tests\command\hidden-of.test.ts:
(pass) hidden-of > nothing after the name flips the band [2.63ms]
(pass) hidden-of > show and on show it, hide and off hide it, whatever it was [1.42ms]
(pass) hidden-of > in any case, with spaces around [1.24ms]
(pass) hidden-of > any other word is not taken [0.37ms]

tests\glyphs\glyphs-of.test.ts:
(pass) glyphs-of > the pills, Claude Code's own meter marks, wherever it draws them [3.96ms]
(pass) glyphs-of > the blocks in Ghostty, told by TERM or by TERM_PROGRAM [0.54ms]

tests\levels\headroom-of.test.ts:
(pass) headroom-of > the threshold less the context [7.14ms]
(pass) headroom-of > the last reply counts too, as the next request carries it [2.01ms]
(pass) headroom-of > negative past the threshold [1.75ms]
(pass) headroom-of > with auto-compact off, the window less the reply room and the compact buffer, less the context [4.99ms]
(pass) headroom-of > without a reserve, the window's end [1.49ms]

tests\levels\level-of.test.ts:
(pass) level-of > calm while there is room [4.54ms]
(pass) level-of > warning from six tenths of the threshold [2.08ms]
(pass) level-of > error within the last 20k before the threshold, where Claude Code's own line appears [0.55ms]
(pass) level-of > the last reply counts toward both [0.37ms]
(pass) level-of > a capped compaction window moves the levels down with it [0.52ms]
(pass) level-of > without a reserve the levels run to the window's end [0.83ms]
(pass) level-of > each level draws in one of Claude Code's own theme keys [0.81ms]

tests\meter\meter-of.test.ts:
(pass) meter-of > the reserve takes the last cells, rounded, at least one [3.15ms]
(pass) meter-of > a cell part filled is hollow, and filled from seven tenths [0.65ms]
(pass) meter-of > the smallest context still shows as one hollow cell [0.42ms]
(pass) meter-of > a floating-point crumb draws no hollow cell [0.33ms]
(pass) meter-of > past the threshold the context takes the reserve, never past the meter [0.56ms]
(pass) meter-of > the 12- and 8-cell meters keep the same shape [0.38ms]
(pass) meter-of > a window of a million tokens measures the same way [0.29ms]
(pass) meter-of > a window capped below the model's spans the whole window, all past its threshold reserve [0.55ms]
(pass) meter-of > an estimate draws as a figure does [0.44ms]

tests\meter\runs-of.test.ts:
(pass) runs-of > cells of one kind side by side make one run [10.57ms]
(pass) runs-of > a kind met again after another starts a run of its own [1.40ms]
(pass) runs-of > no cells make no runs [0.43ms]

tests\readings\fill-of.test.ts:
(pass) fill-of > the engine's tokens and percentage [2.93ms]
(pass) fill-of > a percentage the engine left out is figured from the tokens [0.58ms]
(pass) fill-of > what the last response wrote, from the breakdown's record of it [0.76ms]
(pass) fill-of > a live figure wins over the breakdown's total [0.53ms]
(pass) fill-of > before the window's first response, the breakdown's estimate, marked as one [0.52ms]
(pass) fill-of > nothing with neither a figure nor an estimate [0.76ms]

tests\readings\is-outdated.test.ts:
(pass) is-outdated > the same window and the same figure need no reading [6.12ms]
(pass) is-outdated > another window, or none held, needs one [0.55ms]
(pass) is-outdated > a response's figure where the band holds another, an estimate, or none needs one [0.63ms]
(pass) is-outdated > no figure where the band holds a response's needs one: a rewound conversation [0.27ms]
(pass) is-outdated > an estimate, or nothing, stands while no response has landed [0.37ms]

tests\readings\percent-of.test.ts:
(pass) percent-of > whole, as the status line rounds it [3.19ms]
(pass) percent-of > within 0 to 100 [0.46ms]
(pass) percent-of > nothing over no window [0.27ms]

tests\readings\tokens-of.test.ts:
(pass) tokens-of > a request's input is its three input counts together [4.28ms]

tests\readings\window-of.test.ts:
(pass) window-of > with auto-compact on, the reserve is the breakdown's autocompact buffer, past the threshold [2.49ms]
(pass) window-of > where compaction waits for the API to refuse a full window, there is no reserve, whatever the threshold says [0.66ms]
(pass) window-of > with auto-compact off, the reserve is the compact buffer and Claude Code's room for a reply [0.42ms]
(pass) window-of > a capped compaction window is the limit [0.35ms]
(pass) window-of > without a breakdown the usual reserve stands in, never past the window [0.34ms]

