{"branches":[{"author":{"email":"did:web:projective-bichuwa.test.pds.did.bot","name":"did:web:projective-bichuwa.test.pds.did.bot","when":"2026-10-27T18:16:57.000000Z"},"committer":{"email":"noreply@tangled.sh","name":"Tangled","when":"2026-10-27T18:51:05.000000Z"},"message":"fix(harness): drop a stray conflict marker left in the audit doc\n\nThe previous commit's rebase conflict resolution missed one of the two\nmarkers, leaving a literal \"<<<<<<< HEAD\" line in the merged section.\n\nCo-Authored-By: Claude Sonnet 5 \nChange-Id: Icfb86599f0db301a82c328d37685dd77e42b2d25\n","name":"main","oid":"9e3c505fe5b3c4bf0cd421057ced6a86239c7d25"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-10-20T15:26:34.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-10-20T15:26:34.000000-04:00"},"message":"docs: show what our agents are doing in sds with a vibescrobble badge\n\nThe badge is kept to this repo, so it lists only lines written while working\nin sds.\n\nCo-Authored-By: Claude Opus 5.5 \nChange-Id: If036084e41d8ea276a375c4215e388d28ed72f15\n","name":"claude/vibescrobble-badge","oid":"424a79a92690f57ecf37622392b5b90099fc91e2"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:44:20.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T20:22:47.000000-04:00"},"message":"docs(latency): an epic for a unit-decision budget the worst case honours\n\nChange-Id: I748b6217060fb63ed45a433b524fe719687af428\n","name":"claude/epic-latency","oid":"d51f74b53ec6b2faad6928fa35c2f00814062b6a"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T20:17:44.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T20:18:26.000000-04:00"},"message":"docs(candidates): note that the four-verb generator flag is gone\n\nThe rollout item still named SDS_CANDIDATES, --candidates and their\nflag-off test as if current. sds-bot/src/candidates.rs's module doc\nsays all three were deleted with the four-verb generator they gated.\n\nChange-Id: I3d5ff9dafebabf1998b47383e96e8c37c4a735e8","name":"claude/doc-inaccuracies-fix","oid":"ca8bed9eec08942d2f6832c3da4ba7bb0be05d92"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T06:26:22.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T20:14:10.000000-04:00"},"message":"docs(training): the ablation run properly, and what absorbed the column\n\n+0.000106 of r-squared, agreement worse without it, and all 49 weights moved.\nRemoving expected_criticals flips value_destroyed from -0.0140 to +0.0147 - its\npresence is what inverts a feature with a declared expected sign.\n","name":"claude/refit-crits","oid":"7462a5e43fd8d5d1cb0fd991a3b937fa025413d8"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:43:45.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T20:05:34.000000-04:00"},"message":"docs(puzzles): an epic for recorded positions the bot must not get wrong\n\nChange-Id: I13ab48a52be5b6ec3fe2735234ee7ae520e2d3df\n","name":"claude/epic-puzzles","oid":"1974c93278704835d11b0475b081c98ef064ca3a"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:44:41.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T20:04:47.000000-04:00"},"message":"docs(browser): an epic for seating sds on lance.blue and then in the browser\n\nChange-Id: Ibe8af7332765a6423296959e7993a5c36cb764f9\n","name":"claude/epic-browser","oid":"8f676e914f2a08103cdee888f69cbc02b930cc19"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:52:21.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:52:47.000000-04:00"},"message":"docs(plan): remove references to nonexistent complete/ directory.\n\nThe plan/README.md documentation referenced [complete/](complete/) as if it\nexisted, but the directory has never been created. Removing all stale\nreferences to archival behavior that was never implemented.\n\nCo-Authored-By: Claude Haiku 4.5 \nChange-Id: I40de57713e431b3eab1b36b072b8f52ed29e317e","name":"claude/docs-fixes-v2","oid":"a06cf478e06ec92face95fed2c0b491311217bd9"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:46:44.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:48:02.000000-04:00"},"message":"docs(readme): clarify that matches are not guaranteed deterministic\n\nThe project goals claimed matches replay to the same outcome with the same seed, but the implementation notes say seeded dice \"does *not* make a match deterministic\" due to external factors in MegaMek. Clarify that the bot's decisions are reproducible, but match outcomes may vary.\n\nCo-Authored-By: Claude Haiku 4.5 \nChange-Id: I3501e87040e84e08911c796b695744e42b67b2ed","name":"claude/docs-determinism","oid":"6ecf514da2146540ddc2c06d6c8aa448bb672c8d"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:46:28.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:46:59.000000-04:00"},"message":"docs(readme): correct default bot in example command\n\nThe example commands incorrectly showed random_bot, which is not the default. The default bot is sds-bot, which runs unless --bot explicitly specifies otherwise.\n\nCo-Authored-By: Claude Haiku 4.5 \n","name":"claude/docs-default-bot","oid":"0c6db46c16a76ea307b19a9e95016794e9f5ebb9"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:46:13.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:46:13.000000-04:00"},"message":"docs(readme): remove stale WIP marker from docs section\n\nThe docs/ directory has been developed with comprehensive, complete documentation including FEATURES.md, HIERARCHY.md, PROTOCOL.md, and TACTICS.md. The WIP marker is no longer accurate.\n\nCo-Authored-By: Claude Haiku 4.5 \n","name":"claude/docs-wip-marker","oid":"4a0bd74b3185eac15cb0c942eb3b66af015f517b"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:29:25.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:29:47.000000-04:00"},"message":"docs(readme): correct default bot in example command\n\nThe \\`./sds.sh one\\` command defaults to sds-bot, not Princess vs Princess.\nUpdate the comment to reflect the current default behavior documented at\nline 241.\n\nCo-Authored-By: Claude Haiku 4.5 \nChange-Id: I01421f9670199ac3619e0597a6ab8d2cbe4926b5","name":"claude/doc-fixes-1","oid":"7d34d07da42a232a5ad6dc15434107b5cc785f8c"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:28:38.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:28:38.000000-04:00"},"message":"docs(melee): phaseName() already gives PHYSICAL its own tally\n\nThe epic said counting a physical under \"firing\" was still an open\nproblem. SdsClient.phaseName() has had a dedicated \"physical\" case\nsince 929dc24, and sds/melee.py already reads phases.physical.\n\nChange-Id: If27ac6c3243daad41c8d05d861804589dd80216a\n","name":"claude/sds20-stale-docs","oid":"da9a298c7461423755be250f30d7156e40f1eb68"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:21:43.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:22:06.000000-04:00"},"message":"docs(README): document the ineligible_bot reference bot\n\nThe documentation listed two reference bots, but ineligible_bot.py also exists in bots/ and is used for testing host-level refusal of illegal unit selections. Added it to the documented list.\n\nCo-Authored-By: Claude Haiku 4.5 \nChange-Id: Id7625dca625b15d069c6a0acdaf3d7d9b6707af7","name":"claude/docs-inaccuracies","oid":"d4b2c3969eeaa029d9efdb04ea80046431fdc06c"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:18:01.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:19:02.000000-04:00"},"message":"docs(performance): remove references to removed candidate cap\n\nUnits now report all scored candidates uncapped. References to \"20 candidates\" and \"capped-candidate invariant\" are outdated.\n\nCo-Authored-By: Claude Haiku 4.5 \nChange-Id: I091b0d27ad962b2f0b77f473774479a028d7683b","name":"claude/docs-fixes","oid":"f22decc589ec2b5e209667b04d4ad4c6a6b372b0"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:13:31.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:13:56.000000-04:00"},"message":"docs(training): reflect that units report uncapped candidates\n\nCorrected stale documentation that implied the 20-candidate cap was\nstill in place. Units now report every scored candidate.\n\nChange-Id: Ief146724bed46941d2e0554b514b8aabab97480b","name":"claude/stale-docs","oid":"2315eb768f6ed95bae71d376230b96975a40765b"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:06:12.