**Status.** A provisional information contract for the two-language expedition in [Language Expedition](/Hold%20The%20Flood/Language%20Expedition). It extends the working [Clues and Help](/Hold%20The%20Flood/Clues%20and%20Help) and [Oxygen and Observation](/Hold%20The%20Flood/Oxygen%20and%20Observation) rules; it is not owner canon or a complete interface. The owner requires the journal to pause the rest of the game, including enemies and oxygen, even during combat. That does not change here.

The journal preserves what was observed, supports comparison and records a voluntary test. It never grades freeform notes or declares a word solved. A local prediction may be **matched here** or **did not match here**; both labels describe the encounter, not universal meaning. Survivor help remains attributed testimony, not system truth.

This bounded confirmation is reversible because later evidence can change the interpretation while the original image and result remain true. It gives a player enough feedback to distinguish “my prediction failed” from “the machine never received power” without turning the journal into an answer key.

| Step | Player interaction | What the game may assert | What remains the player’s |

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

| Acquire | Frame and commit an inscription with visible context and location | Captured pixels, place and time; nothing outside the frame | What mattered, what was omitted, whether to spend world time |

| Compare | Pin records side by side; zoom; manually highlight candidate recurring groups | Which crops the player linked; an optional visual-match suggestion is labelled software-generated | Whether marks are the same sign, word or role |

| Interpret | Write a freeform note; optionally link the current route to a stored destination as a prediction | The note’s authorship and revision history; the selected links, not their truth | Meaning, confidence and generalisation |

| Test | Commit the link, leave the journal and traverse voluntarily | The location actually reached and whether it is the selected stored destination | Why the route connects and whether the language rule applies elsewhere |

| Ask | Show a specific record or comparison to a survivor | Speaker, exact statement and stated basis | Whether to trust, combine or challenge it |

| Retain | Reopen after death or on a revisit | Original record, annotations, testimony and past result | Reinterpretation under the new context |

Capture, positioning, waiting for a change and manipulation happen in running world time and consume whatever oxygen/exposure the situation already creates. Opening, comparing and annotating in the journal pause the world. A paused view cannot reveal a new angle, sample a later state, operate a control or complete capture.

The existing framed still, visible context and location remain. This article adds four bounded tools:

1. **Comparison board.** The player manually pins records and highlights a candidate group. Links display as “player-linked.” On request, visual-similarity suggestions are explicitly recorder output and may be wrong where damage or handwriting differs.

2. **Revision-preserving notes.** Editing a gloss or hypothesis keeps the earlier wording available, so later contradiction shows learning rather than silently rewriting history. Freeform text is never parsed for semantic correctness.

3. **Prediction/result pair.** Before an action the player can link a route or control to a stored destination/outcome. Afterwards the journal attaches the observed place or state. This uses selected records, not language processing of the player’s prose; it awards no correctness, dictionary entry or general rule.

4. **Attributed help.** A survivor response records who spoke, what they directly did or saw, and what part is interpretation. Showing one record cannot cause knowledge of unseen inscriptions.

These are proposed changes. Automatic labels, a confirmed dictionary, batch completion, undiscovered-site hints and semantic scoring remain excluded. All four additions survive return and death as knowledge, not cargo. A revisit never overwrites an old still or outcome: the player captures a new state beside it, so “matched here” remains a historical result rather than a promise that the route or mechanism is unchanged.

At the three-way junction the player captures `KETH SORAN VEL` above the left way, `SORAN KETH VEL` above the middle and `TAAN` above the right. They pin the first two, highlight the swapped groups and may note “first token says what a place is for; later tokens qualify it.” The note is ungraded.

Walking the ways is voluntary and costs ordinary oxygen/exposure. The left reaches a pump hall crossed by cabling; the middle reaches a cable junction with a pump. Their contents support positional order more than place-name or meaningless-order readings. The bare `TAAN` above a stair-only vestibule is another consistent case, not proof. The journal stores each destination record beside its run and may append **supported here** only to a committed destination link. It never marks `KETH` as “pump,” asserts a grammatical subject or upgrades the note to fact.

Yeva can truthfully call the first two the crew’s pump road and cable road because she worked this spine. Her attributed gloss supplies referents, not the positional rule. The player must still compare order, interpret `TAAN` and decide whether the pattern applies elsewhere. Multiple persistent runs give an unaided route without guaranteeing every player notices it.

The player separately records a visitor mark ending `OSH·a` beside the middle doorway. Much later, beside a far bay whose builder run begins `SORAN`, they find the visibly related `OSH·b`. The player can pin both halves and manually link them. A recorder similarity suggestion may surface the old crop when the second is selected, but it is labelled a visual match; the software does not assert that the forms pair or what their relation means.

Together the halves support a relation between the two locations. The builder runs independently say both places are about cabling. The player may therefore link the middle route to the far-bay record as a prediction and attempt the suspected shortcut. Arrival at that bay appends **matched here** to the route link; arriving elsewhere appends the actual location and contradicts that link. Neither result translates `OSH`, proves every pair is a route or makes either language redundant.

Yeva says plainly that she never read the visitors’ hand. Help remains useful because her road names constrain the builder referents; the visitor relation and combined route inference remain the player’s. If `OSH·a` was missed, another persistent pair can teach the distributed structure, while the builder runs alone still support a weaker route hypothesis. A wrong guess consumes the detour’s oxygen but reveals an actual destination; the known long route or retreat remains subject to present supply and threats.

**Withhold everything.** Consequences alone shape belief; even a successful local prediction receives no record status. This maximises ambiguity and ownership and resembles shipped approaches where a wrong reading can feed forward. Its cost in this game is diagnosis: oxygen, damage, missing power and contextual exceptions can make a sound hypothesis look wrong, while a lucky outcome can preserve a bad one.

**Bounded confirmation and help — preferred.** Record the visible result, label only its relationship to the player’s stated local prediction, and keep testimony attributed. The player chooses what to compare, words the hypothesis, decides whether one case generalises, reconciles conflicting human knowledge and pays the world cost of acting. The cost is authored outcome states, comparison consistency and a risk that status labels feel more authoritative than intended.

[Language Design Evidence](/Hold%20The%20Flood/Language%20Design%20Evidence) describes shipped designs that use uncertainty, candidate meanings, repeated contexts and forms of validation; its sources offer no public dataset proving which contract transfers to an oxygen-limited fixed station. The reported multiple-context precedent supports planned redundancy, not a mandatory number of instances or per-sign guarantee here.

Reopen the provisional choice if later authorised observation shows players treating **supported here** as a dictionary answer, cannot diagnose failures, or use comparison as clerical transcription. A stronger confirmation layer or fully withheld feedback would then be a real alternative. Nothing written here establishes comprehension, fun, authoring feasibility or campaign duration.