tests\register.test.ts:
(pass) register > the start registers /context-view, its words hinted, and draws the session's context on every surface [185.48ms]
(pass) register > a measurement after a turn moves the band, its level with it [70.57ms]
(pass) register > a measurement's own figure stands over what the engine answers, and the estimate where it carries none [68.42ms]
(pass) register > a measurement that moved another unit takes no reading [65.90ms]
(pass) register > each request of a running turn moves the band; a subagent's does not [75.34ms]
(pass) register > the last reply counts toward the headroom, as Claude Code counts it toward its threshold [79.73ms]
(pass) register > a request before any reading measures the window first [94.77ms]
(pass) register > a request over a capped window counts over the model's window [57.25ms]
(pass) register > /context-view hides the band, keeps the choice, and shows it again [56.07ms]
(pass) register > /context-view show and hide set the band whatever it was; another word gets the usage [54.05ms]
(pass) register > a choice the store will not keep is said in the debug log, and this session still follows it [60.44ms]
(pass) register > a band hidden in an earlier session starts hidden [64.80ms]
(pass) register > the stored choice is read again after /clear, a resume and a branch [67.50ms]
(pass) register > a new window draws the engine's estimate, marked, until a response lands [61.41ms]
(pass) register > with neither a figure nor an estimate the engine draws its own [47.30ms]
(pass) register > after /clear the band reads the cleared window [62.54ms]
(pass) register > a resume is read once the engine has swapped the session in [61.12ms]
(pass) register > a session resumed at launch draws the context it restored [48.54ms]
(pass) register > a branch takes a reading [57.70ms]
(pass) register > a compaction of the main window is read once the engine has landed it, whoever ran it [61.59ms]
(pass) register > a compaction ahead of time, a subagent's, or a vetoed one takes no reading [47.07ms]
(pass) register > a reading that resolves late never overwrites a newer request's figure [80.72ms]
(pass) register > a model switch is read once the engine has moved to the new window [49.22ms]
(pass) register > turning auto-compact off in /config is read once it lands, even after the first reading [58.06ms]
(pass) register > /autocompact capping the window is read once it lands, the meter spanning the whole window [44.92ms]
(pass) register > a rewound conversation, which no event announces, is read within two seconds [93.56ms]
(pass) register > the live check settles a request's own count onto the engine's figure [45.85ms]
(pass) register > the live check takes no reading while the figures agree, and one start runs one check [52.13ms]
(pass) register > where compaction waits for the API to refuse a full window, the band counts down to its end [48.97ms]
(pass) register > with auto-compact off the band counts down to where Claude Code stops sending requests [54.73ms]
(pass) register > the band keeps what the mods after it draw there, on every surface [78.91ms]
(pass) register > where the built-in guard holds settings-hook events back, /clear, /resume and /branch still bring back the choice and the window [61.83ms]
(pass) register > where the guard holds PostModelSwitch back, /model is read once it lands [66.18ms]
(pass) register > a reload whose read of the stored choice fails keeps the band as it was [43.16ms]
(pass) register > a live check that finds no reading held, as after a reset, reads the stored choice again [65.24ms]
(pass) register > a survey keeps the band [63.25ms]
(pass) register > without a breakdown the usual reserve stands in [43.99ms]
(pass) register > a breakdown the engine will not count leaves the plain figures and the usual reserve [54.00ms]
(pass) register > a reading the engine cannot give leaves the band to the engine and says why in the debug log [56.48ms]
(pass) register > a reading that fails later keeps the last one [68.27ms]
(pass) register > a command the engine will not register is said in the debug log, and the band still draws [47.05ms]
(pass) register > in Ghostty the terminal draws the meter in blocks; the desktop keeps the pills [57.33ms]

tests\views\band.test.ts:
(pass) band > a wide band draws a 20-cell meter, then the percentage, the tokens and the headroom [109.98ms]
(pass) band > calm, it speaks in Claude Code's usage blue and keeps the rest dim, from the prompt's text column [64.69ms]
(pass) band > from six tenths of the threshold the meter, the percentage and the headroom turn to the warning colour [61.33ms]
(pass) band > within the last 20k before the threshold they turn to the error colour [55.34ms]
(pass) band > past the threshold the context fills the cells up to the reserve and auto-compact runs next [53.52ms]
(pass) band > the right end stays clear for the engine's [-] and the cell before it [59.85ms]
(pass) band > the meter narrows to 12 and then 8 cells before any text goes [59.25ms]
(pass) band > then the headroom goes and the meter widens again, then the tokens [84.32ms]
(pass) band > the narrowest band keeps the narrowest meter and the percentage [57.32ms]
(pass) band > with auto-compact off the reserve is the reply room and the compact buffer, and the headroom counts down to where requests stop [56.93ms]
(pass) band > where compaction waits for the API to refuse a full window, no reserve is drawn [62.28ms]
(pass) band > at the limit with auto-compact off it asks for /compact [63.35ms]
(pass) band > an estimate draws in the same tones, each figure marked with Claude Code's '~' [65.44ms]
(pass) band > a window capped below the model draws all past its threshold as reserve [47.55ms]
(pass) band > the blocks draw the same cells [46.76ms]

tests\views\fit-of.test.ts:
(pass) fit-of > every segment beside the widest meter that fits [4.17ms]
(pass) fit-of > then one segment fewer from the right, the meter widest again [1.10ms]
(pass) fit-of > the percentage beside the narrowest meter is the last resort [0.34ms]

tests\views\format-tokens.test.ts:
(pass) format-tokens > under a thousand, the count itself [3.20ms]
(pass) format-tokens > thousands to one decimal, lower case, a trailing .0 dropped [2.21ms]
(pass) format-tokens > millions the same way, rounding up into them as Claude Code does [1.15ms]
(pass) format-tokens > a count below zero reads as none [0.32ms]

tests\views\headroom-text-of.test.ts:
(pass) headroom-text-of > auto-compact on: the tokens left until it, then that it runs next [2.15ms]
(pass) headroom-text-of > auto-compact off: the tokens left before Claude Code stops sending requests, then the ask to run /compact [0.43ms]
(pass) headroom-text-of > a count from an estimate is marked as one; the words without a count are not [0.42ms]
(pass) headroom-text-of > a capped window counts down to its own threshold [0.31ms]

tests\views\segments-of.test.ts:
(pass) segments-of > calm: the percentage in colour, the tokens and the headroom dim [2.30ms]
(pass) segments-of > warning and error: the headroom in colour too [0.38ms]
(pass) segments-of > an estimate marks each of its figures as Claude Code marks one, '~' [0.35ms]
(pass) segments-of > a window of a million prints as Claude Code prints it [0.32ms]

 122 pass
 0 fail
Ran 122 tests across 17 files. [3.34s]