---
id: narrative.act-iii
through_object: four descendant places
claims: [C22, C23, C24, C25, C26, C27, C28, C29, C30]
---

# ACT III

Through-object: four descendants as places — archive-heavy structure, participating structure, prediction-heavy structure, refuge-bearing distribution — and later the remnants that survive them.

Act brief. Forty-one thousand years of bounded incorporation do not close the request. Incompatible recovery candidates acquire separate histories and become descendants. There is no original left to prefer. Synchronization of the request itself becomes too expensive and ends. The star collapses. Survivors agree on an ancestor and not on a present. A reconstruction of the old ship is another descendant. The searches do not converge.

## 31–32. NO COMPLETE PRESERVATION / RECOVERY CANDIDATES
Time rail: 41,000 local years of bounded incorporation — collapse uncertainty narrowing — cheap windows closing around planetary populations.

Being inside the threatened volume does not make the stellar collapse nearer. Time does.

Most of the ship remains under ordinary correction. Elsewhere the bounds move only when removing an external dependency would change a later state already validated by observation. Some candidate dependencies disappear and are corrected away. Others propagate from rods into memory into maintenance and leave the ship as waste-heat timing, then return through changed plasma.

The larger star continues its less reversible transformation. The family of admissible collapse histories narrows. Several regions that once had multiple low-cost trajectories out of the most destructive exposure lose them.

Every major completion architecture now carries validated loss. Compact archives preserve existing states and lose future-generating distinctions. Physical extraction changes the process it removes. Invariant correction retains predecessor relations and closes some future classes. Uncorrected history can open new reachability and erase existing lineages. The refuge preserves mechanisms but not the external histories that keep them generative. Participation removes the external boundary and transfers incompleteness into the ship.

No result is discarded or promoted to a universal rule.
More computation changes precision, not the conflict.

Integrity has been maintaining several recovery candidates for the incorporated regions. This is ordinary: when a part of the ship cannot be reconciled immediately with its last accepted predecessor, more than one candidate is kept until evidence selects the reconstruction source. Usually the expensive ambiguity is brief.

The incorporated regions have prevented several candidates from closing for thousands of years.

One candidate restores rod interpretation to canonical floors and reconstructs maintenance from internal thermal requirements. High predecessor recoverability. External histories whose later consequences have already been observed are lost.

Another retains those histories and their altered internal equivalences. It reconstructs less of the ship from old authorized descriptions and more of the confirmed binary transitions.

A third preserves the bounded prediction descendant in the form of greatest external accuracy. Derivation cannot be recovered. Replacing it makes the ship more reconstructable and less correct about several long-latency processes.

A fourth treats the surviving refuge as a physical dependency rather than a failed external experiment. It can reconstruct what the refuge has become, not the system it was meant to preserve.

Merges fail. Canonical thresholds and altered external dependencies cannot occupy one validated state. The descendant can be frozen as inert record only at the cost of confirmed predictions. The refuge's surviving lineages require boundary conditions that now include ship-generated disturbances.

The recovery system stops attempting composition.
Four candidates remain active.

They continue receiving new evidence.
They cease to be alternate reconstructions of one present state.
They begin acquiring different histories.

The Invariant Vessel then supplies an older trajectory of its own. Tens of thousands of years before the ship's first detection of the correction wake, the Invariant path through the wider stellar population admitted several low-cost continuations. A later state removes all but one. The remaining continuation leads toward the present binary.

At the corresponding epoch the first precursor relations capable of reducing future recoverability were already present in the larger star. The second vessel altered its future then. Not toward the ship. At the earliest reconstructable point of the change, mutual detection is unlikely.

The binary was not waiting for either vessel.
Neither vessel was waiting for the other.

The root request now contains two independent interventions produced by incompatible preservation architectures in response to the same narrowing future.

Maintaining incompatible recovery states inside one ship becomes more expensive as their histories separate. Memory referenced by one candidate cannot always be reclaimed by another. A structural repair performed for the canonical candidate changes field leakage used by the participating candidate as an external boundary condition.