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T19:06:12.000000-04:00"},"message":"docs(harness): stop pointing SeededRandom's javadoc at a command that isn't there\n\n`sds check-determinism` was named as the way to measure how far a seed\nfalls short of a deterministic match, but no such subcommand exists in\nsds/cli.py. Point at plan/harness.md instead of a phantom tool, and\ncheck off the tracked TODO for it.\n\nChange-Id: Ib3b5122831e2b7377c199b50926713ea458d91fa\n","name":"claude/doc-fixes","oid":"2cf9aec9c2b9871fe9059b223fc29a512717033e"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T13:09:20.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T13:09:20.000000-04:00"},"message":"fix(harness): main did not build, and its weights had two keys twice\n\n`attacker_self_damage` and `charge_damage` each appear twice in the hand set,\nwhich JSON tolerates silently; the third `Action::Move` in `force.rs` was\nwritten before `dfa_target` existed, so the test target did not compile. The\nfeature page had gone stale with them.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ibf60fbd867868ebe69c8918414136e0beb98f964\n","name":"claude/main-dfa-dup","oid":"368f96a2f2ce6f8084b9115d3fc6a300bf389a31"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T13:02:37.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T13:02:37.000000-04:00"},"message":"feat(estimator): the choice-model spike behind the epic's numbers\n\nPure Python, no dependencies: least squares on differences against a conditional logit, held out by match, on any stored corpus.\n\nChange-Id: Id6351eb3f3017ec8d7fd31b84e381e4029934728\n","name":"claude/epic-estimator","oid":"cab90a9376ac7ed4ec77ec84f6c6974bd6bc3d61"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:48:31.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:48:31.000000-04:00"},"message":"fix(harness): a movement row with no tactic is refused before the binary is asked\n\nretunes() spawned the release bot per sidecar before any row was read, so the fast test hook failed in every worktree without a release build. It is now asked when the first row that names a tactic needs it.\n\nChange-Id: I1010acc9ae33f3b7639b1c6eb9dec40783b80848\n","name":"claude/unit-split","oid":"678a57f423f62e700099f5876e7385c57a9813c5"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T07:45:09.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:41:32.000000-04:00"},"message":"docs(harness): CLAUDE.md carries the test-threads flag now, so tick it\n\nThe second box was open when the branch was written and closed while it waited.\n\nCo-Authored-By: Claude Opus 5 (1M context) \n","name":"claude/test-threads","oid":"9d4df201209e196abe50e36b13e3d0a31a4965cc"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T00:31:46.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:25:14.000000-04:00"},"message":"docs(c3): no network has ever been formed in a match here\n\nThe equipment is aboard and the link is not: a master and a slave sit in\ndifferent singleton nets with no master id, and `exposed_findC3Spotter` returns\nanother unit zero times. A `.mms` scenario cannot express a network - the format\ndocuments none, and the one apparent example is a pilot's name.\n\nForming them is three tasks rather than one. The host can form a net; a formed\nnet is worth 10% more battle value, measured, so every scenario fielding C3 has\nbeen scored wrong and force generation and balance are in scope; and only then\nthe overlay. A change that moves BV moves the measured strata twice.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I917d0e39303ba6ef521c6b58fbb491f8fd314382\n","name":"claude/c3","oid":"abdc436937809681899e9f75591b8bbb6e56ad3b"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T13:42:06.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:19:58.000000-04:00"},"message":"test(infantry): a Gauss and an LB 20-X in the resolved shots, and two platoons to choose between\n\nThe probe's Hunchback carried nothing the Firestarter did not, so it is an\nAtlas AS7-C instead: a Gauss Rifle and an LRM 20 join the 143 shots the server\nresolves and the port predicts.\n\nThe demonstration's two foot platoons become a foot platoon and a mechanised\nhover one. A Flamer is worth four times as much against the first, seven points\nof battle value apart, and no other column says so - which is what makes the\ncontrol with this one at nought differ.\n\nChange-Id: Ibdc0e196d0d6b467e37b40a08276481b006d09d3\n","name":"claude/infantry-firepower","oid":"2a783bb82c764d3a961d4771905cc9406eaf00f3"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T19:18:11.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:18:38.000000-04:00"},"message":"docs(harness): why a bounded column pools badly, with p_kill's bands\n\nChange-Id: If3ac5151c279901d29eab8849e97b726ddf9a66a\n","name":"claude/menu-spread","oid":"c7edb8fd8902385cfea2c1fb80541af079b1df96"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T14:32:28.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T12:18:36.000000-04:00"},"message":"feat(tactics)!: the precedence order is scarcest first\n\nOver twenty thousand random lances the old key-sorted order classified two\nformations as never and left Ranger at 0.8% of the 34.8% it admits. The pinned\norder leaves no formation at zero and raises Ranger to 8.1%.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I1aec8fb4953260641bc9dd201ca4e731fb3478ac\n","name":"claude/formation-band","oid":"245343c0c6ea5ea3c06242d6fa5bb9e3ac9a4e27"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-31T07:35:24.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T08:54:56.000000-04:00"},"message":"docs(harness): the stalling scenario fields artillery contrary to intent\n\nNoted, not changed: `vehicles` is already measured against this suite, so\naltering it would make a future run non-comparable. Removing the scenario is a\nbetween-rounds decision.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I0ea3778c48dc43d0b25ca22afa5ee0b8725b5427\n","name":"claude/princess-bench","oid":"7c0e641b2e58ae8b962d3afd9f2f7831eb57e27d"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T17:23:25.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T08:54:53.000000-04:00"},"message":"fix(harness): a candidate that does not carry a column is not a reading\n\nThe live-menu test counted a firing menu's abstain candidate, whose phi has no\ntarget features at all, as reading nought. Every target column then looked\nvaried on every firing menu, and target_gunnery reported 267 live menus with\n267 immovable - a tautology rather than a measurement. Counted over carriers\nonly, that corpus now says it cannot ask the question. The ablation arithmetic\nis untouched and still reproduces every recorded value.\n\nChange-Id: I83629f1c1190ca4271b49f03e79cbae63f0d99a8\n","name":"claude/live-carriers","oid":"e745fc7f4e090d56149136370cbc2fb37701cf36"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T11:05:47.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T08:54:50.000000-04:00"},"message":"feat(tactics): the lance table says who took each formation's lances\n\n`wins` went to the first formation that qualified, which credited a withheld\none with lances `classify` skips past. It now goes to what classify would\nanswer, and each row names the three formations that took the rest of it -\nwhich is what separates a formation precedence eats from one the criteria\nrarely reach.\n\nChange-Id: I339256e0c47851f7d8ad20f06c53544cf40cbbd7\n","name":"claude/formation-precedence","oid":"e4ade6f6822f07f2920847eb18ed3b7ff19a4f80"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T11:54:40.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T08:54:48.000000-04:00"},"message":"feat(c3): sds scenarios --c3 writes a networked suite\n\nNarrows the draw to forces that can form a net, since two C3i machines in a\ndraw is uncommon. The epic carries the arithmetic, where it was measured, and\na control: the same four scenarios generated unpriced overshoot their tier by\nup to 12.5% and the two sides diverge by up to 408 points.\n\nChange-Id: If424015139f22d585d252a217a9cb6c900eb2fa2\n","name":"claude/c3-pricing","oid":"fad8e16ed2bac69c8b43b49f6331b20d2019c33f"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T11:28:19.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T08:11:16.000000-04:00"},"message":"docs(c3): what the jar says, and why the network belongs in the scenario\n\nCorrects the claim that a scenario cannot express a network: `ScenarioV2` reads\na `c3:` block at `createGame`, before deployment, with no host code.