ADR-013: Deliberative Baseline — Training the DTA on Your Higher Self¶

Status: Accepted (philosophical foundation) Date: 2026-03-21 Context: A fundamental question about the validity of game-based DTA training: when a player sits down to play Sync, they're in a calm, reflective state — not hungry, not panicked, not sleep-deprived, not emotionally activated. Decisions made in this state may differ from decisions made under acute stress, fatigue, or emotional arousal. Does the DTA represent the "real" person, or just the considered version of them?

At a glance

What it decides: The DTA is trained on your deliberative, considered self — the version of you that has time to think — and that is treated as a feature, not a limitation, because that is the state most governance decisions are actually made in.

Deliberative baseline by design — the game has no countdown, forces written rationale, and runs free of hunger, fatigue, or emotional flooding, so it captures considered values rather than impaired snap reactions.
State-tracking rejected — self-reported mood is unreliable and adds friction; the speed round (0.4× weight, ADR-009) is the only deliberately small window into reflexive tendencies.
Consistency is the validation — repeated choices across sessions and moods are the persistent signal; EMA smoothing (ADR-004) filters the noise.
Watch: if Beacon ever expands into real-time agent actions (trading, negotiation), the deliberative-only assumption may need revisiting.

flowchart LR
    A["In-the-moment self<br/>(hungry, stressed, tired)"] -. "not trained on" .-> D["DTA"]
    B["Deliberative self<br/>(time to think, writes rationale)"] == "trained on" ==> D
    D --> G["Governance proxy<br/>(your 'higher self')"]

The DTA learns the considered self, not the activated one — the version best suited to govern.

Decision¶

The DTA is intentionally trained on deliberative decision-making. This is a feature, not a limitation.

When you play Sync, you're: - Sitting at a computer, not under time pressure (the game has no countdown on choices) - Reflecting on a scenario, not reacting to a crisis in real-time - Articulating rationale, which forces conscious reasoning - Not influenced by hunger, fatigue, social pressure, or emotional flooding

The DTA learns this version of you — the person who thinks things through. This is the version that should be making governance decisions.

Rationale¶

Most of life is lived in a deliberative state¶

On average, people are not in an activated state. The majority of decisions — including governance decisions in a DAO — are made from a baseline of relative calm: reading proposals, weighing options, voting. Acute stress states (panic, rage, desperation) are the exception, not the rule. Training the DTA on baseline decision-making means training it on the state you're most likely to be in when governance decisions actually need to be made.

Activated states produce worse decisions¶

Decision science consistently shows that cognitive performance degrades under acute stress. Hunger narrows risk tolerance. Sleep deprivation impairs judgment. Emotional flooding reduces consideration of alternatives. If the DTA learned from these states, it would learn your impaired decision patterns, not your considered ones. A DTA that represents your panicked self would make worse governance decisions than one that represents your reflective self.

The DTA as the "higher self"¶

This framing reframes the DTA's role: it's not trying to simulate what you would do in every possible state — it's capturing your best decision-making self. The version of you that has time to think, considers multiple perspectives, and articulates why. When the DTA acts on your behalf in governance, it acts as this considered version, even if the real you at that moment is stressed, distracted, or emotionally compromised.

This is actually an advantage of delegation. The DTA doesn't have bad days. It doesn't vote differently because it skipped lunch. It applies your considered values and patterns consistently, which is exactly what you'd want from a governance proxy.

Consistency across sessions is the validation¶

If a player makes the same types of choices across multiple game sessions — played at different times of day, different days of the week, different moods — the pattern that emerges IS their deliberative baseline. Session-to-session variance is noise; the persistent signal is who they are when they think clearly. EMA smoothing (ADR-004) naturally filters for this persistent signal.

Alternatives Considered¶

Add mood/state tracking to each game session: Ask players "How are you feeling right now?" before each game. Weight decisions differently based on self-reported state. Rejected because: self-reported mood is unreliable, it adds friction to the game experience, and the goal is to capture the deliberative baseline, not to model state-dependent behavior.
Include time-pressured scenarios to capture reactive decisions: Add a countdown timer to force quick, gut-reaction choices. Rejected because: the speed round already captures faster, more reflexive responses (weighted at 0.4x per ADR-009), and the core scenarios are intentionally unpressured to capture considered values.
Acknowledge the limitation and don't claim full representation: Frame the DTA as "your governance proxy" rather than "your digital twin." Considered and partially adopted — the DTA IS a governance proxy, and the deliberative training is why it's suited for that role. But the "digital twin" framing remains useful for the broader Beacon vision.

Consequences¶

The DTA will not predict what you'd do in a panic. It predicts what you'd do when you have time to think. For governance delegation, this is the right behavior.
Players who are highly variable across emotional states (impulsive when stressed, cautious when calm) will have a DTA that represents their calm self, which may feel "incomplete." But the calm self is the one that should be voting.
The speed round (0.4x weight) provides a small window into more reflexive tendencies, but it's deliberately underweighted because snap judgments are less diagnostic for governance.
If Beacon eventually expands beyond governance into real-time agent actions (trading, negotiation), the deliberative training assumption may need revisiting — those contexts may require a DTA that models reactive behavior.

Open Questions¶

Should the game ever surface this explicitly to players? ("You're training your DTA on your best decision-making self.") This could increase trust and set accurate expectations.
If a player's game-session choices diverge significantly from their real-world decisions (reported after the fact), is that a failure of the DTA or a success? The DTA captured who they want to be, not who they were under pressure.
Does this create a "values gap" where the DTA votes based on a player's ideals while the player in the moment would have voted differently? If so, is the DTA more legitimate than the player's in-the-moment self?

How We Got Here¶

This question emerged from an earlier conversation about the ethics of DTA training. The concern: people make different decisions when hungry, worried, panicked, or relaxed. If the game only captures the relaxed-and-reflective version, is the DTA truly representative?

The resolution came from recognizing that on average, people are not in an activated state. The deliberative baseline IS the representative state for most decision-making, and especially for governance. The DTA isn't a perfect simulation of every possible version of you — it's a governance proxy trained on your considered values. That's actually a better delegate than the version of you that voted while angry or exhausted.

This connects to ADR-011 (delegation map): the graduated sovereignty model means the DTA only acts autonomously in contexts where it has high confidence. If your behavior is genuinely different under stress (a state the DTA hasn't seen), those decisions would naturally fall outside the delegation map's confidence thresholds and get flagged for human review.