Changes touching this path

• loot grep: content search across history via the key oracle (#391) e47d1412 · dbf3dbe6…
• a verb run from a subdirectory finds the repo instead of advising an init that would nest a second one, path arguments rebase from the cwd the way a tester types them while view's globs stay root-relative, and the resolved root is named on stderr whenever cwd is not it eca43819 · dbf3dbe6…diff
• the valued half of the argv door gets a floor: FlagSpec grows value() and message() so a valued flag's payload is stepped over by the spec that declared it rather than found by a bare scan of raw argv, and all 29 readers move onto it along with six more the ticket's list had missed, which is what left loot ferry --dock --git-dir yielding a --git-dir nobody typed while its --json twin was the single case anyone had pinned; and the census meant to stop this class recurring widens from the previous bug's exact spelling to the class itself and gains a positive control, which caught two further hand-rolled scans in help_sub and resolve_tool_path and took the crate from eight offences to zero (#1652) f86ffb25 · dbf3dbe6…diff
• loot grep grows the options that make a search usable, and two of the ticket's acceptance criteria turned out to ask for something this verb already did unconditionally: -n wants line numbers, which loot grep has printed since #391 with no flag at all, so making them conditional would silently change the output of every invocation ever typed and -n ships instead as a pinned compatibility spelling of the default, exactly as -F now spells out the fixed-string matcher that stays the default with the regex behind -E. the same compatibility argument decides the row shape, so the change column is printed only when the selector was a REVSET, a thing this verb could not search before, while a point selector keeps the three-column row it always had. the pathspec is the declared option group and not a fourth hand-rolled one, which cost the leaf its NO_PATHS declaration first: a group's trailing list begins where the leaf's own arity ends, so grep declares two opaque positionals and attaches PATHSPEC, and because its selector is optional a bare trailing path still reads as the selector and the operator says which they meant with the separator. -E is a hand-written POSIX ERE over bytes rather than the regex crate, which is loot-cli's standing dependency-light no-clap stance applied rather than a new decision; that stance is argued in the ctrlc note in the workspace manifest, not by adr 0041, which governs loot-core and the server-side dependencies and does not reach this; and the cost is measured rather than guessed, by resolving regex alone and diffing its closure against this lock: three new entries, regex and regex-syntax and aho-corasick, since memchr and regex-automata are already in the graph through bstr from gix, though regex-automata is linked there with default features off and carries no dependencies at all in this lock, so regex would compile far more of it than anything does today, all to serve one optional flag on one read verb. AC7 is a real requirement and it is keyed on the content address rather than the path, since repo.put mints a fresh address per call and only a carried tree entry reuses one: measured, a two-change revset over three paths each opens four objects where the natural walk opens six and returns byte-identical rows, which is why the pin counts ObjectGets instead of reading the answer. AC6's aggregation needed a fixture the engine refuses to build the obvious way, because a path this identity cannot open cannot be re-captured while it sits on disk, and the count is per change and path, so one sealed path carried through two changes is two skips. the per-object read is loot-core's own readable_tree classification lifted out rather than copied, so grep and surface cannot come to disagree about what a burned path is (#1500) c1ad796e · dbf3dbe6…diff
• three copies of one false sentence collapse into one statement the other five are pinned to: the help for cherry-pick and revert said --continue re-runs the operation while squash, apply and merge said it closes the stop three lines away, so the shipped help disagreed with ADR 0080 section 4, with resume::cont, and with itself, on exactly the two verbs whose re-run was built first and measured wrong. USAGE is a literal and there is no const concatenation, so the fact now lives once as resume::USAGE_NOTE and a new census asserts all five USAGE lines carry those bytes verbatim, which means a reword turns five red together where three hand-typed copies turned none; the comments in pick.rs and in_progress.rs now point at that statement instead of paraphrasing it. ere.rs cited the help of loot grep for a claim the help never made, so the sentence exists now as ere::HELP_NOTE and is pinned the same way: a dot matches one BYTE, and [:alpha:], \d, \b and \s are not implemented. the help for restore asserted that nothing ever recorded those bytes of every run, while the module it documents splits IRRECOVERABLE from OVERWRITTEN precisely because that is false under --source. item 5 was wrong twice: a stop cannot be the FIRST operation of a repo, because loot init records a genesis floor op, and