Hold The Horde — Alien Utterance Discovery

Status. One compact player-facing scene in which a phrase in the second alien language can reasonably be read as an ask rather than a record, with the strongest distinct alternative and a comparison of evidence and cost. It is a proposal for owner reaction: no grammar, no cosmic writer, no story truth. "The writer" below is whatever produces the second language, and every token is a placeholder. Recognising a shape is never treated as translation, and nothing here changes Hospital Story Directions, Language Expedition or Language and Journal.

The gap this answers

The starting correlation is: utterances ending one way (A) sit beside a physical change made at that spot; utterances ending the other way (B) never do. That fits "A asks, B refuses" and "A done, B pending" equally, so a further context is needed before the proposed captive's final line can be read as a request. This scene supplies that context. It does not by itself make the captive's request established; that remains conditional on the story choice and on enough evidence there.

The scene — the regulator gallery

What the player brings. The A/B correlation from four or five earlier places. The working rule for builders' runs from the Language Expedition, so a fitting's own run token is recognisable where it recurs. Two or three builders' workstations at which utterances begin with a slot holding a token that also appears on that builder's own fittings, which supports reading the slot as an address. And, if it was captured on the first expedition, the token stamped on the collar plate at the berth where the ship docked.

The room. A short gallery on a service line whose flow tell runs toward the sealed section. Four regulators in a row, each with its builders' run and, on a clipped plate beside it in the other hand, an utterance whose body contains that regulator's run token. At the far end, a panel holding a column of utterances with different address tokens, in the order of the workstations already found. The last line's address token matches the collar-plate token at the berth, and it ends in A. A blank reply face sits below the column, sized for a plate.

• R1: pulled. A. The pry scar runs through the last sign of the plate. The retaining band is a one-time crimp, cut once and not re-crimped.
• R2: pulled. A. Clean.
• R3: intact. B.
• R4: intact. A. The only A-line in the player's records with nothing done beside it.

Pulling a regulator is ordinary strip work: it yields a part and drops the line's tell one step, visibly, in the gallery. A player who reads nothing has that reason to pull and no reason yet not to.

The ambiguous observation and its rivals. R4 is the anomaly. Record: A means done, so R4 is a false entry. Ask: A means do this, and nobody did R4. Foretelling: A means this will happen. All three are reasonable, and the journal holds them as ungraded notes.

First discriminating context: the scar. The pull at R1 went through the words, so the words were there before that pull, and the one-time crimp is an authored extra clue that there was no earlier pull to record. For R1, "done" is out. What survives is any reading in which the words come first: an ask, a foretelling, or a logged job. The record reading is weakened, not eliminated, and B keeps every status it had.

First act: pull R4. Staged world-time work, noisy in the ordinary horde sense. The tell drops a step, which the journal records as the control's visible state and nothing more. Walking back past the panel, the player finds one more line under the last: the same berth-address token, R3's run token, ending in A. A before-and-after pair of stills shows the difference.

This shows a responsive process that names the one regulator still standing. It does not show a mind, a want or a prediction, and a job display that updates after a fitting changes would do exactly this. The rival is now "a pending-job board that reads the fittings."

Second act: answer without touching anything. The player unclips R3's plate, the B-line about R3, and seats it in the reply face. Nothing in the room changes. A new line appears: the berth token, R3's token, B. The line that named R3 stops being marked as current. Seating any other plate produces nothing.

The reply rule, stated once. The reply face holds one plate. It answers only a B-plate whose fitting token matches the line currently marked, and it answers it once: one acknowledging line per fitting, ever. An A-plate, a plate naming another fitting, or a plate carried in from elsewhere produces nothing and can be taken back. While the plate stays seated the deferral holds and nothing further is written; unclipping it makes the existing line current again only while that fitting's work remains incomplete, without writing a new one, and seating it again restores quiet silently. Completing the work closes its line whether or not it was deferred and advances to the next intact fitting; once none remains, there is no current line and nothing more is written. Removing a plate cannot reopen completed work. There is no rewriting of plates and no validator: the face reacts to which plate is in it and which line is current, nothing else.

That is the observation that separates the two rivals as far as this scene can: a display driven only by regulator state does not answer a plate, and here the words changed because words were given, with the world untouched. What it does not rule out is a job board that also takes manual command input; the player cannot tell a correspondent from a console, and the scene does not claim to. Operationally, the player has found an interaction protocol: a thing that names work, is acted on or answered, and names the next work or goes quiet. The reply interaction costs one reply face and three authored responses. Without it, the gallery cannot distinguish a display driven only by regulator state from a process that also accepts input.

