hold the flood/development and validation handoff.md ..
@@ 25,6 25,7 @@
| **2. Do language clues support a warranted change of belief?** | For each representative chain, list observed evidence, competing readings, the discriminating action, correction route and useful later consequence. Separate an operational response from proof of a request, person or god. | Test a small chain with unfamiliar real players; record hypotheses before and after evidence and whether they can use the result in another context without being told its answer. | If the chain supports several meanings equally, strengthen discriminating evidence or retain honest ambiguity. If players solve by interface labels rather than observation, revise those labels or the learning objective. Agent agreement about a fictional sentence is not player evidence. |
| **2. Does one short rescue create interest in a person and life aboard?** | Resolve one specific living/abort/death crossing contract and write a motive, disagreement and later contribution for the selected survivor. Preserve the owner's short-rescue direction. | Observe the bounded rescue and a short later home interaction. Ask what the player remembers, expects and wants to do with that person; measure the actual crossing burden rather than calling it short. | If the person is remembered only as a mechanical unlock or the rescue feels like an escort chore, change character interaction and crossing before extending the cast. Do not create survivor mortality or a colony upkeep tax merely to manufacture concern. |
| **2. Can essential information work through the chosen accessible controls and cues?** | Produce the cue/interaction inventory from [Player Information](/Hold%20The%20Flood/Player%20Information), including alternative channels, input options and unavailable/unconfirmed states. Keep factual limits equivalent. | After implementation authorization, inspect the actual supported input/display modes and evaluate relevant tasks with real consenting players, including people who use the intended accessibility options. Define representative devices and conditions before claiming coverage. | If a cue only works through one unavailable channel, add a useful equivalent without revealing hidden routes, translations or conclusions. A simulator, checklist, narrated agent walkthrough or a second visual cue cannot by itself establish accessibility. |
+
| **3. Do successive expeditions continue to offer new discoveries and build decisions?** | Specify a representative sequence with connected objectives, meaningful revisits, clue consequences and part acquisition. Record the different decisions each return is intended to unlock; keep repetition and adjacency-hunting risks visible. | Once the required content and phase are authorized, observe real players across that sequence: what understanding and builds change, why they choose another objective, what earlier knowledge they reuse, and where repetition replaces a new decision. Keep this player evidence separate from authoring-cost and hardware measurements. | If novelty or build choices run out, reopen reuse, gates, acquisition cadence or the content band before expansion. Mixed results call for a narrower comparison of the suspected cause, not a favorable average that hides it. A positive result supports this sequence and these players only; it does not establish the full campaign's depth or the30/90/100–200-hour ambitions. |
| **3. Can the finite campaign be authored and run within a chosen production scope?** | Deduplicate the SMALL inventory, name reuse and state complexity, identify asset categories and target-platform constraints, and prepare explicit scope cuts. | After authorization, author representative room/puzzle/encounter/cast content through the actual pipeline and measure revision burden. Run representative technical workloads on specified target hardware, including hordes, connected builds, shifts and save state. | If content or technical cost is too high, cut repeated variants or simplify costly systems before extrapolating to the campaign. Re-estimate using actual distinct work and reuse. External game length, agent writing speed and room counts do not demonstrate our playtime or performance. |
Story endings and resurrection consequences remain owner choices. Evidence can show whether a proposed causal chain is understandable and what tradeoff an ending communicates; it cannot decide the owner's preferred theme by majority vote. Irreversible consequences require a visible decision boundary in the design before they appear in any later experiment.
