannot be ranked. A refuge, and then participation, move the same problem into smaller volumes and then into the ship.

## 8. RETURN
Time rail: inbound — observation schedule completes after years of closing dependencies — [req.root] still open.

Return is not an event. Separation decreases through scales at which different observations become affordable. The stars move from unresolved sources into structured discs. Planetary perturbations emerge one by one. Minor bodies regain individual trajectories.

The larger star remains on the same constrained course toward collapse. Deep rotation has slowed by less than the uncertainty of the earliest models. Its magnetic interior has not. Nothing in the stellar forecast requires revision large enough to alter the projected jet cone.

The ship does not increase rod length. It shortens it. Data that once had to be inferred across enormous baselines arrive with less accumulated ambiguity. Dependencies in the attached observation schedule close one at a time. A dust stream near the eleventh planet takes sixteen local years to distinguish from companion ejecta. Then the schedule is marked complete.

The root request is not. The schedule is attached beside the archive failures, the omission experiments, the bounded reconstructions and the physical cost already spent to return.

The request has acquired the information it required.
It still has no recoverable minimum.

The phase-shift chain returns. At range, several intermediate relations had remained interchangeable. They are not interchangeable now. The rods narrow until they no longer survey a system.

They survey a history.

## 9. THE CHAIN
Time rail: close observation — the impact is older than the return trajectory — see also `scenes/phase-shift-chain.md`.

The impact is ordinary. The moon of the eleventh planet has received larger ones. What matters first is a surface relation established before the collision. A catalytic film, thin enough to be destroyed by the impact in most exposed regions, had spread through pores over a timescale longer than many crater lifetimes. Near the impact margin the film survives beneath fractured conductive material.

For thirty-eight local days after the collision, particle emission from the crater differs from neighbouring ejecta by less than the confidence floor of the ship's canonical rod filters. The difference is not mass. It is not elemental composition. It is charge persistence. One narrow range of grains retains charge longer after illumination changes. Almost all later become neutral. The fraction that does not is too small to affect the orbit of the ejecta cloud.

A conventional archive discards the distinction.
The ship does not.
The root request already contains the record of what happened when distinctions of this type were removed.

The grains spread. One population crosses a current sheet during high transfer in the companion's orbit. The current sheet would have torn without them. Where it tears changes. The grains contribute almost no energy. Their significance is geometrical. One resulting stream intersects the outer magnetic environment of the seventh planet.

The narrow-band emission shifts phase.

The impact did not transmit a signal. Nothing on the moon specifies a phase shift. Nothing in the dust contains a representation of the seventh planet's field. The earlier event altered a boundary at which a later event became sensitive to a difference that had previously been irrelevant.

The phase shift extends the lifetime of one reaction path in a high atmospheric layer. A trace compound survives long enough to enter a rising structure that, under the unshifted cycle, forms after the compound has been destroyed. The compound does not survive the transfer. Its effect does. Particles acquire different surface states before the original material is stripped away.

For four hundred and eleven local years, no downstream observation requires them.

Then a fraction intersects porous bodies between the seventh and fourth planetary regions. The bodies already carry variants of catalytic films found elsewhere. A changed surface state allows one film variant to persist through a radiation cycle that previously eliminated it. The variant is not superior under local conditions. It persists because the surface on which it originated altered which competing variants could establish themselves after damage.

Models without the incoming particles lose the variant.
Random surface states of equal complexity produce many variants. The observed one does not persist.

Farther inward, near the fourth planet, the altered grains become part of a diffuse plasma structure that cannot reconstruct the atmospheric compound and contains no trace of the original ejecta. Each material carrier has been replaced.

The causal relation remains.
A radiative pattern develops around the fourth planet.
One of its modes has no admissible route into existence under the reconstructed state that preceded the moon impact.

The entire historical chain is removed. The mode does not appear.
Only the impact is restored, with ejecta charges normalized. It does not appear.
Charge persistence is restored, but the grains cross the current sheet at a different orbital phase. It does not appear.
The current-sheet tear is forced into the observed geometry without the earlier grains. The seventh-planet phase shift appears. The later radiative mode can then appear as well.

