The plan's top now says it was built, the same day it was written: the
ten commits in four phases at the dry run's numbers exactly, Phase 6's
two native items done, and its other three -- pushing the removal, the
two reverse-pulls, the fork decision -- named as the user's. Phase 6
gains its own Built paragraph: what the refusal changed and did not
change in the DAO's welcome branch, the four re-pointed anchors, and a
measurement the plan had only asserted -- that a Mantra commit offers
itself to the fork cheaply. git format-patch on the refusal commit, four
inverse rules over the patch text, and git apply --check on a scratch
worktree of curated/curated at 29027f2b: it applies cleanly, five files.
That is the shape a normaliser running the other way would take, and it
is recorded so the fork can take the commit in seconds rather than by
hand. The README's reading-order sentence and the first plan's pointer
say "built" where they said "planned".
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
9bb34004, 88577bb5 and 29027f2b landed as 8edb52ff, 72b9d3d2 and
ba5ef723, all clean -- including the two picks into docs/README.md, where
the driver's UNION rule was armed and did not fire: git placed the fork's
row between the npub and jvm-target rows and its reading-order sentence
after the npub note's, beside the sentence this plan's own commit had
added there.
The ten are landed. Exactness at the tip is the dry run's: 135 residual
files between Mantra and the rewritten fork, CreateProfileViewModel.kt
the only file that left the residual since the base, and no other file's
residual moved. Both compilers clean, jvmTest 864 -> 868, testDebugUnitTest
420, every audit budget met -- the dry run's numbers exactly, as a
deterministic rewrite replayed onto an unchanged base should give.
One review point, read: MultipleProfilesRoundTripJvmTest reaches the
saved default through the app's getGlobalPrefs rather than building a
GlobalPrefs of its own, and says why in a comment, so the per-process
cache trap is written at the place a future test would copy from.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
759199f2 landed as 72afe8aa, a clean pick, and with it the database's
first migration since the fork: version 20, two nullable owner columns on
the fetch queues by AutoMigration, the export committed as
composeApp/schemas/press.mantra.compose.database.MantraDatabase/20.json.
The check this phase exists for passed: after a full build of both
targets, git status shows nothing under composeApp/schemas/, and the
committed 20.json is byte-identical to the rewritten fork's, identity
hash 4414c373. Room derived the same twenty tables from Mantra's entities
as from Curare's, which is the second decision's rule -- whichever tree
migrates first owns the number, the other pulls before it adds its own --
holding on its first use.
Exactness unchanged at 135 residual files with nothing moved. Both
compilers clean, jvmTest 858 -> 864, testDebugUnitTest 420, every audit
budget met. Read in review rather than assumed: the inbox sweep runs
inside the canSign branch before the broadcast pump and the live
subscriptions launch, returns zero without a query for a key pair that
holds no private key, re-indexes each wrap in a transaction of its own
and logs rather than throws when one cannot be opened; the DAO's own
placeholder and participant syncs carry no owner; the seal now links to
the wrap it came out of; and the migration's comment says why the
broadcast queue was given no owner column.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four commits landed as 84ac84cc, 73a0e5b9, 5654e462 and ccb9942a. Three
were clean picks; the fourth is the conflict the dry run found and the
third decision explained -- CreateProfileViewModel.kt at 2f420230, where
Mantra's 39fb64b6 had moved one import and upstream's commit rewrites the
whole block and removes the derivation that import served. Taken theirs,
by the rule added to the driver, and the file is now identical on both
sides: the residual between the trees went from 136 files to 135, that
file being the one that left, and no other file's residual moved.
Both compilers clean. jvmTest 837 -> 858, testDebugUnitTest 413 -> 420
(StartupChoiceTest), every audit budget met, and the audit's count of
composable string literals -- reported without a budget -- fell from 40
to 30 as the startup screen's wallet-worded literals moved into the
catalogue; strings.xml 444 -> 455. The review points the phase named were
read rather than assumed: the relay observer is a single collectLatest
child with no per-pubkey map; the switch sets the desired id, sets
startWalletImmediately back to true, clears the identity and only then
stops a node, and only when the profile being left had one; the startup
precedence reads force, desired, "show me the list", the only one, the
default; CreateProfileRoute(withWallet = true) appears once, under
Landing; the jvm target reaches the global preferences through
JvmGlobalPrefs alone; the switcher's bar carries NavigateBackButton.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Phase 0 needed one thing this plan did not predict in detail: the
worktree's library checkout sat at 01489b8, an older master whose object
store did not even hold the pinned 84cc44c, while its nested chain was
already at the pinned four. One fetch from the main checkout's submodule
and a detached checkout made the tree clean. The rewrite was rebuilt from
the fork point and produced the same forty-nine rewritten commits as the
dry run, hash for hash: filter-branch with the normaliser is deterministic
for the same input, so the dry run's numbers are the run's numbers rather
than an estimate of them. Pushing the removal to origin/mantra is the
user's step and is recorded as still open.
Phase 2 landed a5ff2641 and 3008137e as 44324c90 and d957ee8d, both clean
picks. The residual between the trees is 136 files before and after --
the dry run's 135 plus this document -- and no file's residual moved.
Both compilers clean, jvmTest 826 -> 837, testDebugUnitTest 413, every
audit budget met, and nothing under composeApp/schemas/ changed after a
full build. The release note the phase owes is written out in one line,
since it is the one downgrade hazard in the pull.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The upstream moved the day the first pull landed: ten commits on
curated/curated, 86cb876b..29027f2b, that let a device hold several
profiles and move between them, and with them the fork's first schema
migration, version 20. The first plan named them the next pull and said
the schema change earned a decision of its own. This is that plan, and
unlike the first it was written after the dry run rather than before it,
so its numbers are measured: rewrite from the fork point in 103 s, ten
commits replayed onto 776455ec with nine clean and one resolved by rule,
both compilers clean, jvmTest 826 -> 868 and testDebugUnitTest 413 -> 420
with no failures, every audit budget met, and 20.json regenerated
byte-identical after a full build.
The verdict is all ten, because the line is one feature and seven fixes
braided together and four of the fixes are live on Mantra today with one
profile: the relay observer that is never all cancelled, the read-only
inbox that stays closed after its nsec is pasted, the DataStore race on
desktop, and the startup screen's wallet-worded literals. Four decisions:
take the whole line rather than carve the fixes out of five commits; take
upstream's version 20 verbatim and adopt the rule that whichever tree
migrates first owns the number; the one conflict is Mantra's own import
from 39fb64b6, resolved theirs, which closes one of the first plan's owed
reverse-pulls; and two of the device's profiles in one group is a Mantra
follow-up before release, since ChatRoom is keyed by the group id alone.
What changed in the method: Mantra's tree is no longer a superset of the
fork's, so the exactness check becomes "the residual between the trees is
the same before and after the pull, file for file", and the plan gives the
commands. The replay driver gains the one rule the dry run needed, with
its reason. The README gets the row and a reading-order sentence, and the
first plan points forward from the paragraph that predicted this one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Kgothatso NgakoandClaude Opus 5