sds by lance.blue · Tangled

bayes for days

Commits on claude/shape

docs(shape): a feature's weight is a curve, not a

committed 1mo ago

committed 1mo ago

docs: cut the write-ups to what changes

committed 1mo ago

docs(melee): regenerate after distance_to_objective

committed 1mo ago

docs(melee)!: explodable components are not

committed 1mo ago

docs(melee): what the bot cannot see in the physical phase,

committed 1mo ago

docs(harness): check the floor before choosing the sample

committed 1mo ago

docs(melee): two rounds of self_harm, and what they

committed 1mo ago

feat(harness): freeze the weights file, and log the piloting

committed 1mo ago

docs(harness): a running measurement owns its inputs, or it does not own its

committed 1mo ago

fix(melee)!: self_harm is a probability, not a share of

committed 1mo ago

docs(melee): two questions about a piloting roll, two

committed 1mo ago

docs(melee): how prediction 3 gets measured, before the

committed 1mo ago

docs(melee): the self_harm prediction, before any

committed 1mo ago

feat(melee)!: price what an action costs us, not only what it costs

committed 1mo ago

docs(harness): what the unbounded test run cost, and why clippy keeps -j

committed 1mo ago

docs(harness): "when the disk fills" happened, and nothing

committed 1mo ago

docs(harness): the opening distribution closes the effect size, at

committed 1mo ago

docs(harness): the opening effect size is bounded by agreement already

committed 1mo ago

docs(harness): lead the self-discordance section with the curve, not the

committed 1mo ago

docs(harness): the width probe's 24 runs, and what they

committed 1mo ago

docs(harness): what the self-discordance result does to a tactic-layer

committed 1mo ago

docs(harness): scratch does not go in /tmp, and the reason is not

committed 1mo ago

docs(harness): whether a discordant pair carries any

committed 1mo ago

docs(harness): a victory does not always name a

committed 1mo ago

docs(harness): the two readings the seed evidence

committed 1mo ago

docs(harness): what --seed reaches, and why a match still

committed 1mo ago

docs(candidates): record why per-enemy sweep reuse does not

committed 1mo ago

fix(unit-types): a vehicle has no heat scale, and now says

committed 1mo ago

feat(melee): the bot can swing the hatchet it is

committed 1mo ago

docs(harness): a clean rebase is a statement about

committed 1mo ago

docs(harness): two costs that are only visible across

committed 1mo ago

fix(harness): kill test binaries by executable, not by command

committed 1mo ago

docs(harness): a search-and-replace across a rename is not a

committed 1mo ago

docs(candidates): the fast memo is 17.7 MiB and misses cache every

committed 1mo ago

feat(scoring): distance_to_objective, the one distance a caller

committed 1mo ago

perf(candidates): benchmark the path the sweep actually

committed 1mo ago

docs(los): the cache outlives the thing that invalidates

committed 1mo ago

docs(candidates): settle whether deployment needs the

committed 1mo ago

perf(candidates): collapse a sweep without drawing the

committed 1mo ago

docs(candidates): mark what the merged stack settled, and date what it did

committed 1mo ago

docs(unit-types): destroyLocation on a game-less entity, a negative

committed 1mo ago

fix(features): one denominator for value_destroyed, not one per

committed 1mo ago

fix(candidates): the mean collapse carries the criticals

committed 1mo ago

perf(candidates): collapse at the mean without a powf a

committed 1mo ago

perf(candidates): collapse a stand without a Vec per

committed 1mo ago

fix(candidates): fold arms_flipped into the loadout

committed 1mo ago

perf(candidates): hold a unit's volley cache between

committed 1mo ago

perf(candidates): let the level-one volley memo outlive the

committed 1mo ago

docs(candidates): what the half() and store chain is made of, and what came

committed 1mo ago

perf(candidates): size the exchange column up front and stop shrinking

committed 1mo ago

perf(candidates): answer the arc prune per facing, not per

committed 1mo ago

fix(candidates): the hex-pair fixtures build against main's

committed 1mo ago

test(candidates): measure the collapse at the size the sweep actually

committed 1mo ago

perf(candidates): answer a volley once per hex pair and

committed 1mo ago

refactor(candidates): let a caller resolve a side from a bearing it already

committed 1mo ago

fix(candidates): the sweep fixtures deserialise their board and

committed 1mo ago

perf(candidates): split the volley memo into what fires and what it

committed 1mo ago

base: line table (#284) and fast-key mixer (#293), in review

committed 1mo ago

docs(candidates): a memo does not survive a round

committed 1mo ago