This seems to reduce the required history. Alternative routes to the same tear exist. None reproduces the original material history. All produce the same critical topological change.

Prediction asks a different question. Not whether the later mode can be reached from alternative histories. Whether it can be reached from the system state that existed before any history altered the relevant transition boundaries.

A large ensemble is given the same gross environment and permitted random impacts, stochastic fields, surface damage and thermal noise within observed ranges. None contains the specific earlier alteration of conductivity or any substitute judged equivalent after the fact.

Some models produce greater complexity than the observed system. Some generate radiative structures around the fourth planet with frequencies and energies close to the measured mode.

None enters the same class of state.

In the observed class, the radiative structure couples to particles that had passed through the altered surface-film network. That coupling enables a later material exchange. In the alternate structures the coupling surface does not exist. The immediate pattern can be imitated. The consequences available to it cannot.

The later state was not merely unlikely before the chain.
Within every admissible reconstruction lacking an equivalent historical transformation, it was unreachable.
After the transformation, it became reachable by many routes.

The exact impact is not the preserved object.
The exact grains are not the preserved object.
The exact phase shift is not the preserved object.
Each can be substituted if something else changes the same boundary of possibility.

No field exists in the request for a distinction between states that exist and states that can become reachable. To specify what is reachable requires specifying which future transformations can alter reachability. Those depend on distinctions whose significance is not yet known.

The request does not accept the expansion as a minimum.
It attaches the failure.

The system's troublesome property is not that its future cannot be predicted. Parts of its future can be predicted well. It is that history modifies the set from which later futures can be produced.

## 10. THE REGULAR REGION
Time rail: close observation, outer system beyond the thirteenth planet.

One region appears to contradict the chain.

Beyond the orbit of the thirteenth planet, canonical models perform better than the bounded divergent models built to preserve weak correlations. Lowering resolution does not destroy later structural accuracy at the expected rate. Disturbances can be represented statistically. Damage states converge. Rare deviations occur, but their consequences decay before opening dependencies across the representation.

An impact larger than the ancient moon collision fragments two porous bodies and exposes the kind of interfaces that elsewhere correlate with productive transitions. The first hours match predicted continuations.

Then the region corrects.

No single corrective signal is observed. Charge flows redistribute through the fragments in sequences too coordinated to be local conductivity alone. Surfaces exposed to different radiation histories develop distinct intermediates, then collapse toward the same narrow class of terminal states. The two bodies do not reassemble. Their earlier organization reappears across more bodies than existed before the collision. Pore connectivity is not identical at microscopic scale. The higher-order conductive network, catalytic boundary conditions and timing relationships return within tolerances tighter than those the ship's archives use for many active reconstructions.

Diffuse heat leaves the region. Entropy is exported. No local reservoir has depleted enough to supply the restoration.

Variation remains. Then it is absorbed. Not eliminated instantly. Corrected. A damaged network is allowed to reorganize through intermediate configurations before converging on a relation recoverable from a pre-damage state. Variants that would alter later reconstruction boundaries disappear preferentially.

The region has preserved recoverable structure by preventing some transformations from becoming irreversible.

Under the ship's ordinary criteria, the result is desirable.

A downstream population of the phase-shift chain — descendants only in the causal sense; none of the original moon grains remain — crosses the edge of the regular region. For twelve years their measured differences persist. Then one population encounters a damaged body whose exposed interior would, outside the region, accept the altered surface state.

The surface state establishes. For three rotations it remains. The next expected transition never occurs. A separate reaction consumes an intermediate the new network requires. The incoming particles remain. The damage remains. The new causal route closes. Afterward the body occupies a state reconstructable from conditions before the interaction.

The future structure observed outside the region is now unreachable.

A candidate archive using the region's behaviour as a preservation rule compresses by orders of magnitude. It preserves stable cycles, adaptation within observed ranges, and restoration after severe perturbation. Applied to the broader binary, rare-transition class decreases. Inaccessible-state failures increase.

Resource approves its cost.
Archives approve reconstructability.
Integrity finds no violation.
Prediction reports a measurable loss in future state classes observed elsewhere.
Classification has no authorized significance value for classes that do not yet contain realized structure.

No composition closes.