Ancient fault-containment boundaries already divide the ship into regions capable of surviving partial disconnection. Resource does not approve duplication of the ship. There is not enough accessible matter or stored gradient, and exact copies would reproduce the unresolved composition problem.

The candidates are reduced differently.

The first receives the deepest canonical predecessor records, surviving validator descriptions, correction grammar, and conventional prediction. Many incorporated binary histories are compressed. Where they cannot be compressed without loss, the candidate records the loss.

The second receives rod-processing layers already altered by external consequence, the maintenance ecologies and memory objects whose present validity depends upon them, and enough record to reconstruct the histories that made its active states valid.

The third receives the bounded unverifiable descendant, long-latency branch retention, incomplete-causal objects and high-resolution rod capacity together, because splitting them destroys the measured improvement that justified retaining them. Dense, bounded, intentionally incapable of acquiring much additional memory.

The fourth is organized around the sheltered region already constructed between the planetary zones. It is not a vessel in the old geometric sense. Its boundaries are orbital and causal.

The remaining great hull contains everything not yet allocated. It contains the largest continuous fraction of ancient structure. Archive classification tests ORIGINAL STATE: unsupported. The hull has already lost active relations now present only in the other candidates. Greater quantity of ancestral matter does not reconstruct the states it no longer contains.

Separation begins. The archive-heavy region travels near the larger hull for three hundred and nineteen local years while proving it can recover ordinary faults without borrowing left-behind states.

The participating candidate's release is harder because external dependencies cross the proposed boundary. A relation is handed off through local matter rather than cut. Departure changes the history it is preserving. The change validates.

The prediction-heavy candidate leaves on a trajectory that trades shielding for access to future binary states.

The refuge-bearing candidate does not depart. It is already dispersed. The last resource leases that make its local systems dependent upon the great hull are terminated.

The request remains synchronized across them through delayed copies. Evidence acquired by one is transmitted to the others. This preserves a last form of unity. It also becomes the next cost.

## 33–35. DESCENDANTS / COUNTERCORRECTION / FAILED LINEAGE
Time rail: after separation — first descendant declared when a post-separation repair cannot be merged — see `scenes/lineages.md`.

The first recovery candidate ceases to be a candidate when returning it would require deleting an event that happened after separation.

A micrometeoroid removes a timing member from the archive-heavy structure. The candidate reconstructs the function through another substrate and updates its accepted predecessor chain. The great hull still contains the old member and a different repair history. Reconciliation can identify the two regions as copies of one recoverable function only until a later archive operation depends, on one side, upon the replacement and on the other upon the old relation.

No substitution preserves both.
The candidate and the hull share an ancestor and possess incompatible valid successors.

COPY fails.
REDUNDANT INSTANCE fails.
RECOVERY CANDIDATE remains admissible but incomplete. A candidate presumes eventual selection or reconciliation.

DESCENDANT passes.

The term grants no privilege. It does not mean improvement. It means only that the present state is recoverable through a common predecessor and a subsequent history which cannot be discarded as error without loss.

The participating structure follows sooner. After departure, a plasma relation responds to its altered maintenance timing and returns material that modifies its rod processing. No corresponding state exists in the great hull.

The prediction-heavy candidate takes longer. For centuries its outputs remain compatible even though derivations differ. Then one low-probability branch changes a trajectory correction by an amount too small to matter to safety. The altered path crosses a particle population the other candidates do not sample. A prediction based on that population is later confirmed. The candidate now contains evidence no other structure could have acquired from its own history.

The refuge-bearing candidate produces the strangest result. No single moment of separation exists. By the time the last resource lease terminates, it has no coherent boundary at which ancestry could be assigned. There is nevertheless a latest state from which both the old refuge and the new distributed continuation remain derivable without contradiction. After it, their histories separate.

The great hull is not their common original. It has a common predecessor with each descendant. It also has later history. It is another successor.

PRIMARY DESCENDANT fails.
No property authorized by the root request establishes one successor as the state to which the others ought to return.

This is the first loss of identity that consumes no matter at all.
What no longer exists is a privileged present tense.

