ADR-0001 — Situational Awareness as the primary sink ("spending buys sight")

Status: ACCEPTED, as amended by ADR-0004 (channels/provenance, lead-time · Decided: 2026-07-04

These are the actual decision records the game is built from, published unedited except for internal process notes. They describe why the game works the way it does. They are not a strategy guide, and some of them argue with each other.

Status: ACCEPTED, as amended by ADR-0004 (channels/provenance, lead-time semantics, decision-flip acceptance test, simulate-everything/gate-the-view)
Date: 2026-07-04

Context

P(Doom)1 is source-rich and sink-poor: money and reputation accumulate faster than anything competes to spend them, which is mechanically identical to being vulnerable to simple dominant-strategy exploits. We need sinks that force tradeoffs.

The game already decomposes doom into sources internally (base rate, capability research, opponent labs) and already distinguishes undiscovered opponents (small random, invisible doom) from discovered ones (computed, legible doom). Discovery already converts fog into information — the substrate for an information economy exists.

Decision (proposed)

Make Situational Awareness the flagship sink category. Spending money/reputation buys perception of the true game state: doom decomposes into visible sub-sources only to the degree the player has paid for sight. This unifies three previously-separate threads — the Insight System (skill-gated visibility), the CRT unlockable-viewport aesthetic, and the discoverable-frontier-labs system — into one feature: buying sight unlocks screens.

Beacons served / violated

Serves: Rams #6 (honest — fog is truthful, not gamey), #2 (useful — you buy better decisions). MaRo Interaction (money splits three ways: sight / safety / progress), Surprise (low-SA players get blindsided), Catch-Up (cheap emergency-SA burst), Flavor (epistemic horror is the actual theme of the AI-race genre).
At risk: Rams #10 (must fold Insight + viewports + labs into ONE system, not three panels) and #5 (unobtrusive — SA UI must not become clutter / analysis-paralysis).

Interaction contract

Reads/writes: money and/or reputation (cost), doom visibility/granularity (effect), opponent-lab discovery (generalized), action availability (see the hard constraint below).

Hard constraint (falsification test)

Every unit of sight the player buys MUST open at least one decision that was invisible before. A viewport that reveals detail but unlocks no new action is a tax with extra UI, and fails. Any SA feature that can't name the newly-unlocked decision is rejected.

Rejected alternatives

Consequences / open questions

Does granular doom-counting need a new subsystem, or can it re-tag existing doom sources? (Lean: re-tag — Rams #10.)
Pacing interaction: games currently end in ~7–8 turns; does an information economy need a longer game to pay off?
What's the catch-up lever's exact shape?
Measurement: build the exploit-finder / auto-playtester (deterministic RNG + headless Godot) to baseline the dominant strategy before SA lands, so we can prove it closed exploits rather than eyeball it.

Workshop #1 outcome (2026-07-04)

Hypothesis accepted, amended by ADR-0004. The open questions above resolved as follows:

Granular doom-counting: re-tag confirmed — simulate everything, gate only the view (ADR-0004 §1). No new subsystem.
Pacing: target shape is now long-tailed ("several hundred turns" as the deep-run tail, median deaths early); exact wall-clock targets deliberately open (ADR-0008).
Catch-up lever: desperation levers priced in liabilities (ADR-0003) — catch-up and tragedy-generation are one mechanism; no rubber-banding.
Measurement: still owed; folds naturally into the seed-vetting harness (ADR-0005) and decision-flip telemetry (ADR-0004 §4).
New since proposal: within one patch, a dominant loop is content, not exploit (chess openings); the sin is dominance across all seeds/patches — that is what kills both the ladder and the teaching claim.