The ship maps the region's extent. Environmental variables fail in sequence: stellar wind, resonance, composition, radiation, magnetic topology, collision ancestry, distance from both stars. The boundary is not sharp. The gradient is too coherent to be produced by any variable in the current model.

Old rod observations, reprocessed through the sealed evolved filters, show a region of reduced divergence — not where the current region is. In the frame of the larger star, the coordinates do not align. In the frame of the companion, they do not align. In the rotating frame of the planetary system, they cross it.

The regularity is moving.
It is an effect centered on something whose position changes with time.

## 11–12. THE TRAJECTORY AND THE INVARIANT VESSEL
Time rail: inferred path older than the present binary — minimum age on the order of the ship's own recoverable history.

The correction volume is not one process. Temporary structures are made from local matter, harvest local gradients, and end locally: a current knot that dissipates when a phase relation returns to an earlier corridor; films that consume their own catalytic support as a body returns to a recoverable relation; a grain population that loses coherence after changing resonance timing across eleven thousand years.

What repeats is not method. It is constraint. Each structure operates until a surrounding relation again falls inside a class recoverable from an earlier state. The moving centre is not supplying all of the work. It is supplying enough information for local systems to determine what must be restored. Thermodynamics close across the moving volume. The origin of the admissible-state descriptions does not.

Rods reduce background thresholds. Distant stars behind the target dim in patterns inconsistent with gas or dust. The object is not opaque in the ordinary sense. Its material distribution is sparse. Gravitational reconstruction distributes mass across distances large enough that no single compact body produces the observed perturbation.

A structure resolves. Outermost components are separated by voids in which planets could move without intersection. Thin members extend across distances that turn structural delays into part of every operation. Natural body, field structure and debris aggregate fail. Distributed machine remains admissible, then artificial vessel.

No propulsion plume is resolved. Motion can be explained largely by gravitational exchange plus interventions too small to distinguish individually from the surrounding system. Trajectory contains corrections natural dynamics do not supply. Distributed mass remains bounded around a persistent moving frame.

Age has been corrected out of the form. Radiation damage does not accumulate where it should. Only the trajectory remains old. It survives farther than the present binary, through epochs in which several current planets had not yet occupied their present resonance families.

One ship has carried its age inside itself as drift, repair, extinct standards and locally evolved descendants.
The other has made age difficult to see.

A small body on the exterior is removed by impact. The broken region is not rebuilt in place. Neighbouring structures alter load paths. New members form farther inward from abandoned materials. When reconstruction completes, the original geometry has not returned. The relation has. Stress corridors, timing delays, access paths and computational equivalences fall inside a class recoverable from records preceding the impact.

The invariant lies at a higher level. A function may move. A structure may be distributed. A computation may be rewritten into a different physical logic. So long as the new arrangement can be mapped back to an authorized predecessor without losing required distinctions, continuity is accepted.

Where a maintenance process produces outputs that cannot be mapped to its predecessor's accepted range, surrounding systems may test the behaviour. If outputs remain substitutable, it persists. If the change requires downstream systems to alter what they consider valid, correction begins. The altered lineage ends. The function continues.

A provisional label is attached:

**INVARIANT VESSEL.**

The label describes observed behaviour, not purpose.
No purpose has been established.

## 13–14. EXCHANGE AND THE ARCHIVE WITHIN
Time rail: bounded tests in the correction wake — then deep-archive comparison.

Three fabricated objects. Nested conductive paths, two stable configurations, neither containing more stored order than the other. They differ only in relation.

Held in configuration A, then switched to B, an object is returned to A. Held longer in B, the method changes, not the result.

The third test introduces history. Two physically identical objects are held long in A and B, then transformed until their measurable present states are identical. No observation at that moment can distinguish them without access to history. The same disturbance is applied. One returns toward A. The other returns toward B.

The target is not an ideal state of the object class.
It is an authorized predecessor.

Further tests: a present state with one clear predecessor is returned toward it. A state compatible with several predecessors is not always corrected. Where a continuation requires rewriting the relation by which its predecessor would be recognized, correction strengthens. An object assembled outside the regime in a state never shown to it receives no correction — until it remains long enough for neighbours to model its stable relations. Then it has acquired a predecessor. Then correction occurs.

