**Status.** Practical comparison for the owner's selected direction: certain actions connect rooms or groups persistently, and connecting the entire station is one of the deepest/latest puzzles. The owner has not selected the mechanism or answered whether completion stabilises the layout. Everything below is proposed; no story ending, corruption rule, exact machine, or implementation is established.

A death can place two unbound thresholds physically **adjacent**. Adjacency only makes an attempt possible. **What moves is a rigid placement unit, not a loose room:** each authored compact assembly and each set of rooms already latched together preserves one internal pose, and an unanchored group may translate and rotate only as a whole, never reflect. The ship-rooted group containing R01 is anchored and neither moves nor turns, so a shift changes which unbound group meets its free outer frontier rather than which room touches the airlock. See [Death Shift and Rescue Continuity](/Hold%20The%20Flood/Death%20Shift%20And%20Rescue%20Continuity) for the full placement obligations. A persistent join is **latch-compatible** when the visible service signatures at those two faces can exchange the required pulse. A mismatch refuses before the latch completes. A successful mechanical latch opens a walkable threshold and survives later deaths until deliberately released.

Power is an activation tool, not a global ordering rule. A source may charge seam A, be recovered without undoing A, then charge seam B. Both latches remain. Scarcity alone therefore adds travel, not depth.

The owner-selected opening—one reusable part operating both carried equipment and station machinery—can prepare the player to look for shared causal behaviour. It does not prove that room machinery uses personal-rig ports or the same protocol. That relationship remains analogy unless separately adopted.

Rooms visibly accept one service-pulse pattern and emit another after their machinery acts. Reversing or reconfiguring a rare room changes its exposed pattern. The late realisation is that previously “incompatible” regions need converter rooms between them: the station can be assembled as a causal network, much like a build at room scale.

**Earlier clues.** The player has seen Decontamination's forward mode accept amber-single and emit blue-double, then its reverse mode accept blue-double and emit amber-single; a Freight Split branch one pulse to a shutter and rack; and a charged Foundry–Archive latch refuse while both faces visibly pushed blue-double outward. The reused source was sufficient—the directions were not.

**Strongest case.** It gives the player a predictive rule and an action beyond finding matching doors. Familiar rooms acquire new uses, and multiple converter sequences can support authorship rather than one prescribed solution.

**Cost.** If every room has modes and transformations, the station becomes an enormous circuit editor. Signals, directions, walking paths, and latch states could overwhelm the discovery. Authoring must guarantee readable converters and viable constructions without quietly giving the player a solved network.

Most thresholds belong to visible signature families. A bridge room has an A-family face on one side and a B-family face on another. It can join an A group and later present its B face to a B group.

The evidence must actually show the difference: inspecting Glass Theatre reveals an A signature on its west face and a B signature on its north face; records across different shifts show the west face beside A-family rooms and the north face beside B-family rooms. Two identical marks on one face would not establish a bridge.

Binding the A/B bridge to B first normally leaves its A face available, so order alone does **not** create a failure. The real discovery is which rooms span families and which exposed face is needed, not a hidden “correct first side.” A mistake occurs only if the player uses a needed bridge face on an irrelevant same-family branch and thereby buries or occupies that face; that must be visible and reversible.

**Strongest case.** Families make old adjacency records useful without requiring every room to be programmable. A rare bridge can explain why two broad regions never meet directly.

**Cost.** Once bridges are identified, patient matching may solve the rest. Its late insight may be classification rather than construction, and invisible family boundaries risk making ordinary bad luck look like impossibility.

Every threshold displays one counterpart signature. When compatible faces become adjacent, the player charges and latches them. The whole station is complete when all regions join one network.

**Strongest case.** This can still support discovery: the player learns where a rare counterpart appears, recognises a previously ignored face, and chooses which current opportunity matters. It is clear and inexpensive to teach. It has fewer stateful mistakes than transformer modes, but still needs the same authored guarantee that a mistaken join has a reachable release.

**Cost.** The deepest puzzle risks becoming the longest collection of pair matches. Knowledge helps search, but it does not substantially change how the player acts.

Use **simple signature compatibility as the common rule**, but give a small number of bridge rooms a visible transformation and, occasionally, a reversible mode. Most rooms remain static. The late discovery is not “every room is a circuit” or “bridges have a secret order”; it is that a familiar service room can convert the family presented by a growing group and thereby join regions that no direct seam can connect. This provisional recommendation assumes rigid mechanical joins and is therefore compatible with eventual layout stabilisation; the pending owner answer can reopen it.

This combines the family model's legibility with the transformer model's actionable payoff while accepting more complexity than unique matching. The main unresolved cost is authored combinatorics: converter scarcity could still create one mandatory sequence, and mode displays could become a disguised solved-map interface. The design should allow more than one viable converter placement where feasible, but prose cannot establish that the complete content set supports it.

**Failed.** The Foundry Block group and Archive Stack group are physically adjacent after a shift. Each exposed threshold shows blue-double flowing outward. The player brings a charged source and attempts the join. The latch preview displays two opposing output arrows and refuses final engagement. The source remains available and no existing bind changes. More charge or sequential source reuse cannot turn either output into an input.

**What the later insight changes.** Synchronization Gallery shows only its own locally observable behaviour: a test pulse enters each of its current thresholds, and the display shows the pattern before and after the gallery's present machine state acts. It does not import, curate, or explain the player's old captures. In the paused journal, the player may compare that behaviour with their own Decontamination, Freight, and Foundry–Archive records and infer that familiar rooms can be converters. The player can then predict that a converter sequence may expose the receiving pattern Archive needs.

**Successful sequence.** Proposed example:

1. Reverse Decontamination so its free face accepts Foundry's blue-double and its other face emits amber-single—the exact inverse pair the player observed earlier. This machine state persists.

5. Send a local diagnostic pulse through the new path to verify that the chosen converter states behave as predicted. This is useful evidence and troubleshooting, not a second condition for geometric completion.

Because new adjacencies arrive on death, the player may deliberately die to request another arrangement after finishing useful work in the current one. The owner's carried-property rule makes that a legitimate late strategy rather than a recovery penalty: carried equipment and loot return, consumed supplies remain spent, and installed work stays in place. The player may instead pursue other expeditions until a natural death shifts the station. Hunting one exact face orientation could still become repetitive death cycling; its frequency is an unresolved pacing and authored-distribution cost, not a guarantee this document solves.

- **Walking connectivity:** a completed latch leaves a traversable connection. Changing a converter mode does not close it.

Thus a player may have every region physically joined even while part of the diagnostic signal is unpowered or misconfigured. Physical inclusion is the base objective; tuning the pulse is optional local evidence and troubleshooting unless a separate later proposal gives it a distinct payoff. A mode change never silently destroys permanent work or turns a joined region into a false “incomplete” result.

Late station-wide construction should extend a **ship-rooted walkable network**, or merge an encountered island into it. This is a proposed authoring rule, not something prose proves about every possible shuffle.

Suppose the player latches Imaging by its red face to an irrelevant branch, occupying the face Archive needs. Because the new join extends the ship-rooted network, the player can walk from the ship to the release lever on the network side. They may first try a mode change; the walking latch remains while diagnostic coverage changes. If the wrong physical face is occupied, they deliberately release that one seam. Other latches and machine states remain. The released rooms stay adjacent for the current life and become eligible to separate only on the next death-triggered shift; random power loss erases nothing.

- **Why:** a concrete failed prediction becomes a solvable adapter problem, earlier machinery gains new meaning, and all regions can form one player-authored network.