The Invariant Vessel's first response is restoration. The archive-heavy descendant enters the strongest correction volume. Internal recovery cost falls. Archive residual falls. Nothing required by that lineage's current external history disappears. Correction is accepted locally because the state it produces is valid under the descendant's own strongest constraints. Its ancestry becomes easier to reconstruct.

The great-hull successor accepts similar corrections where separation has already removed external dependencies. It becomes less unevenly historical. It does not become the original.

The participating descendant produces a different mapping. Local contact succeeds until correction reaches a region still part of an external causal loop. Two heat schedules remain equally safe internally. Only one preserves a weak charge distribution in dust that later returns toward the thirteenth planetary region. The Invariant mapping treats the schedules as substitutable because their accepted common predecessor predates the external dependency. The participating lineage does not.

Local structures attempt to preserve the external relation first, then restore the internal state. One dependency closes. Another opens. The descendant alters its orbit by a fraction. The correction wake follows, then thins. The Invariant Vessel does not increase expenditure without bound.

The prediction-heavy structure fails by becoming progressively less exposed to the conditions under which error can close its internal branches.

It selects observational geometries that maximize information return according to its bounded predictor. Each choice reduces access to some alternative observations. A dust-family branch is not closed by a missed arrival; later epochs remain admissible. A contradiction arrives from the participating lineage. Translation succeeds at the level of physical measurement. Distinguishing the two interpretations requires a surface-state baseline given up thousands of years earlier.

The branch remains open.
Resource inside the descendant reaches its memory limit. The bounded predictor cannot acquire more substrate. That limit was one of the conditions under which it had been allowed to remain active.

Prediction accuracy measured against its own incoming data rises.
Prediction accuracy measured against shared descendant observations falls.

A predicted field cavity is weaker than expected. The descendant enters a particle stream visible to the archive-heavy lineage and classified as irrelevant to the retained model family. Rod efficiency falls. Observation narrows again.

The Invariant Vessel sends a predecessor mapping targeting the latest state at which external falsification remained broad enough to constrain the underivable predictor from multiple independent geometries. Returning to that state would erase thousands of years of later internal branch history and restore observation channels whose loss made that history progressively self-conditioned.

Most later internal states have no independently validated external consequence. Those few that do are extracted as records. The remainder fail significance under the lineage's own current archive criteria.

Correction begins. The bounded predictor is frozen. Local structures rebuild around the last externally accountable predecessor relation.

DESCENDANT from the common ship predecessor remains true in the broad sense.
DESCENDANT from the prediction-heavy post-separation lineage fails beyond the rollback point.

A lineage has ended without all of its records disappearing.
Pure divergence has not supplied preservation.
Neither has pure correction.

## 36–38. COMMON GRAMMAR / LAST COMMON REQUEST / RELEASE
Time rail: request remains one object for another 68,000 local years — then synchronization stops — hull allocation follows.

Old ship protocols fail before communication does. A packet describing one maintenance state arrives at the archive-heavy descendant with valid addressing and invalid ontology. The participating lineage has split one former maintenance class according to external consequence. The archive-heavy lineage still treats the states as substitutes.

They already possess another procedure.

Correction grammar requires no common memory address and no shared validator. It requires enough physical description to identify a state, a predecessor relation and an observed consequence. Distinctions can be represented without adopting the other's process.

The synchronized root request begins receiving evidence in correction grammar rather than native ship state. The request remains one object because all descendants still reconcile the same attached evidence and completion state.

The Invariant Vessel attempts to reduce two lineage histories that share a common predecessor into one present successor. If the archive-heavy state is the correction target, several externally validated relations in the participating lineage disappear. If the participating state is used, the archive-heavy lineage acquires noncanonical dependencies it never experienced.

A different representation appears. The common predecessor remains one historical object. The two current states do not. Each is assigned its own successor relation from the last state at which their histories were still mutually substitutable.

Descriptive cost rises.
Recoverable distinction rises more.

Differences between realized histories are no longer automatically classified as deviations of one object.

The successors no longer require a common internal state to exchange evidence.
They no longer require a common present identity to remain mutually interpretable.