\n\nRecords what forming one costs - 5% of the network's total, per member, before\nthe crew multiplier - and that the scenario generator balances on a design\nrating that cannot know about it.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ied78bf14570a8f0a28f1b33dd2defa765cc73a07\n","name":"claude/c3-overlay","oid":"7de955da36f6c8f6770bc7e98f5759559d937c71"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T18:01:59.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T07:47:41.000000-04:00"},"message":"fix(melee): a target that has not finished moving is not a target\n\nMegaMek's `MovePath` overload refuses one with \"Target must be done with\nmovement\"; the `Coords` overload the quote uses does not, so the server accepts\nthe declaration and drops the attacker on the floor for nothing. Carries what\nthe demonstration measured.\n\nChange-Id: Iaca3be93dcc7c393429f1aab82600bbb47bb0311\n","name":"claude/dfa-attack","oid":"9469274b08baf084513e8efd6f9b6aa8424a3a2c"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T01:01:27.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T07:39:28.000000-04:00"},"message":"feat(ev): a missile shot at an AMS carrier takes four off its cluster roll\n\nMegaMek's own figure - `bipush -4`, three times, in\n`MissileWeaponHandler.getAMSHitsMod` - applied where the target mounts one and\nthe shot is one that handler would resolve.\n\nApplied at both places a `ShotKey` is built. `volley::gather` scores the twist\noptions and `Volley::keys` feeds the location features, and a modifier on only\none of them moves half the basis with nothing failing; patching the first alone\nchanged no figure in a match, which is how the second was found.\n\nOn `ams-asymmetric.mms`, one missile boat against an Atlas that mounts an AMS\nand one that does not, in the same firing decision: value 3.864 against 3.178,\n`p_breach` 0.000320 against 0.000089, `target_breach` 0.3472 against 0.1157.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ie07ebea2453cd57c6f29ba7e5ce811b76dcd8425\n","name":"claude/ams","oid":"befbc796d4cfbfec14a98f6e36db114224a8094e"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T00:31:46.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T07:39:28.000000-04:00"},"message":"docs(c3): no network has ever been formed in a match here\n\nThe equipment is aboard and the link is not: a master and a slave sit in\ndifferent singleton nets with no master id, and `exposed_findC3Spotter` returns\nanother unit zero times. A `.mms` scenario cannot express a network - the format\ndocuments none, and the one apparent example is a pilot's name.\n\nForming them is three tasks rather than one. The host can form a net; a formed\nnet is worth 10% more battle value, measured, so every scenario fielding C3 has\nbeen scored wrong and force generation and balance are in scope; and only then\nthe overlay. A change that moves BV moves the measured strata twice.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I917d0e39303ba6ef521c6b58fbb491f8fd314382\n","name":"claude/c3-epic","oid":"7eafae67265e82052b2dbaed715abd8c5012bf3e"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T21:48:23.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T07:39:28.000000-04:00"},"message":"fix(artillery): an artillery cannon's direct hit is its rack size\n\nThe cannon reached the firing menu and was priced at nothing: MegaMek states\nartillery damage through the `DAMAGE_ARTILLERY` sentinel, the resolver answers\nzero rather than declining, and the round was then reported unmodelled, which\ntook the weapon out of the range table. `getRackSize` is the rules' direct-hit\nfigure - 5 for a Thumper, 10 for a Sniper, 20 for a Long Tom - and MegaMek will\nnot compute it: `calcDamagePerHit` hands the sentinel back unchanged where an\nAC/10 answers 10.\n\nTwo sites, because two independent exclusions each sufficed: the damage, and the\nname list that caught \"thumper\" the way the declaration guard's `instanceof`\nnever did. Cannons only - a tube lands a later round through an action type\nthat does not exist, and its sentinel stays clamped.\n\nOn `cannon-asymmetric.mms` the volley with the cannon went from 7.549 against\n7.565 without it, to 7.304 against 7.099, and four declarations now carry two\nattacks where every one carried one.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I08bd03bd08e43367efe28238a7dcf0ddb4fb8b5a\n","name":"claude/cannon-pricing","oid":"a3c8a43b30f4a9d2241166c408577bfec1414ea9"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T23:33:26.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T07:39:28.000000-04:00"},"message":"docs(features): a preferred mode must leave the to-hit alone\n\nThe bridge computes a shot in whatever mode the mount is settled in and reports\none shot; the volley then scales damage, heat and `ammo_spent` by the rate. That\nholds only because a rate of fire changes how many shots there are and never how\nhard they are to land, which is true of Ultra and Rotary and is now measured at\nthe site.\n\nA mode that moves the to-hit would be computed in one mode and declared in\nanother with nothing failing. `Indirect` is such a mode and is deliberately a\ndifferent mechanism: its own shot, its own to-hit, and the declaration carries\nthe mode that was scored.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I2d8048a50dde4fc51fb768a96c7bc6f8cd442b02\n","name":"claude/mode-invariant","oid":"b9d0f95c03ba26d7ea0e10dd677433445417b404"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T02:03:36.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-05T02:03:42.000000-04:00"},"message":"feat(training): bench the current-basis vector too, and it is another null\n\n40 of 66 decided against the hand set's 38 of 65, +2.1% with an interval of\n-14.3% to +18.4%. Three arms, three overlapping intervals.\n","name":"claude/price-the-unpriced","oid":"6988196fa872e58007cbf7de015210bf40f374fc"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T23:06:43.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T23:06:43.000000-04:00"},"message":"fix(harness): main did not compile after the swarm offer landed\n\n`hold the swarm` constructs `Action::Move` without `charge_target`, and the\ntest beside it patterns the same variant without it; both were written before\nthat field existed and applied cleanly onto a tree that had it. The\nreachability test reads `formation.opening`, which `Offers` replaced.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I08f1b5afedbff0309072fee6510a3866b459382b\n","name":"claude/main-charge-target","oid":"54b8f6e6b2ee3de9e0a15da5aeaa4d516475943f"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T18:30:07.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T22:41:55.000000-04:00"},"message":"test(harness): pin the recording, including the failure path\n\nThe offline half: the recording parses and still holds the facts the code\ndepends on, and the comparison reports a moved, gone and new fact rather than\npassing quietly. A check nobody has seen fail is not known to work.\n\nChange-Id: Id8f3bdc3e1017c27b3b70c9c9704089be42c4a3b\n","name":"claude/jar-conformance","oid":"3dcd08ea3c00d51970e0fa34f5b1befe797f3cb3"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T14:45:51.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T22:41:55.000000-04:00"},"message":"docs(melee): the push acceptance scenario, and what the probes settled\n\n`push-duel.mms` is a Javelin a side with both arms empty, so the push is on\nthe menu every contact turn rather than refused for an arm that fired.\n\nThe epic records what MegaMek answered: a trip is gated behind\n`tacops_trip_attack` and undeclarable through `PhysicalOption` either way, and\na push needs the target directly ahead of the attacker's feet rather than in\nthe forward arc a kick uses.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ibf6a3b2fcbc9f212331c5e9d1f740a9c0ff48df9\n","name":"claude/push-attack","oid":"013fb38621b9a56ae6c454bae22f4e6eac541afc"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T22:04:56.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T22:04:56.000000-04:00"},"message":"docs(harness): drop the counts of commands, bots and basis columns\n\nThree headings and sentences carried a number over a set the code owns: the\ncommand list, the reference bots in `bots/`, and the shape of the stance blend\nmatrix.