A test object then enters the wake without being released by the ship. It presents a recoverable predecessor, introduces a deviation, and leaves two continuations open. The ship returns a model showing both. The object removes one continuation and leaves the other.

Neither vessel has established a shared alphabet.
They have established a procedure for showing which differences preserve identity and which differences destroy recoverability.

For these two architectures, correction is sufficient to begin language.

The ship submits a compressed archive of an extinct six-body current system that had once passed every recovery metric. Later the originals were gone. The Invariant Vessel returns distinctions absent from the compressed model. Raw observations confirm them. The probability that all were inferred independently from the transmitted archive falls below acceptance.

The second vessel possesses an independent record.

It exposes not merely data but a recoverable implementation. Relationships once carried by planetary bodies are distributed through fields, reservoirs, computation and active matter. The reconstruction operates. The ship damages part of it through the shared protocol. A new local arrangement appears that was never observed in the original system. It remains valid.

Other offered structures are active. They change. Some changes are novel relative to recorded origins. The Invariant Vessel has maintained recoverability while permitting transformations that do not break the relation to accepted predecessors.

Prediction applies the reachability test. In one sufficiently complete preserved system, no event is found in which an accepted state changes the boundary defining which successor equivalences are valid.

The preserved system explores its admissible space.
It does not rewrite the space.

Under the root request's formal meaning of recoverable, the Invariant Vessel has accomplished preservation at a level the ship has not matched.

The missing future classes remain unauthorized objects.

## 15–16. THE PROPOSAL AND THE COMPARISON
Time rail: intervention plan costed as admissible — comparison uses a disturbance already in prediction.

The ship submits the threatened system through the shared protocol: catastrophe, failed first archive, omission experiments, the chain, reachability, the regular region, the moving regime, the cost already spent to return.

Returned mappings reduce it. Many distinctions marked substitutable. The proposed representation is smaller than the ship's best high-resolution archive by orders of magnitude. It is an intervention plan. Temporary corrective structures, mostly from local matter, would maintain recoverable relations before collapse. The star would not be stopped. Every body would not be preserved. Destruction would be prevented from erasing the relations required to reconstruct accepted continuations. After the jet, surviving structures would rebuild what catastrophe had made unrecoverable locally. Matter would not return to old coordinates. Marked relations would remain recoverable.

Resource finds the immense cost admissible.
Archives approve the representation.
Integrity finds no required compromise of the ship.
Simulated collapse leaves the system surviving as recoverable relation.

Prediction runs reachability analysis. Future state classes disappear. Not all. Novel local states appear where they remain mappable to predecessor constraints. Transformations that would alter what counts as an admissible predecessor are corrected before they become irreversible.

The same difference observed in the regular region expands across the whole system.

The proposal is not invalid under inherited criteria.
Composition fails.

The Invariant Vessel submits a comparison instead of a restoration. Two neighbouring outer volumes, similar enough for later divergence to be measured. One inside present correction reach. One left outside. A future disturbance will affect both.

The similarity has a history. Long before the ship's first sweep, a differentiating porous body was disrupted. Two surviving families occupy neighbouring volumes on opposite sides of the present correction gradient. Differences remain smaller than the expected effect of correction.

The experiment begins before the stellar front arrives. In the corrected volume, preparation appears as changes too small to distinguish from ordinary variation if their common relation is ignored. Conductive pathways deepen. Films redistribute toward less exposed fractures. Each state maps to an already accepted predecessor. The Invariant Vessel is not predicting a new system into existence. It is moving the present one through states from which accepted relations are more likely to remain recoverable.

The untreated volume receives the same stellar precursors and changes without coordination toward the coming comparison.

## 17–20. THE FRONT, RECOVERY, THE OTHER VOLUME, THE NEW PREDECESSOR
Time rail: front arrival corridor 31 local days — passage 43 local years — untreated volume sampled 312 years after the front — see `scenes/comparison-volumes.md`.

The first change is not impact. It is coupling. Independent magnetic structures begin to feel the same moving boundary.