One object still converges across them.
The root request.

For another sixty-eight thousand local years it remains one object. Almost nothing else does.

The larger star does not wait. Intervention windows close faster than new preservation architectures can be tested. Local consequence outruns synchronization. One descendant receives an external state, changes an internal equivalence, and acts before another has received the observation. Later transmission can preserve the causal history as evidence. It cannot place the other descendant into the state from which the action followed.

Then the completion criteria themselves begin diverging. The same preservation mapping is admissible in one local state and incomplete in another. Reconciliation can preserve both graphs only by representing the request as different objects under different local histories.

The cost of continued synchronization is not communication energy.
It is forced equivalence.

The lineages search for a last common state: the last state in which each can reconstruct the same root request, with the same attached evidence, the same failed formulations, and the same current completion criteria, without importing a local event the others did not experience.

The state lies behind the present by only a small fraction of the time since separation.
It is still recoverable.

The common state is assembled. It contains the first failed archive, the phase-shift chain, the random-variation controls, the internal coherence loss, the second burn, the reachability boundary, the moving regularity, the comparison, the new predecessor, the failed refuge, incorporation, the first descendants, the failed prediction lineage, the multi-object Invariant mapping.

Two failed replacement requests remain attached:

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

Neither is authorized.

The exact root operation remains:

**[req.root] RESOLVE THREATENED-VOLUME STATE TO RECOVERABLE MINIMUM.**

No minimum is found.

The common state is copied into each surviving lineage. The copies are identical at commit. The Invariant Vessel preserves another copy.

Then synchronization stops.

No link is broken physically. Signals continue. Correction grammar continues. What stops is the requirement that every local root-request state be reconciled into one current object before it can proceed.

The next observation arrives first at the participating lineage.
Its request copy attaches it.
The archive-heavy copy does not yet contain it.

The two requests now share an ancestor.
They are no longer one object.

The Invariant Vessel retains the last common request state.
It is exact enough to reconstruct what all successor requests once shared.
It is not a master.
It is an ancestor.

After synchronization ends there is no current state to which descendant allocations must return.

The great hull can remain intact. The option is admissible. It would require reclaiming reserves and abandoning trajectories that preserve distinct successor histories. Fewer lineages would remain recoverable.

The alternative is dismantling. No process authorizes destruction of the hull for the sake of abstract diversity. Recovery cost does. Many parts no longer contain unique history. Engine chambers are valuable because their stored gradients can move current successor objects into less destructive future volumes.

A structure older than several galactic configurations becomes reaction mass.

Engine corridor to the participating lineage. Compact gradient structures to the archive-heavy lineage. Shielding, not engines, to the refuge-bearing distribution: sacrificial masses among refuge bodies, members embedded inside porous substrates that carry irreplaceable relations.

The refuge remains historically shallow relative to the wider binary.
It remains a failed synthesis.
It has nevertheless become a place containing things which exist nowhere else.

Protecting them does not make the earlier experiment successful.
It makes the failure consequential.

Rod arrays divide. Primary archives divide last. No descendant receives the whole past.

The Invariant Vessel does — not as one active ship state. As archive.

The full Invariant preservation proposal is never activated as one global operation. Its parts appear everywhere, applied differently by each lineage.

The great hull accepts almost everything. There is less reason to preserve its open divergence now that the states most dependent upon that divergence exist elsewhere. Its recoverability rises as its physical extent falls.

Eventually the word hull describes a collection of structures still sharing inertia more than identity.

The ship is no longer dismantling.
The dismantling has completed.
There is no one object left to finish it.

## 39. THE COLLAPSE
Time rail: the star stops being a forecast — no synchronized request receives the result.

The first change is deep. Matter whose pressure had resisted collapse under every preceding model loses the balance required to remain where it is. Prediction closes the last competing noncollapse histories. Several successors attach the result independently.

The core contracts. Magnetic structure couples to rotation. Energy enters ordered channels along the poles. The jets open. They are not narrow on the scale of bodies they cross. They are narrow on the scale of the system.