\n\nChange-Id: I31ebf97cdc03c07d0d5606ff86343060f4c8cf6d\n","name":"claude/prose-restatements","oid":"e26b2cae7874089f5b69a622f390b990bd729ab5"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T18:48:12.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T18:48:12.000000-04:00"},"message":"docs(unit-types): the vehicle gap re-measured, and what is left of the column\n\n6.9x rather than 17x over 436,416 real firing candidates, and after the double\ncount the reading on a vehicle is identically `p_kill`.\n","name":"claude/lethal-union","oid":"383986b9bdf95e5f4b537fe3ea200ccfd99c5694"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T18:40:02.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T18:40:02.000000-04:00"},"message":"docs(harness): the compression item fired twice more before it was closed\n\nRecords the measured ratio, why compressing at the source would break the live\ntail, and that two of the three plain-only readers failed by reporting zero.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ib79773f1630d65445cbf87aed6cd3a666a2fba3d\n","name":"claude/compress-decision-logs","oid":"f11ef56e6a05e00a241d2c7145bf80ddaa7bfb59"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T16:17:39.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T16:17:39.000000-04:00"},"message":"docs(battle-armor): what refused the movement orders, and what it cost\n\nFive illegal orders in one match, all one squad that had swarmed. MegaMek's own\nanswer, the two readings it rules out, and the six-seed pair either side.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: If89cbcc799020abdcff91280a5e8a8408e17f490\n","name":"claude/ba-move-legality","oid":"9b917aca63b2703c656dda2174f9df199992f73a"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T16:14:49.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T16:17:26.000000-04:00"},"message":"perf(melee): do not quote a charge at a target no run reaches\n\n`chargesFor` asked `ChargeAttackAction.toHit` six hexes times two gaits for\nevery enemy on the board, per eligible mover, on every movement observation.\n\nThe bound is the mover's own run allowance plus the step into the target's\nhex, which is a fact about the machine rather than a rule: reaching a hex\nadjacent to a target that far away costs at least what it has, and terrain and\nturning only make it cost more.\n\nChange-Id: I085f25e49a1612dc1c2ec544baf1cde41ecd2121\n","name":"claude/charge-attack","oid":"36808e4c592e6d50e630958f098ad9d10cff2b1d"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T15:00:56.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T15:00:56.000000-04:00"},"message":"docs(battle-armor): why no squad would play, and what it was\n\nChange-Id: Ifa5d8ea857b9fe9317200096ebc7c3a23cc74bb8\n","name":"claude/ba-deploy-stall","oid":"ea9c72aeb9ae367d0c5b4f9aa4a2da1de5acd8d6"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T12:53:18.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T12:53:18.000000-04:00"},"message":"test(tactics): a tactic nothing can reach, found by recomputing the set\n\n`situation::orderable` is what decides whether a tactic can reach a force, and\ntwo other places state the same fact independently: a formation's `Opening` and\na tactic's `Status`. Both can disagree with it silently, and the force then\nplays `Engage` under another name.\n\nChange-Id: Ifd7a45688f623297d3c8f87567181e13bded0a5b\n","name":"claude/tactic-reachability","oid":"ea68f4559cdec58fc5c7b3b7bb6594b154837bcc"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T10:45:00.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T10:47:47.000000-04:00"},"message":"docs(tactics): what twenty thousand random lances came out as\n\nAnvil and Rifle, which uniform sampling reached zero times, are 5,284 and 781\nof 20,000 - so the corpus can produce the cases the formation work needs. Order\nis the one row of thirty-three it cannot: it wants four of the same model, and\nthe generator fills each slot independently.\n\nTwo findings for the precedence order: Battle and Berserker/Close have\nidentical satisfies counts at every seed, so Battle is never what a force is,\nand Ranger and Heavy Battle qualify in thousands and win in dozens. Row\nstability is measured against a second seed rather than computed, because\nlances are clustered by faction and year and a binomial reads too tight.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ie4bdb13a06ab54c2689e1154cc785625508b2cef","name":"claude/lance-corpus","oid":"4779adff803b47358797822625266c2e19bbd246"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T10:18:57.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T10:18:57.000000-04:00"},"message":"feat(tactics)!: a formation can offer nothing, and say why\n\nFive formations stay in the catalogue and are never handed to a force:\nAnvil, Rifle, Hammer and Order are gated on factions nothing here models, and\nAnti-Air has no behaviour to price. `Offers::Withheld` carries no `Opening`, so\nthere is nothing to turn into an assignment or fit a weight to.\n\nChange-Id: I156934d92b908a9236430389e2877a0a8fe02f07\n","name":"claude/formation-exclusions","oid":"1278d31dc0ed8dcca1bde7191cf723bf1c9ff872"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T07:28:09.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-04T07:28:09.000000-04:00"},"message":"test(ev): pin that a declaration puts the concentrated guns first\n\nMegaMek resolves attacks in arrival order - `handleAttacks` walks\n`getAttacksVector()` calling `handle`, with no sort or comparator - so the order\nthis bot declares is the order the shots land in, and a gun that opens a\nlocation resolving first means the cluster weapons behind it fire through the\nhole.\n\nThe ladder already orders that way. Its sort exists to choose which subsets to\noffer and descending damage-per-packet happens to be \"most concentrated first\",\nso the property was incidental; reordering the ladder for some other reason\nwould have handed the cluster weapons intact armour with no figure moving enough\nto notice. The test fails if the sort is reversed.\n\n`plan/ev.md` said nothing sorts it. Something does.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Iba05520d0be12392bd631a05d775316e36b44fe1\n","name":"claude/declaration-order","oid":"115775096477fb1fd7794950afc168ef4b3cf23b"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T20:50:17.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T20:50:17.000000-04:00"},"message":"docs(ev): three armour types reduce damage per weapon, not per location\n\nReflective, reactive and ballistic-reinforced key on the shot as well as the\ntarget, so `Location` cannot carry them. Recorded with what is established -\nthey are not applied in `damageEntity` - and what is not.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I41f46a4dc80222a9afcfa466dc993edb627c7abe\n","name":"claude/armour-type","oid":"b5c6872110b2e65d29a51f3b0d2cb715e5d48153"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T19:57:44.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T20:35:41.000000-04:00"},"message":"docs(infantry): a platoon is a headcount, not a body plan\n\nNo hit table, no armour, and strength carried as a trooper count, so the body\nplan catalogue does not extend here. The probe behind that is committed beside\nit.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I8c0bad065e53f5f65c77a180c76ada825ddcdd37\n","name":"claude/infantry","oid":"3a02d1c0ec541e9c3cbd6fc94d58f6c40a688c9f"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T18:56:52.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T18:56:52.000000-04:00"},"message":"feat(features): the movement corpus records a helicopter that flies\n\nEvery VTOL row was taken at elevation 0 - a landed machine MegaMek refuses a\nforward step in every clear-ground cell. The probe now flies the specimen and\nscores it against the flier model, and the block agrees on all 49 cells.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ia316bf1828b14d894d8c0bba73de849f65bcf97c\n","name":"claude/vtol-move","oid":"bd80397030a1a735516c5e3c590c3459d6352ad8"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T17:51:08.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T17:53:19.000000-04:00"},"message":"feat(features): a torso-mounted cockpit is a fact about a design, not a plan\n\n`getTransferLocation` answers LOC_NONE for such a head where a standard design\nanswers LOC_DESTROYED, so the head stops being lethal. One field read off\n`Location.cockpit`, and the three torso-mounted quads need no plan of their own.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I493962323e58f86f41180f6a14eab05f45a26bc3\n","name":"claude/vtol","oid":"6b1d8237b25d4679479c3fe842ca32640bfb7541"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-31T07:35:13.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T17:24:26.000000-04:00"},"message":"docs(artillery): artillery is three modes across two weapon families\n\n`ArtilleryCannonWeapon` selects direct or indirect as a weapon mode and needs no\nnew action type; `ArtilleryWeapon` has no modes and reaches only through\n`ArtilleryAttackAction`, which carries the flight time MegaMek computes itself.\nHoming loads need a TAG unit, which is a force-level dependency.\n\nCorrects the claim that `brackets::Excluded::Artillery` keeps artillery out: it\nholds for pricing, not for declaration.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: Ic667d1331ea27d28b5da49ffb1f74542bae2bd35\n","name":"claude/artillery-declaration","oid":"295575e213ccd82e9ffb5e4c63157d4bd4edb379"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T17:06:48.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T17:06:48.000000-04:00"},"message":"fix(artillery): an artillery piece is not a shot the bot can declare\n\n`SdsClient` builds a plain `WeaponAttackAction`, and MegaMek routes an\n`ArtilleryWeapon` to a handler that casts it to `ArtilleryAttackAction` on the\npacket pump thread, hanging the match with every counter clean. The firing menu\nleaves them out until the action type exists.\n\nThe class test rather than `F_ARTILLERY`: an artillery cannon is a sibling that\ndirect-fires through its own handler and does not crash.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: If8d8defcfd600b48551e71dfd59b7b2513baf643\n","name":"claude/artillery-containment","oid":"ce0acaa82e7ef37eb7b40675446b14f6fc1a23b1"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T22:01:42.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-09-03T16:30:17.000000-04:00"},"message":"fix(harness): a round report that cannot be formatted must not hang the seat\n\n`receiveReport` runs on MegaMek's packet thread, so a throwing formatter killed\nthe handler, the acknowledgement was never sent, and the game sat in `*_REPORT`\nuntil the stall timer fired. The seat now falls back to MegaMek's own report and\ncomplains once. Movement failures print their stack trace.\n\nCo-Authored-By: Claude Opus 5 (1M context) \nChange-Id: I6cf1beb942e4cf96369e7322ff537e503b2ee2bd\n","name":"claude/hang-fixes","oid":"336561e8018408b82bd300a541fa4bf61034d593"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T20:55:46.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T20:55:46.000000-04:00"},"message":"docs(candidates): the key collapse is worth 1.7-2.1x at the lookup, ~5% of the sweep\n\nTwo maps of the two sizes over the keys a real sweep produced, four million\nprobes each. A volley pays two lookups out of about 120ns, so halving one is\nnanoseconds - worth having, and not what closes the gap.\n\nChange-Id: I8fb745ff70bf82a877006015382ecd8140a2c7bf\n","name":"claude/los-prefill","oid":"234d89bf3833b454c766c33af1250d8e9586431c"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T10:44:10.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T19:44:50.000000-04:00"},"message":"feat(protocol): count what a match actually did, not what defaulted\n\njmm asked for a rejected count and a never-answered count. The first existed as\nrefusals with the right semantics; the second could not be computed, because\ndefaulted fired both when no reply came and when the bot deliberately passed -\nso a dead seat and an immobilised tank were one number.\n\nChange-Id: I170ce12016ce513b6fe5c993b308ba330a2439ce","name":"claude/movement-stack","oid":"adbd5df90f6f753fe339a1f33d582e90211758ef"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T14:55:07.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T19:04:22.000000-04:00"},"message":"feat(features): force_share_destroyed, and retire the invented scalar\n\n`value_destroyed` divides by the target's own battle value, so it says\nhow close a volley comes to killing that target and nothing about scale.\nThe new column divides the same expected battle value by the whole\nenemy force's, so it says how much a volley takes off the board.\n\nBoth, because neither can be had from the other: `Weights::score` is a\nlinear sum, and the absolute figure is a product of the two factors.\n`target_current_bv` is a fraction of a target's own undamaged self and\n`target_original_bv` is a rank, so no magnitude was available either.\n\n`Location::value` has no callers left and is deleted.\n\nCo-Authored-By: Claude Opus 5 \nChange-Id: I94b5b8fa6f59816959d4c8afd68ad55c50965da3\n","name":"claude/component-bv","oid":"10fd695f38ef5556888056c8423a14671d049fac"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T19:00:21.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T19:00:21.000000-04:00"},"message":"docs(harness): remove the git cherry entry, which does not survive its own test\n\nThe entry says a patch-based merge rewrites patch-ids. It does not: three\nlanded pulls have patch-ids byte-identical to their commits on main, and\na landed four-commit branch is reported correctly. No mechanism and no\nsurviving instance, so the rule goes rather than gaining a caveat.\n\nThe summary-table row stays. It makes a different claim that never\ndepended on the mechanism.\n\nCo-Authored-By: Claude Opus 5 \nChange-Id: Idc0912f3d1a3fd75e8bd9fcc9719a17d7eb65c59\n","name":"claude/cherry-audit","oid":"be9dd89243ef2212f082cf848349152548267e79"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T18:37:58.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T18:37:58.000000-04:00"},"message":"docs(shape): a feature's weight is a curve, not a number\n\nWeights::score is a dot product, so one number per column says more is\nproportionally better forever. Damage saturates, heat has a knee, a piloting\nroll has a cliff, and overkill already exists as a hand-built column for a shape\na weight could not express.\n\nBlocked on training, and the checklist puts extending basis_fingerprint to cover\nknot positions first: it is already blind to a changed computation, and two fits\nwith identical column names and different knots would be indistinguishable.\n\nChange-Id: Iec5f43479b5251fe9290ea1785b659b50edd1807\n","name":"claude/shape","oid":"5738442a67bf7875972d7a1f5e9b356ad54534c9"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:50:50.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T18:36:08.000000-04:00"},"message":"docs(unit-types): the ruling on pricing a location, and the differential that cannot be computed\n\nReplaces the three options and the recommendation with what was decided:\nMegaMek's per-component battle value, carried as `component_bv`.\n\nTrims the `destroyLocation` note to the fact that stops someone\nreimplementing it - outside a Game it marks a location doomed and stops,\nso no criticals, no MP recalculation, and no usable differential.\n\nCo-Authored-By: Claude Opus 5 \nChange-Id: I6651c7bf500ff158bd984262454b03b8a2023934\n","name":"claude/vehicle-basis","oid":"6c01796567ee9dc8bb1e239221b73d431835bb2a"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T16:28:59.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T16:28:59.000000-04:00"},"message":"feat(melee)!: attacker_self_damage\n\nRenames the column to sit with the basis's other attacker_* readings, so a\nreader scanning the vector sees it is a property of the shooter.\n\nChange-Id: Ia4704fea057df87553f065ff4df6d880d74912ef\n","name":"claude/self-harm","oid":"2404e405fbad6967e274e28c88cb0f3aeeaf9d9d"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T08:49:32.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T16:10:58.000000-04:00"},"message":"docs(harness): what the unbounded test run cost, and why clippy keeps -j 2\n\nA concurrent benchmark went 66-74s a match to 113 and back to 44-55. And the\nclippy hook is correct as it stands - clippy is a compiler, with no second\nbinary downstream for -j to miss.\n\nChange-Id: I448bf2a778237bdb1daf16bf81ff8d96db5a0292","name":"claude/perf-docs","oid":"72cc0cb89cd55315a3b463532cf41fa736688d20"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T13:20:58.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T13:20:58.000000-04:00"},"message":"feat(melee): the bot can swing the hatchet it is carrying\n\nThe capability was already on the wire and the shot was not. Observation's unit\nblock enumerated getClubs() with its retracted-blade guard, so the movement\nprojection could price a hex for a hatchet; physicalsAt looped the fixed four\nand nothing else, so the physical phase never offered one. A Hatchetman could\nwant to be next to something for a reason it was then unable to act on.\n\nsds club-probe answers the two things worth measuring rather than assuming: a\nclub's damage scales with the attacker's tonnage, exactly tonnage over five and\nso the kick's own figure, and getClubs returns an empty list rather than null.\nThe probe set separates the two - three hatchets across the tonnage range, and\nan Atlas at the Berserker's own hundred tons carrying none.\n\nA hatchet is a strictly better kick: same damage, a flat 4 to hit, and no\npiloting roll for missing. Same seed on hatchet-duel.mms takes the bot from six\nkicks and no hatchets to three hatchets, four kicks and a punch. Princess was\nalready using them.\n\nChange-Id: Ia3c1812c5621a1715065fcf57cf6725674f993f0\n","name":"claude/clubs","oid":"8827a85c7eec5ad9e6628b06ca4890ce649e3fb0"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T10:18:58.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:54:13.000000-04:00"},"message":"fix(unit-types): a vehicle has no heat scale, and now says so\n\n`getHeatCapacity` answers `Entity.DOES_NOT_TRACK_HEAT` for a combat vehicle -\na sentinel whose value is 999 - and the bot divided by it. Every heat figure\ncame out near zero and correct, which is why nothing ever failed.\n\nNo figure moves: `heat + heat_buildup + extra` never approaches 999, so the\nsubtraction already floored to zero, and the test computes the old route\nalongside the new. The comparison is made in Observation.java where MegaMek's\nown constant is in scope, so nothing on our side writes 999. `Unit::heat_load`\nhad no callers and is deleted.\n\nCo-Authored-By: Claude Opus 5 \nChange-Id: I3a76e4e2bbabb91932dc148163514f8d4a0ec823\n","name":"claude/vehicle-sentinels","oid":"0e49db9899d19c63265bfbde08e269e500d89ee2"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:38:34.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:38:34.000000-04:00"},"message":"fix(features): one denominator for value_destroyed, not one per phase\n\nThe firing phase divided by the profile - the locations `hitloc` names -\nand the movement phase divided by every location the observation carried.\nIdentical on a Mek, where both are the same eight, and different on\nanything else, so one column meant two things according to which phase\nasked.\n\n`LocationProfile::value_at_stake` is now the single implementation and\nboth call it. The movement path computes it once per enemy instead of\nonce per candidate, which it can because a location's value takes no\nfacing.\n\nCo-Authored-By: Claude Opus 5 \nChange-Id: I15652f634314baa6540b49e0aa9d2f175c5db926\n","name":"claude/at-stake-denominator","oid":"bc93bb17ee20b0211ac6a76cd4dd27d78bc39973"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:29:44.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:29:44.000000-04:00"},"message":"docs(unit-types): destroyLocation on a game-less entity, a negative result\n\nPricing a location by the battle value its loss costs looks like it works\ngame-less: an Atlas yields eight well-ordered differentials and no error.\nIt does not. `destroyLocation` marks a location ARMOR_DOOMED and stops -\nisLocationBad stays false, no criticals are destroyed, and MP does not\nchange - so the movement term never enters and a leg prices below an arm.\n\nAlso records the detector both of tonight's omissions would have failed:\nan omitted term shows up as a ranking a player would call wrong, not as a\nzero or an exception.\n\nCo-Authored-By: Claude Opus 5 \nChange-Id: I179612332627845ccd6d8589271b31566b03441b\n","name":"claude/bv-doomed-negative","oid":"7e986908127a5447420e3610151314274216f75d"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:00:18.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T12:00:18.000000-04:00"},"message":"docs(harness): a clean rebase is a statement about text\n\nNineteen commits onto a main that had moved twenty: no conflicts and two compile\nfailures, mirrored - a field addition against a new literal, a signature change\nagainst new call sites. Neither has textual overlap, so git cannot flag either.\n\nChange-Id: I41b4782010118106f6b03e80ee1300b02257c0d2\n","name":"claude/replace-is-not-rename","oid":"03a58e198f6d659c883503c299d7c7fb6622909e"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T03:03:11.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T10:24:08.000000-04:00"},"message":"docs(unit-types): the calibration measured, and not confirmed\n\n+3.84 points against a registered \"under 10, would not be surprised by\n3\", at z = +1.06 and per scenario up 7 down 4 same 7. Three further\nlimits named. Kept in its own file so the branch below keeps its own\ncomparison.\n\nCo-Authored-By: Claude Opus 5 (1M context) \n","name":"claude/vehicle-motive","oid":"91d85a24ac66ff6744650fcfc803a08d8d00d621"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T09:02:15.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T10:23:52.000000-04:00"},"message":"feat(unit-types)!: expected_criticals is optional, and a vehicle has none\n\nA vehicle is described by a hit table and still has no critical model:\nMegaMek resolves its criticals as unit-level results and it reports\nsystemSlots=0, so the Mek 2d6 table does not apply. `None` says that,\nthe feature imputes zero at the boundary, and no figure the fit sees\nmoves. Replaces the pinning test left for whoever gave vehicles a table.\n\nCo-Authored-By: Claude Opus 5 (1M context) \n","name":"claude/vehicles","oid":"0e3c827d62ec87bd1337603b45619e56c946ac14"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T08:41:03.000000-04:00"},"committer":{"email":"noreply@tangled.sh","name":"Tangled","when":"2026-08-30T14:20:04.000000Z"},"message":"feat(protocol): a refused move is answered again, not overruled\n\nThe host refused a move, stood the unit still and told the bot nothing, so the\nnext ask was the identical question and got the identical answer - one unit sent\nthe same four steps for thirty-seven rounds. The observation now carries what\nwas refused and the bot re-decides with it excluded, bounded by a count.\n","name":"claude/move-profiles","oid":"0ffc6fcb55c8d3dc6ab63727cbc9f5c7b38665f6"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T09:35:32.000000-04:00"},"committer":{"email":"noreply@tangled.sh","name":"Tangled","when":"2026-08-30T14:19:36.000000Z"},"message":"test(candidates): record the corpus with the turn cost the wire now carries\n\nEvery specimen was parked when its unit line was written, and Observation skips\nturnMp for a unit with no position - so a squad that turns for nothing was\nrecorded paying a Mek's hexside for as long as the field has existed. Placed at\nelevation zero, which keeps the clear-ground frame stands is measured in.\n\nlance-advance.svg is regenerated rather than churn: rebasing onto #287-on-main\nbrought battle armor's free turn in, infantry turning free changes the reachable\nsets the vignette draws, and the committed SVG is an artefact of behaviour that\nlegitimately changed. It is evidence the fix propagated exactly as far as it\nshould have.\n","name":"claude/move-legality","oid":"f3467a84eab74ae0ab371a525559338f06286aa2"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T10:01:54.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T10:01:54.000000-04:00"},"message":"fix(candidates): expected_criticals through the mean fast path\n\nMain added the channel to Prospect while this was in review, so mean_collapse\nneeds a seventh accumulator on both the gain and the loss path. Added to the\nfixtures and to the bit comparison too, so the new channel is exercised rather\nthan merely present.\n","name":"claude/rank-cost","oid":"fdfe9dd6f49b522273b7fdf8dc89b541abcc039c"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T01:40:21.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T09:56:31.000000-04:00"},"message":"perf(candidates): collapse a stand without a Vec per column\n\nCherry-picked from #294, with expected_criticals threaded through Collapse::of\n- main added that channel while the PR was in review, and a channel missing\nthere is one the ranking silently stops weighting.\n","name":"claude/stands-bookkeeping","oid":"16173bcca5ecfcdbc587bad3b4f2243172edc5a6"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T09:47:43.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T09:47:43.000000-04:00"},"message":"docs(features): a file with a regeneration command is not a file to edit\n\nPromotes the tell to the heading it deserves. It is the whole rule, it\ncosts nothing, and five things carried a stale answer tonight of which\nonly one was code.\n\nCo-Authored-By: Claude Opus 5 (1M context) \n","name":"claude/target-described","oid":"fe93f0869393e068647a5b7e5344fe27339cfe71"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T08:41:03.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T09:38:44.000000-04:00"},"message":"feat(protocol): a refused move is answered again, not overruled\n\nThe host refused a move, stood the unit still and told the bot nothing, so the\nnext ask was the identical question and got the identical answer - one unit sent\nthe same four steps for thirty-seven rounds. The observation now carries what\nwas refused and the bot re-decides with it excluded, bounded by a count.\n","name":"claude/move-retry","oid":"1b0d70143874ae8dd32c1a75fdc73d07351531a2"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T08:48:01.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T08:48:01.000000-04:00"},"message":"docs(training): the asymmetric corpus was replaced, not reconstructed\n\nBERSERKER supersedes it with a different population at 24 matches rather than\n330. A destroyed measurement is not always worth rebuilding: what a second\ncorpus is for is disagreeing with the first.\n\nChange-Id: I165c54fd73399422d70d62a366a10cf80475c154\n","name":"claude/corpus-rescue","oid":"bc6517ea978394957bde749a9ca184ccaca7c26c"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T08:38:54.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T08:39:31.000000-04:00"},"message":"docs(unit-types): (c) for a vehicle, and what the zero sweep could not find\n\nCorrects the recommendation. (a) is ruled out by my own sweep, which lists\n`expected_criticals` at 0.000 for a tank as one of three sites where zero is an\nabsence; recommending it would have been recommending the catalogued defect.\n\nThe mechanism for (c) is two parts and one alone is cosmetic. `Option` on\nthe producer changes no figure the fit sees, because a feature must yield a\nnumber and imputing zero is what happens today. The distinction has to reach the\nvector: a `target_described` indicator, built for the family rather than for one\ncolumn, since `value_destroyed` and `at_stake` have no gate to make optional.\n\nAlso records why the sweep found 3 of 5: the predicate was \"reads zero\", which\nis the costume. The shape is a value arrived at without the thing it claims to\nmeasure, and it can err upward or be accidentally right.\n\nChange-Id: I383875378eba1e78086f0cab65e1f6dc16e5c61f\n","name":"claude/vehicle-criticals-mechanism","oid":"75fb4567f62598231aa60d3f25802ab6a664fb1b"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T03:20:05.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T03:20:05.000000-04:00"},"message":"docs(candidates): what the half() and store chain is made of, and what came off\n\nThe 10.4% is a chain rather than a copy: `Outcome::of` on a memo miss, 1.4\nmillion weapon-twist arc tests, 620k pushes and 3,288 shrinking reallocations.\nTwo of those came off, for 3 to 4 points measured against the current code in\ntwo quiet-box runs.\n","name":"claude/stands-half-narrow","oid":"1be2d399436ac1247c4d3c8795360dcb49c27cb1"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T02:54:21.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T02:54:21.000000-04:00"},"message":"fix(training): a matching basis is not evidence two corpora are comparable\n\nThe fingerprint hashes a feature's name, normalisation and description, so it\nagrees across a change to what a feature computes. The commit that would have\ntold the two apart was already in the manifest and nothing read it. check() now\nnames both commits whenever the fingerprints agree - a note, not a refusal,\nbecause every corpus older than the working tree trips it and a hard error would\nbe switched off within a day.\n\nsds tactics validated nothing at all, which is the reader that produces the\nagreement matrices and so the one comparing two corpora. It now prints the\ncorpus identity and the review before the numbers.\n\nChange-Id: I1cbf559e480ae910c14e6a58c2c3576143866bdd\n","name":"claude/advance-tactic","oid":"aa4730ef3ab79927da7d4a001dfa81146cfdf61f"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T02:19:02.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T02:19:02.000000-04:00"},"message":"docs(candidates): where the sweep spends its time, by ablation\n\nThe volley call is 84.7% of `score_stands` and splits 58 points of first-time\nwork against ~27 of memo lookup, so the four changes that attacked the lookup\nwere attacking the smaller half. `count()` and the range fold are free, and the\nmarginal and cumulative readings of bookkeeping disagree by eleven points for a\nreason worth writing down.\n","name":"claude/stands-decompose","oid":"3a405f2fa59c2141cb0bd9a355e264031cb13708"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T02:17:11.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-30T02:17:11.000000-04:00"},"message":"test(candidates): decompose the sweep by replayed ablation\n\n`score_stands_ablate` no-ops one step of the sweep at a time and `perf\n--ablate` runs it beside an unablated sweep in the same process. Steps that\ndecide something the loop branches on are replayed from a recorded sweep\nrather than no-oped to a constant, so an ablation removes work instead of\nchanging the workload, and the sweep's own counters are asserted identical\neither way.\n","name":"claude/stands-ablate-harness","oid":"3e6df9eb93a63bb57c1b09b1bea10f6e4b1f3ec4"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T01:14:59.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T23:58:33.000000-04:00"},"message":"perf(candidates): mix the fast key instead of SipHashing it\n\nFastKey is one packed u128 and SipHash is more than a memo lookup needs. The\nobvious replacement was worse: los::BuildFx measured 15% slower, because the key\npacks its high-entropy fields at fixed bit positions and a weak mixer clusters\non exactly that, trading cheap hashing for long probes. SplitMix64's finaliser\nis 5-9% faster over 20 pairs with a control at 1.006.\n\nChange-Id: I22214b0243e599b0e335a02807caeca2f119bd31\n","name":"claude/fastkey-mixer","oid":"3efef97711bb59f0a9a6845a441bbe6ae613302f"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T23:51:19.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T23:51:19.000000-04:00"},"message":"feat(harness): kill your own match containers, not everyone's\n\n`sds clean` kills every sds-* container on the machine. This kills only the\nones whose /work mount is the caller's worktree, and names the owner of each\none it leaves.\n\nChange-Id: I98c1f212604af4a01e70e6ca6fd6e3097ad308ba\n","name":"claude/kill-my-containers","oid":"cf9908e23bdeb7322ff42b31fe74fa94d72106d2"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T23:34:23.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T23:34:23.000000-04:00"},"message":"fix(features): rebase onto main, and restore what the merge dropped\n\n`Excluded::for_weapon` came through the rebase without the `anti_mek` check:\nboth branches changed that function from opposite sides, and taking one\nwholesale compiles and passes both suites while losing the other's work. That is\nthe hazard this branch documented at the site, and it landed exactly as\ndescribed. The order is now round, flag, handler, name, and\n`a_swarm_attack_is_refused_by_the_flag_not_the_name` makes losing it a test\nfailure rather than a silence.\n\nThree more the merge would have swallowed quietly: `Observation` keeps both the\nswarm-damage substitution and `packetsPerLocation`, which sit in the same block;\n`cmd_formation_dump` lost its tail to a shared `return code`, caught by ruff;\nand melee's new `Shot` needed a grouping, which is one - a fist lands in one\nplace.\n\nChange-Id: Iacd1679e0d0980cee1da10cf74577f959c91988b\n","name":"claude/expected-crits","oid":"b9e9a43390a78c10b82d3e5c9b76a5eb694597d5"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T22:00:38.