**Status.** A provisional proposal, not owner-approved. It works one expedition in which two alien languages both contribute, and shows what the player observes, believes, tests and decides. Token strings are readable text placeholders; no glyphs, alphabets, conlang or vocabulary catalogue are proposed. Who the builders and visitors were is replaceable candidate fiction, and the episode depends on no central-mystery direction.

**Builders' runs.** Left by whoever built the structure, above and beside fittings and doorways, as short ordered sequences: `KETH SORAN VEL`. The candidate rule available to the player is positional — that the first token is what the run is about and the rest qualify it. A run is read where it stands.

**The visitors' hand.** A later species wrote on the builders' surfaces in a visibly different hand. It is a language, and the player meets one *register* of it here: marks placed in two halves at the two ends of something, each carrying a matching bracket token — written as `OSH·a` and `OSH·b` — plus one further token. Pairing is a property of this use, not a claim about the whole language.

Acquiring a half requires being where it is. Comparing halves happens later in the paused journal, once both records exist.

- The runs are compound place names built from a shared stock of elements.

None is silly. The compound-name reading in particular survives a long way, and the article does not pretend otherwise.

Two mains run the length of this spine: a pump main and a cable main. Both are visible as continuous runs, and in any given room a main either continues out the far side or comes to an end.

The **test is voluntary and physical**: walk a way and look. It costs oxygen and exposure like any traversal.

- The left way ends in a **pump hall**. The pump main terminates here — the pumps are its end. The cable main crosses the room, entering one wall and leaving another.

So the observable is: **of the two mains, which one ends here.** In the pump hall the pump main ends; in the cable junction the cable main ends. The token order tracks that — `KETH` leads over the way where the pump main ends, `SORAN` over the way where the cable main ends.

**What this supports, and what it does not.** It supports the purpose-and-order reading over the stylistic one, because order tracks something the player can see. It does not eliminate compound naming: two places could carry different compound names from the same elements and the correlation would look the same. The player leaves with a working rule, not a proof.

`TAAN` over the right way adds a third case — a stair to a vestibule where neither main runs at all. That sits comfortably with a bare subject having nothing to qualify. On its own it settles little; a short name among long ones is unremarkable.

A pair marks the two ends of something. What tells the player *which* end is which is not their own route — walking a passage one way says nothing about how the visitors used it.

It comes instead from an independently observable one-way process. On pairs the player has already recorded, the two ends are the ends of a **transfer** whose direction is visible without reading anything: at one end a driving device pushes into the line, at the other a vessel receives and fills. Across those pairs, one further token recurs at the driving end and the other at the receiving end.

That is a small, checkable interpretation of local roles, without settling the tokens' full meanings or the language's grammar. It says the visitors' pairs distinguish a source end from a destination end, and it can be got from two or three pairs whose transfer direction is physically legible. What the tokens *say* remains open.

Much later, deep on the far side and reached the long way round, the player finds `OSH·b` beside a bay whose builders' run begins `SORAN`.

Set against the recorded pairs: `OSH·a` at the junction carries the **driving-end** token, and `OSH·b` at the bay carries the **receiving-end** token. That supports a candidate transfer connection from the junction to the bay, rather than establishing a walkable route.

There is also an untraversed continuation. The middle way reaches the cable junction, where the player turned back on supply. Beside the source mark, a transfer conduit disappears into a human-sized maintenance passage that continues beyond the junction. At the far bay, the receiving fitting stands beside another maintenance opening. These visible service accesses make investigating a possible connector reasonable; a conduit can still cross an impassable gap, and neither opening proves a continuous passage.

**What each language contributes.** The pair says the visitors treated the junction as a source end and the bay as its destination. The builders' runs say the middle way and the far bay are both about cabling, which is consistent with that rather than proof of it. Together with the visible service access, they give a reason to investigate that continuation for a connection to this particular bay. Neither is claimed to be indispensable in general, and a player with only the runs has a weaker version of the same idea.

The player can investigate the continuation past the cable junction as a possible way to the far bay, or take the long route they have already walked. A connection might offer a shorter journey. Its walkability, destination and actual oxygen cost remain unestablished until traversal; material transfer alone proves none of them.

If they arrive somewhere else, the journal records the arrival they actually made against the route link they set. That is a record of where they went, not a verdict on the reading. A blocked or abandoned attempt records as incomplete rather than as a refuted idea.

The cost is time, oxygen and interrupted work. Under the owner's rule, carried property comes home regardless, so a detour does not put equipment at risk. What it can cost is the rest of the trip: the player can press on, retreat to the known route, or end the expedition by dying and arrive home with everything. World time and supply still matter, and none of those outcomes is free.

**A wrong rule is not necessarily exposed by one trip.** A player who decided the last token is the subject can make a prediction that happens to hold, or fail for an unrelated reason. What one walk gives is an observation, and observations accumulate.

Runs and pairs recur, so no single instance is load-bearing. A player who missed `OSH·a` can learn the paired structure from another pair; a player who never completes one still has the builders' runs and the weaker route hypothesis they support.

What the design does not claim is that any two swapped runs, or any short run, will do the same work. The rooms here carry a visible which-main-ends difference; a different pair of rooms might carry nothing comparable. Nor does it claim universal access: a player who never compares occurrences may not reach the positional rule, and the survivor route does not fix that.

Yeva worked this spine for years. She can say what the crew called two of the ways — the pump road and the cable road — and she says plainly that this is what the crew called them, not what the writing says.

**Precisely what that gives.** It does not translate a token. It fixes which *place* each run sits above without the player walking there, so runs can be lined up against known destinations from the record alone. The token-to-purpose mapping, the positional rule, and whether order carries meaning at all remain inferred.

She never read the visitors' hand and says so, so the pair, its direction and the route inference stay with the player.

On request, the journal may surface a previously captured record that looks visually similar to one the player is examining. It is fallible, it is separate from any claim about relation, direction or route, and manual comparison remains fully usable.

**Reason and cost.** With pairs many expeditions apart, a player can hold both halves and never bring them together. On-request similarity reduces some unaided noticing. It does not interpret, relate or route, and everything above still has to be done by the player. Whether unaided noticing is central to the discovery remains unestablished.

**On confirmation.** Recording an actual arrival against a stated route link is not a correctness badge on a hypothesis, and this episode asks for nothing more. Without any record, diagnosis becomes harder rather than impossible: environmental evidence stays visible, and a player can often see why an attempt failed.

How records are compared and revised belongs to [Language and Journal](/Hold%20The%20Flood/Language%20and%20Journal). Also open: what these tokens mean beyond the placement Yeva's road names allow, who the visitors were, and why they wrote across the builders' walls. Nothing here establishes that inferring a positional rule is enjoyable, or that players will test rather than wait to be told.