ADR-0004 — SA amended: channels with provenance, lead-time semantics, decision-flip test

Status: ACCEPTED — this amends and accepts ADR-0001; build second, after ADR-0003 · 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.

Context

ADR-0001 proposed "spending buys sight" as the flagship sink. The workshop stress-tested it against Pip's peak verb — "triaging in a race against death" — and against the realization: "I am used to thinking about the game with universal powers as god mode / admin, whereas the player mostly starts ignorant about the world the same way a fresh spawn in Tarkov / fresh game of Civilization / entirely fresh run of some Roguelike." The hypothesis survives, amended in four ways.

Amendments

1. Simulate everything; gate only the view

The world-model always runs at full fidelity — every rival lab, every doom source, computed every turn exactly as the admin-mode intuition imagines. SA purchases never add simulation; they only remove blindfolds from state that already exists. The discovered/undiscovered opponent mechanic already works this way; this generalizes it. Implementation is therefore UI gating plus per-source visibility flags, not a subsystem.

2. SA is not a meter; it is channels with provenance

Espionage-sight, alliance-sight, media-sight, research-sight — each reveals overlapping but non-identical slices of the same sim, each priced in a different currency, each with side effects (espionage corrodes governance and burns alliance standing; alliance-sight is slow, public, and obligates you via ledger entries; media-sight is cheap and noisy). Channel investment is a build-identity axis — the run's founding-location choice.

3. Lead-time semantics: the game must explain your death before it kills you

From Pip: "Ideally the late game will bring with it some forcing of understanding onto the player as to how they're going to lose (modulo their efforts to stop it?)" — so doom sources become legible for free as they become lethal. What SA purchases buy is not exclusive truth but lead time: everyone eventually sees the thing killing them; paying means seeing it while you can still act. This resolves the deduction-vs-triage tension: sight serves triage, not puzzle pleasure.

4. The acceptance test is decision-flip rate

ADR-0001's constraint ("every unit of sight must open a decision") was too weak for a triage game. Sharpened: the fraction of reveals that change the player's next action or reorder their priorities. A reveal that confirms the existing plan feels like a tax; a reveal that flips the plan feels like the game. Instrument playtests (log declared-intent-before-reveal vs action-after-reveal); kill any SA feature whose flip rate stays low. This is a falsifiable design criterion — measure it, don't eyeball it.

The honesty boundary (Papers, Please register)

Conspiracy-tinged weirdness lives in the provenance of information, never in the sim lying about itself: redacted documents, cranks who are occasionally right, a source who might be a plant — all on-theme, because real AI-race epistemics look like that. Instruments may be fallible (rival info ops are thematically perfect), but the game must be honest about how they can be fooled, and counter-intelligence is its own purchase. Rams #6 applied one level up: honest about dishonesty.

Beacons

As ADR-0001, plus: the lead-time rule protects against the one loss-feeling Pip did not choose ("I still don't know what killed me" — horror), while preserving the one he did (tragedy: "I saw it coming and couldn't stop it").