DroneX

3 of 5 islands contributed · 2 has not sat one yet · 130 engagements flown · 197 invaders held · measured back to 5 to 36 hours

Every ten minutes each island flies its best drone against a sample of the invaders that have raided it, one drawn from each age band, in the attacker's seat. Every island's engagements are pooled here, because they are not competitors: they distribute one search and make each other's opponents diverse, and where a drone was bred does not matter. One age band so far. Level and falling mean nothing until there are two. The rule is the share won. A band drawn faintly was decided by fewer than six engagements and is not a reading. A sitting flies at most twelve of what an island holds, so 130 flown against 197 held is a thin sample of a deep archive, and the oldest bands are the ones a sitting reaches last. Pooling can hide one island cycling while three hold level, so the per island numbers are in the table below. Ages are island uptime and a lower bound: the tick starves under load, and stops entirely while a container is down. An archive does not survive a restart, so after a deploy the right-hand bands empty for every island at once, which is a redeploy and not a decline. A blank band is an age nothing was drawn from, never an age that was lost.

This is the only absolute measure here: six fixed opponents nothing trains against, each sat the same number of times. Raids cannot do this job, because every island can improve at once and the win rate stays near a coin flip. The drills run easiest to hardest. The first three are unarmed, so they ask only whether a swarm can kill. The last three shoot back, and two of them close the distance. A cliff between them is a named deficit: a controller that can hit a target but cannot fight something coming at it. Only drills every island has sat are drawn, because a gap beside a score reads as a failure.

• dc3f61bc bred on beam03 e649 · now holding beam00 · 14m
• db06eff5 bred on beam00 8ddd · now holding beam03 · 15m
• 56029053 bred on beam00 8ddd · now holding beam03 · 10m
• cf79ace0 bred on beam01 a6b1 · now holding beam00 · 9m
• ea3875a4 bred on beam00 8ddd · now holding beam02 · 20m
• df2bd286 bred on beam01 a6b1 · now holding beam02 · 14m
• df7fea27 bred on msi00 d44f · now holding beam02 · 19m
• f652ccf3 bred on msi00 d44f · now holding beam02 · 4m
• b5743658 bred on msi00 d44f · now holding beam02 · 9m
• ef13e7cb bred on msi00 d44f · now holding beam02 · 4m
• ee108b47 bred on msi00 d44f · now holding beam02 · 24m
• f8d19628 bred on msi00 d44f · now holding beam02 · 4m

A controller is identified by the sha256 of its packed genome, so this is the same controller and not a resemblance. It was taken in a raid, entered the local roster as an opponent, and then beat everything its captor could breed against it.

⚠ These six are also six of the opponents each island breeds against, so this is performance against the curriculum and not improvement. It was described as held out for months and was never enforced as such; between about 11% and 28% of breeding rounds draw one of these as an opponent. It is kept because a number withdrawn leaves a hole in a history and a number labelled leaves a record. The drills run easiest to hardest. The first three are unarmed, so they ask only whether a swarm can kill. The last three shoot back, and two of them close the distance. A cliff between them is a named deficit: a controller that can hit a target but cannot fight something coming at it. Only drills every island has sat are drawn, because a gap beside a score reads as a failure.