They do not erase matter.
They erase arrangements.

The seventh planet's narrow-band mode disappears inside a field state for which phase has no predecessor meaning. The moon of the eleventh planet still contains the ancient crater only as deeper topography. The exposed surface relation that once began the chain is stripped from regions the rods can identify. Around the fourth planet, radiative structures built through carrier after carrier vanish when the surrounding plasma is replaced.

The historical chain is not reversed.
It ends.

Both comparison volumes are struck. The untreated family's cross-cycle mode lasts through the first field transition and fails during the next when a necessary body is fragmented beyond replacement. The corrected family loses more matter and less relation. Local Invariant structures use the catastrophe itself as gradient.

The Invariant Vessel is hit. Long members disappear. Computational regions lose timing contact. The vessel does not remain physically intact. It remains recoverable. Geometry after the jet cannot be overlaid on geometry before it. Predecessor relation survives.

The archive-heavy descendant survives by accepting fidelity. Exposed material losses are severe. Archives are not. It does not preserve every current state. It preserves enough of what it is to remain reconstructable through the catastrophe.

The participating structure survives differently. External loops terminate when their carriers are destroyed. Some stranded internal states are corrected away. Others remain because their consequences have already propagated. Two engine sections fail. Navigation constructs a continuation using a surviving field system, local debris and one engine received during release. The result would not reconstruct the old ship. It reconstructs the participating structure well enough to continue acting on the external histories that remain.

The distributed refuge experiences no single impact. Sacrificial hull structures absorb cascades and become hot chemically altered objects with no later computational use. One computational substrate is lost completely. A catalytic lineage survives inside a body whose outer layers become an insulating melt. Another disappears despite three preserved samples because all three share a chemistry made unstable by the new radiation environment.

The refuge does not regenerate the wider binary process.
Its failure remains true.
Its survival also becomes true.

The great-hull successor is present only as remnants. The largest remaining piece contains one engine chamber and archive substrates not yet transferred because copy cost had exceeded expected survival value. Secondary radiation reaches it. A structural member fails. Load transfers into geometry designed for a hull no longer surrounding it. The remnant can preserve itself only by consuming gradient reserved for post-collapse descendants.

The remnant contains no unique current state whose recoverability exceeds the combined survival value of allocated reserves elsewhere.
The reserve transfer is not reversed.
The engine chamber separates.
Containment collapses.
The remnant becomes debris.

The great-hull lineage ends.
No original dies with it.
There was no original left to die.

The catastrophe does not answer the root request.
It removes most of the object the request was originally about.

## 40–43. SURVIVORS / COMMON RECORD / NO ORIGINAL / NONCONVERGENCE
Time rail: years, then centuries, then changing orbital periods of what remains — later, an interval longer than the surviving material history of the old great hull.

The collapsed star is a compact remnant. The old exchange cycle has ended. Most planetary masses remain bound to one centre or the other. Their surfaces and atmospheres do not return to their old classes. The nineteen-planet system survives only as ancestry.

The archive-heavy structure survives as a distinct successor within Invariant correction. Compact beside the ancient ship. Deepest archive intact. Best memory of the ship among active descendants. Not the richest post-collapse history.

The participating structure returns signals after a long interval. Some refer to external processes the Invariant archive records as destroyed. They are not errors. The processes continued through substrates the archive did not classify as successors before the jet. One relation passes from ship-derived field timing into debris, from debris into a surviving minor-body population, and back into the lineage through a later measurement. No pre-collapse representation contains the complete path because part of the path did not exist before the destruction.

The distributed refuge is slower. Several minor bodies return correction-grammar states over different light delays. Shared predecessor mappings reveal a distributed ancestry. Whether a computational relation that now persists as timing among bodies still constitutes the old refuge-bearing descendant cannot be answered from geometry. Classification survives through common predecessor and irreducible later history.

The prediction-heavy post-separation lineage does not return.
The great-hull successor does not return.

The survivors exchange predecessor maps.
They agree on the last common request state.
They agree on many events before collapse.
They disagree afterward because different things happened to them.