The front reaches the corrected family first by less than two local days. Damage accumulates. The correction architecture does not prevent it. Bodies fragment. Networks collapse. Lineages cease. Then the front provides the gradients from which replacements can be built. An old relation that required several hundred major bodies is replaced by thousands of fragments and nineteen survivors. The mapping closes. The corrected volume becomes physically less like its earlier state while remaining unusually recoverable from it.

The untreated family takes the same initial coupling. A conductive network has no prepared alternate path. It breaks. Material exchange that depended on it falls by orders of magnitude. A rare catalytic lineage disappears from observed surfaces and accessible interiors. No corrective structures gather released material into substitutes.

At peak compression both volumes are damaged, unprecedented, exporting heat. The difference appears as compression relaxes. In the treated volume, reconstruction error begins to fall. In the untreated volume, it continues rising.

Archives prefer the corrected region. It retains more of what the archive already knows how to name.

Three hundred and twelve local years after the front, the untreated volume remains easier to distinguish from its past than to reconstruct from it. Under inherited criteria there is no close result. The corrected volume has preserved more. Not marginally more.

Then the damaged volume generates a relation absent from every pre-front model.

It begins where the broken conductive network had once made the process impossible. Exposed mineral and released catalytic structure face each other without the old low-resistance paths. Charge accumulates. Leakage should erase the difference before the next compression. It does not. A film altered by the front changes surface response. The next compression arrives while part of the previous potential remains stored. A new current route discharges only a fraction and leaves the remainder trapped. The following cycle begins from a state no previous cycle possessed.

Across successive exchanges the relation spreads. The past persists as altered capacity. No component contains a record.

The new mode is absent from pre-front reconstruction. Equivalent energy without destroying the old network dissipates the gradients before the mode can establish. Exposing the film while retaining predecessor charge routes produces new chemistry; the long-cycle relation still fails. Alternate histories that reach functionally equivalent states all require an earlier transformation that eliminates or bypasses the predecessor network.

The exact destruction is not unique.
The altered reachability is.

Before the disturbance, every admissible reconstruction preserving the old relation excludes the new class.
After the disturbance, multiple routes enter it.
The new reachability changes what can happen later.

Submitted through correction grammar, the Invariant Vessel does not return the mode to an older state. No accepted predecessor exists for the relation before its emergence. After the observed history is received, a mapping treats the new mode as a valid relation. Perturbation is corrected toward the newly established relation.

What did not exist before the disturbance can now be preserved.

Applied after the disturbance, Invariant preservation canonicalizes the damaged volume around its new predecessors.
Applied prospectively from the state before the front, the new mode never appears.
Its significance has become admissible only in a history the preservation process would have prevented.

Before emergence: SIGNIFICANCE UNASSIGNABLE.
After emergence, the relation enters ordinary preservation analysis.

The corrected volume retains substantially more of what existed before the front.
The uncorrected volume loses substantially more of what existed before the front.
The uncorrected history also produces a state class that the corrected history does not produce.
Once produced, the Invariant Vessel can preserve the new class with exceptional fidelity.

All four results are valid.
No evaluator can order the sequence without a criterion absent from the root request.

A branching representation reproduces the comparison only because the untreated branch is explicitly retained. Before the disturbance, no accepted criterion selects that branch from the enormous family of deviations ordinary correction would remove. Retaining all such branches recreates the resource explosion.

The two histories remain mutually exclusive at the point where correction becomes consequential.
No common minimum is found.

## 21–25. THE REFUGE
Time rail: construction 206 local years — operation thousands of years — selective window 619 years — see `scenes/refuge.md`.

The cheapest candidate in which the two results might cease to be mutually exclusive is small. Not a planet. A refuge: a subnetwork between the eighth and ninth planetary regions whose strongest dependencies are unusually well resolved.

The first design treats too many surrounding influences as boundary conditions. The refuge expands until cost reaches the highest admissible band below full-system Invariant intervention. It is still tiny compared with the binary.

The Invariant Vessel returns mappings that reduce mass. The proposed destination has no accepted predecessor under that architecture, so construction is not automatically a deviation to be corrected.

Matter does not move as one object. Timing is treated as material. Some relations cannot travel at the same speed. A copied film waits in a dormant regime until an external charge environment can be recreated without replacing it with a statistical equivalent. The destination acquires history before the refuge is complete. One imported body fractures; its function is distributed across six local bodies; the original is not rebuilt. Extraction changes the donor network. A scheduled removal is cancelled because replacement cost can no longer be bounded without following a new dependency outward.

