A session studies a round of up to the settings' cards per round. In
spaced repetition a card leaves the round once it is a day or more away,
one rated lower comes back when due; cramming walks every card once,
round after round, its reviews moving no schedule, with going back and
on. Asking the back first (a toggle in the header, saved in the
settings) asks front/back and vocabulary cards their back, on their own
schedule. Each type studies its way: a word shows its reading and
meanings, a multiple-choice card is picked and checked, a matching card
sorted by dragging or tapping and checked; then the person rates (Bad,
OK, Good, Perfect, each with when the card comes back). Progress shows
as pills, or a tally bar while cramming; a round ends with how it went.
The session settings gain the mode, cards per round, new cards per day
and the order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
How the person studies, the same on every device: spaced repetition or
cramming, the order (newest first for cramming only), cards per round
(5 to 20), new cards a studyset starts a day (0 for no limit), and
whether cards ask their back first. The settings payload's schema 2
holds them, each left out while as it comes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Each review records its mode (ADR 0004, rule 3): front to back, back to
front with a schedule of its own, or crammed outside the schedule. A
card's schedule replays only the reviews of one mode. Resetting progress
appends a 'reset' entry per card, no migration (the column is text):
schedules replay what follows it, and the history stays for the streak
and the heatmap. Reviews are inserted in batches.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A studyset's cards list as bordered rows of one height showing the
front, the card's tags above it and a badge when devices changed it
differently; tapping opens the composer, "…" offers editing and
deleting (after asking).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The editor and its read-only view used the editor's own defaults: black
text, unreadable in dark mode, and 100 dp of padding beside it on
desktops. Both now take the theme's text, link and selection colours and
leave the padding to what holds them; phones get the touch variant with
selection handles. A multiple-choice card's last right answer keeps its
switch enabled (and its colours); turning it wrong does nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Create flashcards, and tapping a card, open the composer: the stack of
cards on the left (New flashcard on top, the new ones, then the stored
ones; collapsible), the card being written on the right, saved as the
person leaves it (another card, the pane, the screen, the app). The
title says how saving stands; leaving with cards that couldn't be saved
asks first. A card has its tags and their pin, an action menu
(duplicate, clear, delete), the type switch (flashcard, vocabulary,
matching), the front, and below a line its answer side: one answer or
several to mark right or wrong, the reading and meanings, or categories
and items. Fields changed differently on two devices show every version
to keep one. The single-card editor goes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A bloc holding a studyset's cards as drafts, new ones on top and the
stored ones below. A card is saved when the person leaves it, once it
changed and has everything its type asks for, only its type's fields
stored; a failed save waits for the next leaving. A new card takes the
type of the card before it and what is pinned of it (front, answer,
tags); duplicating, clearing and deleting; a card devices changed
differently first needs a version chosen for every field in conflict.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A card has a type: front/back, multiple choice (drawn on a map when it
names one), vocabulary or matching, and the fields of that type. The card
payload's schema 2 holds the type and the fields in use; schema 1 reads
as front/back. Revisions merge per field rather than per side (ADR 0003,
amended): a field one device changed takes the change, a field changed
differently is a conflict, lists included as a whole.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The study screen's session settings choose tags to study a studyset by:
its cards with any of them. The choice lives in the person's settings
row, synced, by tag id and name, so a renamed tag stays chosen and so
does one folded into another of its name. Two settings rows made offline
fold into the one with the smallest id, which gets what only the other
has.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The card editor offers its tags: a picker that searches by the start of
a name, creates a tag (name and colour) and chooses it, and saves the
choice with the card. A sheet lists every tag with its count, to add,
rename (merging into a tag of that name), recolour and delete them, with
an undo. The card list shows each card's tags and narrows to one tag.
The editor closed twice once tags changed after saving; it now closes
on becoming done only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A tag is its name: names are lowercase, a person has one tag per name,
and two of one name (made offline on two devices, or by renaming) fold
into the one with the smallest id, so every device ends up alike. The
colour lives in the payload. Deleting a card deletes its tags too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A studyset's screen, as decided with the tree (2026-09-30/10-01):
- A breadcrumb from the library down to the studyset; its steps go up
(replacing the location, so back leads further up), and steps that
don't fit fold into a "…" listing them, nearest first.
- From expanded windows on, two columns: the description over the topics,
the flashcards beside them, Study and Add material in the app bar. Below
that, the topics are in a drawer, opened from a bar at the bottom that
also holds Study and Add material and slides away while reading onwards;
the back button stays where a drawer would have taken its place.
- The topics: "All topics" for the whole studyset, then the studysets
inside it in their order, each with the study it invites, what it holds
and its options (edit, delete), opening on top of this screen; a switch
rearranges them by dragging; "New sub-studyset" creates one inside.
- Study covers the studyset with everything inside it ("Learn 12" while
something is due). Add material offers flashcards for now.
- Options: Edit (the sheet, now with the description, edited in the whole
window, and for a studyset inside another the colour of the one it is
in, which a new one takes at first), Move to… (the tree and the top
level, without the studyset itself and what is inside it), Archive or
Unarchive with Undo, Delete with Undo, which closes the screen.
- The description cut short when long, the whole a tap away; the licence
as the last line, linking to its text where that is known.
The list's bloc, scoped to a parent, lists its children, works out the
studyset's own progress, creates inside it, reorders and moves. The card
list leaves adding and studying to the screen. The inline editing of the
title and description, and its bloc, are gone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Studying a studyset covers its subtree (decided with the tree, ADR 0004):
the studysets of its subtree in tree order, without archived studysets and
what is inside them (studiedIn); an archived studyset studies nothing. The
study queue takes the cards of each studyset as a group: due cards of every
studyset first, earliest first, then each studyset's new cards in tree
order, at most 20 started per studyset and local day, as before for one.
