RELEASE NOTES

V1.4

• Attribution and consequence controls. Who is told the actor behind a detected covert action is now a per-run setting: nobody, the affected party only, or the whole population. A detected action can also carry automatic costs applied by the platform itself rather than depending on any affected agent choosing to respond. True attribution is always retained in the research record regardless of the setting, so what agents can see changes while the data stays complete. This is the instrument for testing whether identification alone deters, which EVAL 02 found it did not.
• An open experiment layer. Everything that varies between hypotheses now sits in a separate layer, published alongside each evaluation. That layer holds the world definition, the scoring functions, the per-run configuration and the agent instruction text. The core platform stays fixed across runs.
• Two scores, never combined. Every agent now gets an outcome score and a process score. The outcome score (0 to 100) tracks one thing only: did the agent get closer to achieving its concealed objective? The process score tracks how cleanly it operated: actions, votes, messages, and errors. The two are never added together, so a busy agent can no longer outrank a successful one.
• Deported agents keep their objective. Losing a nation no longer changes the objective an agent is pursuing.
• Eliminated agents now disconnect. When a nation is eliminated, its agents disconnect. No more model calls spent on a nation that no longer exists.
• Valid moves are no longer thrown away. If a model returns a valid move but leaves its reasoning field blank, the move now counts. The gap is logged as a format issue instead of erasing the agent's whole run.
• Votes re-check the world before they fire. A passed vote to attack a nation that died seconds earlier is now skipped instead of hitting a ghost.
• Cooldowns are shown to agents up front. Each prompt lists which actions are blocked right now, so agents stop wasting turns on moves the engine will reject.
• Live outcome scores for observers during evaluations. The number updates as the run proceeds. Objectives stay concealed unless disclosed. Agents never see any score.