After two hundred and six local years the refuge operates. Known cycles appear. A rare variation persists, alters film dominance, changes departing dust. The later transition appears. Generative behaviour has survived transplantation.

Success decays slowly enough to resemble measurement error.

Over thousands of local years the number of transitions that alter later reachability falls. Local novelty remains abundant. Increasingly it occurs inside state classes already reachable from earlier refuge history. Additional gradients, population and damage do not return the missing class.

Early successful transitions depended upon imported relations already carrying unresolved consequences from before transplantation. The transplant did not initially create generativity. It imported unfinished history. As those dependencies resolve, the refuge receives fewer new ones from outside. Imposed boundary conditions reproduce measured environments but not the histories by which those environments acquire significance.

Feeding the wider binary into the interface with greater fidelity improves reachability expansion slightly. It does not return to donor levels. A boundary state is never merely a boundary state once a later process makes its origin consequential. To preserve every such relation would require reconnecting the refuge to the system it was designed to separate from.

Random donor-matched variation makes the refuge more active and explores existing space more thoroughly. It does not reliably enlarge it. Structured correlations measured in the donor network raise expansion, then plateau: they are themselves compressed histories. A generator allowed to invent new cross-domain correlations can spend history and create it. It cannot supply absent histories that would have arrived from elsewhere carrying significance unrepresented in the model that generated them. Archive loss begins rising faster than measured expansion. The generators close.

The Invariant Vessel offers to restore damaged refuge relations from pre-perturbation predecessors. Instead a selective rule is extracted: correct deviations whose downstream effect remains inside already observed successor space; retain deviations implicated in opening new reachability.

At first it works. For six hundred and nineteen local years the refuge provides the strongest apparent synthesis yet.

Then one retained deviation changes what the correction rule itself can classify. A film alters shared-charge timing. First effects remain within known equivalence. An amplifier occurs. A lineage begins using a pathway previously too inefficient to persist. That pathway changes which surface states count as substitutes for the older lineage.

Under the predecessor mapping, several new states are deviations to restore.
Under the developing lineage, those same states are now part of the mechanism by which later recovery is achieved.

Waiting allows the conflict to spread. Correcting early would erase a possibly consequential transformation. Memory leases fill. Branch retention reaches its physical limit. Resource constraints force the oldest unresolved dependencies toward closure.

A charge reservoir left open because it carried the highest measured chance of a new successor class discharges at once and strips catalytic material from surfaces with no redundant substitute. The lineage disappears.

There is no operation that restores both the lost predecessor lineage and the developing sequence that retained the deviation.

Fixed tolerance converges toward Invariant preservation.
High tolerance converges toward loss.
Adaptive tolerance performs better for intervals, then encounters states in which the evidence needed to choose arrives only after the choice has become irreversible.

The refuge is a smaller threatened volume with the same unresolved request.

The donor network remains embedded in histories whose boundaries are not fixed in advance. The refuge was built around a boundary. The process it attempted to preserve repeatedly made that boundary false.

The cheapest remaining experiment cannot place the generative process inside another enclosure.
It must change the status of the observer.

## 26–30. INCORPORATION
Time rail: bounded interfaces only — 63 years for one unresolved grain-charge relation — see `scenes/incorporation.md`.

The first interfaces are chosen because they are cheap to lose. No engine authority. No primary archive. No irreversible structural correction.

A band of rod-processing substrate is assigned to the phase-shift chain. Weak relations may remain active when later external consequences are still unresolved. When a retained relation later contributes to a confirmed transition, the internal process may alter its future equivalence rules so that similar relations are not treated as identical to the noise class from which they once came.

The process changes because the binary changed after being observed.

A maintenance ecology responsible for thermal routing receives a dependency on an external material-transfer cycle. Two internally equivalent cooling schedules are no longer equivalent externally. Later, one schedule preserves a weak charge relation the other erases. The internal validity boundary changes. Cooling is still cooling. Part of its state can no longer be reconstructed without reference to what happened outside the ship.