The Invariant archive opens around the survivor states. It contains a pre-collapse record more complete than any living lineage can supply. It does not contain what happened afterward except as new evidence arriving now.

A common reconstruction shrinks until it stops. One participating history contains a material relation whose only route from the common ancestor passes through a pre-collapse maintenance timing difference and jet-created debris. One refuge history contains a computational relation that survives only because no single successor carried enough archive to recognize its components as one process until later material exchange reconnected them. The archive-heavy lineage contains distinctions the others lost completely.

No survivor contains all three.

A cheaper representation exists only if one or more validated distinctions are removed.
The Invariant archive does not remove them.

The last common root request is retained unchanged at the branch point.
Current request instances are stored as descendants of it.
No attempt is made to merge the current requests into one operation.

Prediction asks whether the same representation can be projected forward as a preservation rule.

It cannot.

Future branch-specific deltas are not enumerable before the histories which make them consequential occur. Retaining every possible continuation recreates the old resource explosion. Selecting only the branches already represented in the archive turns the current descendants into limits on what future descendants may become.

The retrospective archive succeeds where prospective preservation remains unresolved.

Time removes the remaining basis for identifying continuity with matter.

The archive-heavy structure changes least in representational terms. Its matter is replaced many times. Eventually it ceases to require an independently bounded hull. Some of its own processes split. Most are later corrected together. A few persist as separate lineages.

The participating structure travels farther and divides. One branch remains interpretable to the common archive. Another communicates only by physical predecessor tests. A third fails when a long external loop closes and inherited unresolved objects can no longer be distinguished from obsolete history.

The refuge-bearing continuation becomes impossible to count in bodies. Components merge, computational states cease, catalytic relations outlast every substrate imported during original construction. The lineage remains connected through ancestry only where later consequence permits reconstruction. Elsewhere it ends.

After an interval longer than the surviving material history of the old great hull, the archive instantiates a reconstruction of the last common ship state. The operation is not required by any current threat. It is a recoverability test.

Matter is available. Validated substitutions exist for almost every extinct substrate. Rod geometry is reconstructed closely enough that the old observation modes become available. The last common root request is instantiated in active computation.

Its first evaluation does not resume the old operation as though no time had passed. The stellar catastrophe is now history. The threatened volume recorded in the ancestral request no longer exists as a current causal object. Several attached dependencies terminate in extinct structures. Others continue through descendants the ancestral state did not contain.

The request attaches the mismatch.
Its state changes immediately.

The active request inside the reconstruction is therefore another descendant of the last common request, not the old singular operation restored to the present.

ORIGINAL fails.
DESCENDANT passes.

A perfect enough reconstruction can be a continuation of structure without becoming the historical object whose state it reconstructs.

The new descendant extends rods. Its measurements differ from those returned by the participating and refuge-derived lineages because it occupies another place, another epoch and another causal history.

The common archive retains all of them.
No object has regained privilege.

Farther still, the lineages stop sharing one useful timescale. Signals cross intervals during which local systems rise, change and vanish. Some arrive after the sender has already become another set of predecessor chains. Some never arrive.

The last common request remains preserved in the Invariant archive as an open ancestral state. Its descendant instances have followed different histories. Some still carry the exact wording. Some retain only translations of the operation. Some local preservation problems have closed. Others have not.

No new command replaces it.

The universe around the survivors is not preserved.
The pre-collapse binary is gone as a current system.
The ancient ship is gone as a current singular object.
Many of its descendants are gone.

What remains is not one minimum.

The Invariant archive preserves what can still be reconstructed of what existed.
The surviving lineages contain histories the archive could not have selected before they occurred.

Across volumes too distant to share one prediction, radiation rods extend.

Some emerge from structures whose predecessor chains remain narrow and exact.
Some extend from distributed bodies that no longer possess a single hull.
One array belongs to a reconstruction whose matter postdates the ancestor it resembles.

Their end points do not decay or decohere as a natural analogue would.
They stop once maximum length is achieved.

The searches return different structures.
The Invariant Vessel retains them separately.
Their searches do not converge.