the refusal is not an off-by-one either, since op_index answers 0 for an empty log, no operation 0 exists to hold a captured view, and the pre-operation pointer bytes were never snapshotted, so no other number reaches them. what is true is that .loot/ops is lane-owned and nothing plants a floor in a spawned lane, making the unabortable case an ordinary one: the first view-changing command an agent runs inside a fresh lane. so the refusal stands, the message now says that and names --continue, and it is pinned through a really spawned lane whose empty op log is the control. the ATTACHED census control is now derived from the census rather than kept as a second hand-written list, and resolves each censused group name to its canonical const so an empty group of the same name cannot satisfy it vacuously, proven by a mutation that leaves the census green and turns the control red on grep, the one attached leaf that had no control at all (#1799) 51aea3d1 · dbf3dbe6…diff
• the shorthand whose NAME states the path axis while its SHAPE states the arity axis is DELETED, so the terse spelling is now the one that REFUSES - which was this ticket whole thesis, that the wrong declaration was cheaper to write than the right one and that is why the silent-drop class kept recurring. every site that meant it now types the open constructor out, and the only shorthand left is the one that takes nothing. option A beat option B on BOTH axes, measured rather than preferred: B would have changed the constructor signature, so EVERY open call site owed a reason string - including the path-taking and the genuinely variadic ones - and the two dozen dispatch verbs would each have written the SAME sentence, which is boilerplate that teaches nothing and is itself a hand-maintained population. so A has the smaller blast radius AND the stronger property. the blast radius is ZERO BEHAVIOURAL, because the retired constant was literally that expression: no verb declared arity, no slot kind and no refusal moved - thirty-five declaration sites, eleven imports and about twenty-five prose sites, with the workspace check clean and no new warnings. the exemption list was ALREADY down to its two legitimate names before this began, since #1569 narrowed the other twenty-one hours earlier, so nothing was added to it or taken from it, and the #545 refusal that earns those two their place is untouched by construction - pinned rather than incidental, because the mutation that hands one of them the no-arguments declaration reddens all three censuses. the rename then exposed two more counts standing beside sets that MOVE, and both are fixed rather than carried: a fixture doc claiming all FOUR verbs it exists to serve are exactly this shape, where there are FIVE production attachers and NONE of them is that shape, and a line naming the four verbs that used the retired constant. both now state the rule and count nothing. the new guard refuses BINDING the zero-slot open claim to a name, which is the single edit that would undo this, while deliberately NOT refusing a leaf that spells the claim out at its own spec - the two told apart by what PRECEDES the constructor, with both run through the predicate before its answer is read. its limits are in its own header. and the control that mattered is the second: with the comment-strip removed AND the predicate control disabled, the tree scan names the flags file itself, over the retired declaration QUOTED INSIDE THE SURVIVING CONSTANT OWN DOC - so the strip is load-bearing rather than decorative. the first and third mutations are each other discrimination, one reddening only the tree arm and the other only the binding-versus-spelling arm, and the fourth proves the floor fires at zero files rather than agreeing silently (#1675) 123fdbd4 · dbf3dbe6…diff
• a verb arm now receives the argv door answer instead of the raw slice: VerbRun takes an Admitted value, the rebased tail bound to the spec of the row it was dispatched from, and its readers are the only way in, so argv[0] or argv.first() in an arm fails to compile (both planted, read as E0608 and E0599). spec_of is test-only and its 104 production lookups read through the admitted spec; six of the seven raw-index arms (bundle, grant-status, embargo-status, attest, resolve, completions) read positionals through the door and relay reads its leading word through a named exception; subcommand families resolve their leaf into a new Admitted, and bisect and buoy are handed one ahead of the table. the two named exceptions, leading_word for relay and bisect and pass_through_tail for the bisect run command, plus the one process argv read, are held to a named set of functions by a new census that went red on a planted leading_word read in cmd_bundle, and the source walker consumer header now counts it. refusals and output are unchanged by every suite and by a read of each migrated arm, with one exception found on the way: resolve with a trailing --tool and no command read --tool as the file and now refuses with the usage line, pinned by a new test. CONTEXT.md gains the Admitted argv term; loot-first keeps its own spec_of and DECLARED_READS is untouched (#1929) d92e95f2 · dbf3dbe6…diff

Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.