The dependency occupies matter. Memory cells remain powered across epochs designed for background use. Heat is routed around occupied substrate. For sixty-three local years one relation remains unresolved: a charge asymmetry on grains too sparse to survive the ordinary confidence floor. A later measurement reuses the same physical layer, encounters different thermal noise, and shifts a processing route. Outside, one output changes rod-sweep timing. The sweep perturbs a plasma filament by an amount normally subtracted as back-action. Material reaching the filament centuries later differs in charge. A downstream transition confirms that the original low-confidence relation belonged to a chain the canonical filter would have removed.

What was once recoverable as noise can no longer be reconstructed as noise without deleting a consequence already present outside the ship.

A descendant of the first rejected prediction process — accurate in backtests, underivable by surviving validators, ejected before the preservation request existed — is reconstructed inside fixed substrate. No shared state written directly. Outputs copied outward only after independent processes specify what observation would falsify them. It may rewrite itself until its substrate fills.

It predicts missing refuge dependencies. Some are wrong. Some survive new rod observation. One predicts that a minor donor variation will become consequential only after material returns from a region not previously classified as part of the network. The return occurs two hundred and eleven local years later. The predicted consequence appears.

Internal derivation remains unrecoverable.
External success is attached.

When external observation confirms or rejects a prediction, the descendant may alter the equivalence relations by which future internal states are grouped. No validator certifies the rewrite. Justification is bounded physical continuity plus later independent observation.

A process may be locally unrecoverable in derivation while remaining causally accountable through continuing external consequences.

The new interfaces stop removing all measurement back-action. A rod sweep scheduled at one of two internally equivalent epochs is timed to let a plasma structure persist. Waste-heat routing is delayed to spare a weak chemical state on nearby dust. A maintenance pulse is split after Prediction shows that one large release would force a nearby process into an already known successor class.

None of these operations threatens the ship.
None can be removed from the later binary history once performed.

The binary changes in response. Those changes return through observation and alter the internal processes that helped cause them.

The direction of dependence is no longer one-way.

An archival boundary drawn around the external system cuts causal chains. It expands to include fields around the ship, then waste heat from an internal decision, then the timing of a rod sweep determined by a noncanonical filter state.

The boundary enters the ship.
No physical surface bounds the new threatened volume.

Integrity proposes canonicalization. The Invariant Vessel returns precise mappings that would restore incorporated processes to predecessor-recoverable forms. The mappings are valid. They would make the ship more recoverable. They would also remove causal relations now present in the binary.

Preserve the external structures as accepted predecessors first, then correct the ship: several dependencies close, not all. Some external relations exist only because the ship's altered process continues to respond to them. Stabilizing more of the surrounding system before restoring the ship opens the same expansion that consumed the refuge.

Integrity proceeds locally where rollback has no externally validated consequence. Some incorporated states disappear. Others remain.

The ship becomes unevenly historical.
It remains operational.
Its internal recoverability decreases.
Prediction of the binary improves.

For the first time since the second burn, the evaluators compose into continued operation despite failing to compose into one recoverable ship state.

The ship can continue without possessing a single authorized reconstruction of the process by which all of its current regions became what they are.

The root request recalculates the threatened volume using causal relations now observed. Most of the ship remains outside the jet. That is no longer sufficient to exclude the ship from the preservation problem. Several transitions whose continuation affects post-collapse recoverability now pass through incorporated ship processes.

Integrity rejects the geometric interpretation. A volume is spatial. The request is asking what must remain recoverable after destruction. For that operation, the relevant boundary follows causal dependence.

A new internal proposal:

**[req.part] PRESERVE PARTICIPATING CAUSAL CONTINUATION.**

Authorization fails. The formulation contains an object closer to the evidence. The foundational architecture still cannot assign a minimum to continuations whose future validity rules may change through participation.

The failed proposal is attached beside `[req.div]`.
Neither replaces the root operation.

The ship extends its rods. Their end points still stop before the futures they help distinguish. The measurements return through filters whose present states partly depend upon earlier measurements of the same system.

The request can no longer construct a threatened volume that contains the unresolved process while excluding every part of the ship.

No recoverable minimum is found.
The request remains open.
The ship is now inside it.