Skip to content

Fly pirate-attack-e1 against g3 and decide whether it ships #2

Description

@cgreening

pirate-attack-e1 is the first brain trained on the game engine after the
separate simulator was deleted. It took a 100% validation kill rate and is
deliberately not shipped.

That number is the reason for caution, not for confidence: it is the same
profile as generation 1 and generation 2, both of which won every measurement
in docs/TRAINING-LOG.md and lost the only one that counted, because the
optimal way to hold a firing line is to stop moving and evolution finds it
every time. CLAUDE.md's north star: lethality is a proxy for threat, and a
brain that wins every measurement can still be the wrong brain.

What to do: fly it. T at any station, same scenario, same seed, e1 as
the opposition, then pirate-attack-g3, and compare the two exported reports —
time on each other's six, engagement range, shots per engagement, whether it
makes attack runs or hangs and snipes.

state.brains.engine = true flies it in the live game
(__game.state.brains.engine = 'pro' from a console for professionals only).

Done when: there is a human-flown answer on the record, and e1 either
ships in brains.ts or the log says why not.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions