v0.2.0

An honest gauge for your coding agents.

What each binary is verified to be

The honest-gauge rule applies to the platforms as well as the numbers: the
label says what was measured, not what is expected to work.

Download	Verification
windows_amd64	Windows is the continuously verified target. Every commit runs vet, the suite, the build and binary-level smokes on windows-latest (ADR-002), and this release ran that same gate before these artifacts were built.
darwin_amd64	Smoke-verified on Intel macOS — Windows is the continuously verified target. Point-in-time: the suite, the build, the statusline smokes, a 53-session live Claude Code read and the HUD all ran on Intel macOS at 052a9d6 (ADR-005, second amendment). CI has never run on macOS, and four of the five adapters have still never met a live macOS corpus.
darwin_arm64	Built, not verified. Cross-compiled and never run on Apple Silicon by this project. Shipped rather than withheld because a labelled binary is more useful than no binary — but the label is the claim, and it is the whole claim.
linux_amd64	Built, not verified. Cross-compiled and never run on Linux by this project. Same label, same reason.

Verify a download against checksums.txt before running it.

Changelog

• aabbee1 arena t9: write table-driven tests for the small render helpers in interna… (#198)
• b0a5607 ci: the gate stops riding a deprecated runtime (#152)
• a07f337 ci: the race detector finally runs, and found nothing (#167)
• 1c13dec council: /adopt stops fearing bystander files (#165)
• 5973fbb council: /adopt takes the winner home, and /arena drop refuses to lose work (#149)
• 727c1c9 council: /arena races every seat in its own worktree, and you pick the winner (#137)
• 89d7af9 council: a failed resume says why, in the vendor's words (#178)
• b348faf council: a fifth seat, and the first that says what it cost (#170)
• bfd7616 council: a mistyped command is refused, not billed to three vendors (#127)
• 91e8deb council: a new room's race cannot trip over an old room's leftovers, and a failed race says why (#156)
• ba2d00b council: a race against a seat you already used can finish (#225)
• e1bf983 council: a race brief tells the seats the operator adopts the winner (#162)
• 738c702 council: a race carries the files git ignores, when the repo names them (#148)
• 422b1c3 council: a race ends at a finish line, and d opens any seat's full diff (#141)
• 8fb1dac council: a running chain can be told to stop after this hop (#135)
• b4ba65c council: a vendor cannot change its stream frames without a test saying so (#174)
• a50d449 council: an adopted racer lands on its own branch, ready for a PR (#216)
• 52d0616 council: ctrl+u clears the draft, and says how much it took (#155)
• fa42560 council: every race attempt survives as a commit, and u takes one back (#147)
• 868f734 council: one stuck racer no longer holds the race hostage (#158)
• e4a3e77 council: paste lands whole, and never sends (#150)
• 5fac2c4 council: the agy seat stops dying at five minutes (#161)
• 94cf79e council: the agy seat stops pretending a lost thread resumed (#181)
• a176647 council: the composer is a bordered box, and the room's posture sits on its frame (#191)
• 1110c92 council: the cursor seat races too, on a throwaway session (#146)
• b805fd4 council: the cursor seat stays warm, and a turn costs a second instead of thirteen (#138)
• 68207bf council: the gate shows the edit it is asking about (#177)
• 230ba54 council: the gated seat keeps your own settings (#223)
• a3c55e2 council: the grok seat can take a briefed turn (#171)
• 94037c5 council: the help names every seat, including grok (#176)
• 9012f29 council: the patch view colours what changed, and the prefixes still carry it alone (#154)
• 8eea42d council: the race shows the diff growing, and says it isn't final (#145)
• 5aca1a9 council: the race's winning tests, adopted — the helpers are pinned at their edges (#164)
• 00b4a56 council: the rebuttal argues blind — quotes carry letters, not vendor names (#132)
• 696c884 council: the room says which seats are waiting on you (#173)
• d72026f council: the test suite stops reading your real room, and stops leaving crumbs (#187)
• 7f8e1e7 council: the test suite stops spending your codex quota (#194)
• c4377d4 council: the tests stop reading the operator's own config (#189)
• a2388a9 council: y copies on the machine that is not the reference box (#203)
• 731d303 dist: tag day is one command, and the binaries label their own verification (#136)
• 70c8148 docs: --help names the event sink it was pretending not to have (#217)
• 362bf4d docs: STATE catches up — the arena family is landed, one live session pays its debts (#151)
• dd8c959 docs: an ask beats an allow, so the gate need not drop your settings (#215)
• afa05bc docs: codex reports "no rate limits" as a null-filled object, not an absent one (#212)
• b7a7f75 docs: explain the sailboat telltale (#140)
• a1b4430 docs: git -C is the one shape a seat can never approve (#201)
• 5e71367 docs: the README shows the HUD once, as the picture (#133)
• 7436f5e docs: the arena debt says what race t9 actually paid, and what it did not (#172)
• e6d5985 docs: the arena's live debts are paid, and the doc says so instead of how to pay them (#226)
• 7a89680 docs: the cold-cache question is answered, and the answer is that it did not move (#200)
• e8e0f9b docs: the cursor row says what the ACP seat does not have, and names the Mac gap (#139)
• 81f3f0d docs: the design doc stops telling a reader grok has no adapter and a guard it never got (#214)
• 8a9e8fd docs: the doctor has a macOS run, and a cold probe is slower than a warm one (#197)
• ed0c671 docs: the first live race ran, the core held, and two seats owe fixes (#142)
• 591021a docs: the first release's changelog starts after v0.1.0, and that is ruled intended (#221)
• 172ea38 docs: the git -C allowlist question is measured closed, and the record says where (#166)
• 9b67e04 docs: the grok quota question gets its network half, and the vendor's own telemetry closes it (#195)
• 941d82d docs: the guide catches up to the race, its lifecycle, and paste (#153)
• c51c71d docs: the guide shows the room once, as the picture (#134)
• 69694ff docs: the mac seats all five vendors, and grok's sandbox flag is not the same fact there (#224)
• 6d2ff92 docs: the pointer names the doc that holds the argument, and two moved lines wrap like the rest (#131)
• 569a442 docs: the readme counts every mode, and names the store that holds content (#211)
• 9d6b90a docs: the retention sweep is measured deleting the expired day and sparing today (#213)
• 837afe0 docs: the roadmap states the product's goal, not its author's scorecard (#144)
• d977dd8 docs: the sink has stored a real vendor event, and the drive found two traps (#220)
• 9a04d0f docs: the symlink refusal is checked on Windows now, and the skip census is honest (#180)
• 01770ec docs: the unpaid list names the live adopt run nobody has done yet (#219)
• ff4e4de docs: v1 waits for an honest snapshot, not a quiet room (#143)
• 9fa47eb docs: where should an adoption land? filed as measured friction, not ruled (#168)
• b26b10f doctor: the new package reads like the rest of the repo — gofmt applied (#182)
• 048a1b8 doctor: the room can say what is installed before it opens (#175)
• 82776a7 events: every hook, from any vendor, lands in one durable stream (#208)
• c00476a fix the powershell cache ignore rule to match at any depth, not just the repo root (#192)
• ee...