**Status.** An early-reaction comparison of two proposed ways to make permanent connections, for the owner's selected direction: certain actions join rooms into groups that stay connected through shifts, and connecting the entire station is one of the deepest and latest puzzles. Nothing here is canon — not the machinery, not the architecture, not any ending. Early shift uncertainty and persistent selected connections are preserved by both.

**Seams.** Every coupling face carries a short signature in the builders' cut script. Two sections whose facing signatures match were built adjacent. Binding is engaging the coupling lock at a matching seam; a mismatch visibly refuses.

**Live service continuity.** Rooms stay together when a service main runs live and continuous between them, because the machinery that re-seats fabric will not cut a running line. Binding is restoring the run.

They answer **different questions**, which is the finding of this comparison. Seams answer *which pairs belong together*. Continuity answers *what makes a join hold, and in what order joins are possible*. Each original supplied one answer and improvised the other.

Not as a merge to satisfy two authors — as a combination in which each part does the job the other underspecified.

- **Legality is a seam signature.** Match or refuse, visible at the flange, cheap to test, easy to teach.

- **Persistence and ordering are service.** A bind holds because a run is live across it, and what constrains the *order* of binding is the plant.

**The one visible constraint that makes order matter: a main carries one live source.** Each main has a source fitting that is driving or dead, and a trip indicator on the line. Put two sources on one main and it trips at both ends and nothing stays live.

**A working bind.** Blue Ward's east flange and Lift Service's west flange carry the same signature. The cable main between them is dead at a pulled contact. Seat a relay in the contact: the run goes live from Lift Service's source, the coupling holds, and the pair is bound.

**A failed one, and why.** Cold Storage's flange does not match — it refuses at once, costing the trip and nothing else. But the instructive failure is the legal-looking one: Cold Storage sits on a stub with its own live feed, and joining it to the same main would put two sources on one line. The signatures match, the lock engages, and the line trips visibly at both ends. The bind does not hold until the other source is killed — and that fitting may be in a section currently somewhere else entirely.

**This is why binding everything greedily fails.** Order matters: some joins require de-energising a source that is presently the only feed for a region already bound. A player who binds whatever matches will strand regions and have to undo work. The global insight is that the station needs a **source plan** — which fitting feeds which region — and that is reasoning, not collecting.

| | Seams alone | Continuity alone | Combined (recommended) |

|---|---|---|---|

| Before | Read facing signatures | Read stamps and find the break | Read both: does it match, and can it be fed |

| Action | Engage the lock | Restore the run | Restore the run at a matching seam |

| Next death | Pair moves as one unit; rest re-seats | Same | Same |

| Late game | Predict *when* a true pair will be adjacent | Solve *what order* joins are possible | Solve the feed for an arrangement you can already identify |

Signatures describe the builders' original arrangement. The mains describe how that arrangement was fed. The discovery that makes both useful at station scale is that **the original arrangement is the one in which every main can be fed from its own source** — so completing the station is not matching pictures until they run out, it is solving the feed for the whole plant. Every stamp and signature already recorded becomes evidence for one problem.

**Brute force is not blocked, in either.** With seams, trying every adjacency fails cheaply at refusals, so a patient player can die repeatedly and test their way through without ever grasping the cycle or the feed. The late insight changes the *cost*, not the possibility: it turns an unbounded search into a planned move. **Accepted cost, stated:** someone can finish this without understanding it, and the design should accept that rather than gate it with anything artificial.

**Correcting a dismissal.** My original called the registration alternative bureaucratic data entry. That was a label, not an argument. A terminal that refuses with a stated reason is exactly as visible as a tripping line, and registration ties binding directly to the language requirement. The real difference is only *where the constraint lives* — in a readable device, or in the plant — and either can be legible or opaque depending on execution.

**Predictability does not survive unchanged.** If unbound sections advance through an order, binding removes sections from the pool, so records taken under one lock state cannot be trusted as-is afterwards. The proposal that keeps this honest: the order runs over **coupling points**, bound groups occupy a point as a unit, and the relative order among still-free sections is preserved. Then old records still help, re-read against the current free set — which the player knows, because they made it. A **release** is different: it puts a section back into the order at a position nobody can know without observing again. Undoing therefore costs predictive knowledge as well as a trip, and that is a real cost, not a formality.

**Reachability of a mistaken bind's release point.** Reversibility is only meaningful if the undo can be reached. It can: the release point is the seam or contact the player installed at, and that lies *inside* the group the bind created. A group's interior is permanently connected — that is what binding means — so reaching any room of the group reaches the release point. The group itself keeps arriving somewhere, because it still occupies a coupling point among a re-seating remainder, and the pathological case of a group sealed away from everything cannot arise: if everything were bound, everything would be one group and all of it reachable. The cost of recovery is the trip, the retrieved part, the lost prediction, and the re-scattering of what had been stabilised.

**Completion is two different burdens.** One connected network is a *spanning* condition — enough live runs to join every region. Every original seam is strictly larger, since a spanning structure needs fewer binds than the full original arrangement. Both satisfy the owner's requirement that every region really belong, and neither may be reduced to a few hub rooms, but they are not the same amount of work and the choice should be made knowingly.

**Stabilisation is a proposed exception, not a footnote.** In both, when nothing is left unbound, nothing re-seats. That is a change to the confirmed every-death shift rule and it is proposed here, not assumed: it needs the owner's acceptance. Contents would still vary — population, hazards, machine states — so the place is stable in *shape* rather than inert. **An alternative worth putting alongside it:** designate a remainder that is never bindable, so shifting never fully stops and the confirmed rule survives intact, at the cost that "entire station" then means the network rather than every last section. Either way, completion is separate from any ending; the owner has not said it frees anything or is required.

**Proposed authored requirements, not established facts.** A unique original arrangement, a discoverable order among re-seats, and a feed problem that admits a global solution are all things a designer would have to *build*. None is demonstrated, and each is a real authoring commitment.

Whether the one-source constraint reads as reasoning or as arbitrary plumbing. Whether the order among re-seats can be made legible without a tracker that solves the map. And which completion burden the owner actually wants, which is the question I would put first.