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T22:00:38.000000-04:00"},"message":"docs(harness): -j 2 never bounded a test run\n\n`-j` limits compilation jobs; libtest then parallelises across every core, so\nthe rule was silently half-effective. One test binary was measured at 710% CPU\nwhile a benchmark was recording.\n","name":"claude/test-threads-doc","oid":"381e6f2d58c84bb1f7a1637b9b91aae2c0bb236e"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T06:09:27.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T06:09:27.000000-04:00"},"message":"docs(tactics): the firing phase measures the basis, not the tactics\n\nFive of seven tactics report \"never diverges\" on a firing menu because it\ncarries no positional feature. Recorded so the measurement is not repeated in\nthe belief it might say something.\n","name":"claude/tactics-precision","oid":"af5c096f595f245221733a20c9a48dfb2d40456e"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T03:10:12.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T03:10:12.000000-04:00"},"message":"test(battle-armor): pin the expected_damage seam melee opened\n\nprojected_shots puts a fist in the volley and ceiling counts only guns,\nso the fraction is clamped and a lightly-gunned machine saturates on\nevery contact hex - which is where Advance lives. Measured, not fixed.\n\nChange-Id: I280d51b8c76f41b9374a089baca3659a683ad939\n","name":"claude/battle-armor","oid":"0e2173c34e521f9c193fea28cdb117f25352fec3"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T02:48:34.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T02:50:37.000000-04:00"},"message":"feat(tactics): load a fitted tactic weighting, and scope what is left\n\n`--tactic-weights `, in the same envelope `--weights` uses and read as\nstrictly: an unknown name is refused rather than skipped, because a file the\nbot shrugged off would have a benchmark measuring the hand-authored set under\nthe fitted one's name. The epic gains the two ways to connect the fitter with\nwhat each costs, and the label question written up as jmm's to answer.\n\nChange-Id: I56089aa2590fc2bb27086b418de893a94d0842e1","name":"claude/formation-tactic","oid":"50ff95fdb0c4d628ff72d44a38d2612a3f4e94ee"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T00:43:01.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-29T00:43:01.000000-04:00"},"message":"feat(harness): report both statistics and flag their disagreement\n\nThe raw and the normalised readings are printed side by side, and a gap of more\nthan five points is called out: two readings of the same seven samples came out\nat 7% and 24%, and the gap was the only thing saying the sample was too small.\nThe end-to-end tests now drive a fake binary under a moving machine factor, so\nthe refusal is checked rather than assumed.\n","name":"claude/ab-harness","oid":"1c4c1e6ed1e5f102320b433955c2ea7b4d68f6cf"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:54:51.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:54:51.000000-04:00"},"message":"docs(los): why a region test cannot beat one bit per pair\n\nThree quarters of the sweep's pairs end at no line, and the cheap test that\nwould reject them is dominated by an exact bitmap it costs the same to build.\nRecords the blocker census, the height table that makes a single table unsound,\nand the ~8% ceiling the whole idea is worth.\n\nCo-Authored-By: Claude Opus 5 (1M context) \n","name":"claude/los-regions-finding","oid":"68c301652139f444cd2504488be4c96c01662ce7"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:47:44.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:47:44.000000-04:00"},"message":"docs(candidates): the line memo did not pay, and why\n\nThe sweep asks the same hex pair 8.5 times over, so memoising the line on a\npacked integer key looked free. Against an unchanged control it measured\nnothing: the memo runs to megabytes and a lookup that misses cache costs about\nwhat it saved. Reverted, and recorded so nobody tries it twice.\n","name":"claude/line-memo-negative","oid":"374ddc09221c909575d7a8e4159460bed02b598a"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:41:41.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:41:41.000000-04:00"},"message":"docs(protocol): a weapon is its properties, never its name\n\nEvery property on the wire is one the bot handles, and an unknown weapon\nworks completely. When a rule is not in the fields, the fix is to put it\non the wire rather than to branch on a name.\n","name":"claude/weapons-are-properties","oid":"2a1395a731e00f41876679babe1c4937ab5364ad"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:07:50.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:07:50.000000-04:00"},"message":"docs(harness): the control for the seat split, 19-18\n\nPrincess against itself on the seat-per-process harness comes out 51.4% to\n48.6% with overlapping intervals, so the split did not bias it - which matters\nmost because each seat now draws from its own dice stream. 39 victories and one\nstall in 40; the 3-of-40 \"undecided\" the report prints is one stall and two\nvictories with no winner it can attribute.\n","name":"claude/control-seat-jvms","oid":"865f3df949dd941bf5b027220bb57bba143e24c0"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:02:52.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T21:04:31.000000-04:00"},"message":"perf(candidates): answer a volley without building its key\n\ngather runs before the memo because it is what builds the key, so a 99% hit\nstill walked every weapon at every twist, built a 150-byte LocationProfile and\nsorted an allocation. FastKey packs what gather reads into one u128 and answers\nfirst; the slow path is unchanged and is what the agreement test compares\nagainst. 32-39% off the sweep, with unchanged gather as the control at 1.01.\n","name":"claude/volley-prekey","oid":"c053fd9381455d3926c69bb2db0d036998d49d7e"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:52:01.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:52:01.000000-04:00"},"message":"feat(candidates)!: backing up is the default, and its rules are measured\n\nMegaMek accepts every route the search emits - 37,166 of 37,166 through\nisMoveLegal - and the corpus is re-recorded with BACKWARDS in the\nenumerator's step set. A Mek may not run backwards, may turn while\nbacking, and may not change elevation, which is level minus depth.\n","name":"claude/backing-up","oid":"233f09a27451c581d5de366526eac59b9e709b12"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:15:34.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:15:34.000000-04:00"},"message":"docs(candidates): three quarters of pairs never reach the weapon walk\n\nThe sweep benchmark counts them: 75% of pairs end at NoLine, 10% at\nNothingBears, 15% produce a volley. The pre-key cannot skip the line of sight\nask, so its win is concentrated in the quarter that gets past it rather than\nspread over the sweep.\n","name":"claude/volley-measures","oid":"526e2b36adaa03430c8d3bbf2168da341e311184"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:13:39.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:13:39.000000-04:00"},"message":"docs(los): correct the end-to-end figure to what the mechanism predicts\n\nA first measurement said 14% off examples/stands. Taken again over twenty-five\nalternating runs it is 5-7%, which is what 18% of the sweep losing 64% of its\ncost has to give. The 213ns to 76ns per get stands; the whole-binary number had\nfound a lucky window on a shared machine.\n","name":"claude/los-cost","oid":"e1b4f2464f0813c543f0e403a674b9c5131e2f09"},{"author":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:10:10.000000-04:00"},"committer":{"email":"did:plc:nlzmjyfv6loqtxyzvdcznwgf","name":"@permadeath.com","when":"2026-08-28T20:13:03.000000-04:00"},"message":"perf(candidates): time the benchmarks on CPU, not on the wall\n\nWall clock on this machine measures the other agents: the same binary\ntimed at 59.5ms and 129.7ms an hour apart. `/proc/self/schedstat` counts\nwhat the process was given; batches grow past the millisecond tick before\nthey are divided. `docs/PERFORMANCE.md` gets the sweep's figures and a\nsection saying which clock produced them.\n","name":"claude/perf-sweep-bench","oid":"14d7aa3a56d3b774ebd21e9266caf752faa0c0f4"}],"cursor":"100","defaultBranch":"main","total":120}