What the player can now predict, and do differently. At a later panel that carries a current A-line naming a fitting, the gallery supports a prediction to investigate, not a guarantee: that doing the named work draws the next line, that seating the fitting's B-plate draws an acknowledgement and quiet, and that anything else draws nothing. So they can defer an ask without disabling a room, do the work to see what is named next, or simply ignore it. Before the gallery, the lines were a ledger and there was nothing to do. After the scar, an A-line beside untouched work is somewhere to go and look. After the reply, the last line at the seals, addressed to the berth and naming the seal controls, is something the player can answer, defer or ignore, and can also close for practical reasons without reading a word, as the story page already allows.

What honestly remains. Whether A is "do", "next" or "will", and whether B is "no" or "not now": the player has learned the force of two endings in one exchange, not their meaning. The berth token is a modest cue that the address means whoever is at the berth; it does not prove the writer knows the astronaut, and a token absent from the player's records is not absent from the station. A responsive process is not shown to be alive. What the line feeds and why anyone wants the regulators down belongs to the story.

Consequences, run through the rule. R4 first: the next line names R3, and seating R3's B-plate is the valid reply. R3 first: R3 was not the named work, so nothing is written for it; pulling R4 afterwards finds no intact regulator to name, so nothing more is written, and B goes less tested; stated as a cost. Ignoring R4: the line stays current and nothing happens; R4's own plate is an A-plate and draws nothing if seated, which is fine, since a player may simply not answer. Invalid replies: R1's or R2's A-plates, or a plate carried in from elsewhere, draw nothing. Death and return: completed pulls, a seated plate and every written line persist as finished work and placed objects, and the gallery is one room, so a shift moves it whole; re-entering writes nothing by itself. A plate found still seated on return keeps its deferral only if its fitting's work is still incomplete; unclipping restores that same unfinished line. A completed fitting stays closed whether its plate is seated or removed. Missing the scar, the plates or the panel: the regulators still yield parts, the seals still close by hand, and no basic interaction is gated on reading. Plates can be unclipped again; nothing here is one-way except the crimps.

Journal and help. Capture and plate-seating are world time; comparison and notes are paused, as the owner requires. The journal records stills and the control's visible state; it never labels a line, grades a note or announces a translation. Yeva can truthfully say the crew never touched the regulators and found the pulled ones pulled, and that the panel has more lines than she remembers. That is weaker evidence than seeing a line appear, and the scene never depends on it; she reads neither hand and says so.

The strongest distinct alternative — the exchange chamber

A service chamber with two labelled shutters and a device that holds traces of past exchanges: who spoke, which control was offered, what the recipient did, what the shutter then did. Early traces show a phrase followed by a shutter closing; later traces cross the factors, the same phrase in the same state answered once by closing and once by a short reply that leaves it open, and one trace with sender and recipient swapped. A reversible practice circuit then lets the player take the recipient's place and reply with a copied utterance.

Its evidence is cleaner than the gallery's: two responses to one phrase in one state is a direct contrast, and the practice circuit makes every guess cheap and reversible. Its costs: a trace device and four to six authored traces, all of which report exchanges the player did not witness; the crossed traces read as a prepared lesson; and the first discovery arrives as data on a device rather than as changed text found after the player's own act. A cheaper variant in either scene is an attributed partial gloss left by researchers, explicitly fallible, which moves the first insight to testimony and turns the activity into verification.

Comparison

Preference

The gallery with the reply face. The two things it lets the player predict, that named work draws the next line and a seated B-plate draws quiet, are earned by acts the player chose, and the discovery is finding that the text moved. The exchange chamber is the honest fallback if later authorised observation shows players never return to the panel or read the scar as damage. Neither is a linguistics proof: both let the player predict a responder's behaviour and reserve what the responder means.

Evidence limits

The scar dates the words before that pull only, and the crimp is an authored clue, not a general property. "Only A" means only in the player's records. A token never recorded is not a token proven absent, and the berth cue makes the address probable, not certain. Text that answers is a responsive process, not a living writer; Yeva's memory is weaker than a witnessed change. A status board could update after a pull, which is why the plate is the decisive act; the plate rules out a display driven only by regulator state, not a board that also takes manual command input, so the player learns an interaction protocol and not a speech act, and a later panel is a prediction to test, not a guarantee. Every mistake keeps work; no escape interaction needs a reading. Nothing here establishes fun, comprehension or authoring cost; later authorised observation would test whether players who pulled R4 return to the panel unprompted, and whether they describe the last line as asking them something.