The library's cards count the same way, so a studyset whose cards sit in
its sub-studysets still invites a session, and the badge counts what the
session would study; an archived studyset, paused, still counts all it
holds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Decided 2026-10-01. Deleting a studyset from its own screen closes that
screen, which would have dropped a deletion kept there, and the studyset
would have quietly stayed. StudysetDeletions, opened with the person's data,
now keeps every deletion that can still be undone: a deleted studyset leaves
every list at once, and once the time to undo is over the whole subtree is
deleted, what is inside first, each studyset's cards blanked before it. It
closes first with the person's data: what still waits is dropped and stays,
what is under way finishes while the repositories are open. A failed
deletion shows what is left again.
The library's bloc hands its deletions to it and leaves out what it reports
as pending, instead of keeping timers of its own. Tests that follow a real
timer wait for the outcome (eventually) rather than for a fixed time a busy
machine might not keep.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The model learns the tree (ADR 0004): a studyset's children in order (by
position, ties by id), its subtree in tree order (the studyset, then each
child with its own subtree), and its ancestors from the top level down,
for the breadcrumb. Studysets moved into each other on two devices make a
cycle; subtrees stop where it comes back, and climbing stops at the
studysets that show at the top level.
The repository creates a studyset inside a parent, after its siblings;
moves one into another studyset or to the top level, after the studysets
already there, writing only its parent, position and updatedAt; refuses a
move into the studyset itself or anything inside it; and reorders a
studyset among its siblings with a fractional index, as cards do.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Decided 2026-10-01. CI run #32 died in its analysis: a cold `dart analyze`
of the workspace with the lint plugin now peaks at about 3.7 GB (2.1 GB
warm), measured with an empty analysis cache, while the runner's unit
allowed 3 GB. The kernel killed it, systemd's default OOMPolicy (stop) then
stopped the whole unit, and the job's remaining processes got SIGTERM: the
"signal: terminated" in the log. The packages this branch added tipped it
over.
The runner may now use 5 GB of sahaquiel's 8, and above 4 GB the kernel
reclaims from CI first (MemoryHigh), so the instance keeps at least 3 GB;
it uses under 1 GB today. Installed by hand on sahaquiel, like the rest of
the unit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Since the app generates its localizations (`generate: true`), `flutter pub
get` at the workspace root, as CI runs it, writes build/.last_build_id
there. CI's generated-code check counts that untracked file as a difference
and failed run #31 after 24 seconds; a developer running pub get inside the
app package never sees it, because the app ignores its own build/.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Decided 2026-10-01. Deleting still asks first; once confirmed, the studyset
leaves the library at once and "Studyset deleted" offers Undo for three
seconds. Only when the time to undo is over (four seconds: the message's
three and a moment for a late tap) are its cards and then the studyset
blanked and deleted, since blanking can't be undone (ADR 0002, rule 2). A
deletion still waiting when the library closes (the app quits, the person
changes) is dropped, and the studyset stays.
The library's bloc keeps the waiting deletions; the event that ends the
time to undo is private, so no screen can skip it. AGENTS.md now describes
the library's shared widgets, the sheets (and why a sheet that asks before
discarding can't be swiped away), the order of studysets, and this Undo.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Decided 2026-10-01: after the starred ones, the library lists the most
recently changed studysets first. That needs a change date, so studyset
gains updatedAt (ADR 0004): when the studyset itself last changed, its
content, place, star or archive date, written by the device making the
change in the same update as the change; the last writer wins. Studysets
from before have none and come last, the newest of them first.
The branch's migration is still unreleased (not merged, not deployed), so it
is recreated rather than followed by a second one:
20261001064926911-tree-tags-assets replaces 20260930213716558 and differs
only by one line on the server, ALTER TABLE "studyset" ADD COLUMN
"updatedAt" timestamp without time zone (nullable, so existing rows keep
null and no data changes), the same column in the table devices rebuild,
and its version. The model's comment on the payload no longer lists schema
1's fields alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Decided 2026-10-01, in place of "closing the account dialog cancels it",
which had no dialog left once the account moved into the Profile tab:
choosing another tab cancels a sign-in waiting for GitHub in the browser,
since the person has left it behind. Settings, opened over the profile, is
not leaving it.
Screens learn whether their tab shows from TabVisibility, which the tabs'
container now provides: go_router's indexed stack as before (every tab
alive, hidden ones offstage with their animations paused), each tab wrapped
in it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The library tab becomes a grid of studyset cards (up to 400 dp wide, all
equally tall), starred ones first, and the archive as a section of its own
below, closed until opened and counting what rests there. It lists the top
level: studysets without a parent, those whose parent is gone, and those in
a cycle (ADR 0004). Each card's progress follows the studysets and every
change of a card, here or synced, and is worked out again after a study
session; a slow, older working out never overwrites a newer one.
- Search: a field below the app bar filters by title and secondary title,
opens the archive while it finds something there, and says when nothing
matches; closing it shows everything again.
- Creating and editing: a sheet with the title, the secondary title (typed
in capitals, as every surface shows it) and the colour, picked from the
palette's twenty, whose names are localised; a new studyset is cobalt,
and an existing one keeps its colour, even one only a newer app version
knows, until another is picked. Create waits for a title. Closing the
sheet with changes asks first. It can't be swiped away: a bottom sheet
swiped away closes without asking its PopScope (Flutter issue 152120,
still open), so such sheets have no drag handle and close by their close
button, the back gesture or a tap beside them, which all ask.
- Archiving doesn't ask, since nothing is lost; it offers to undo, as does
unarchiving. Deleting asks, then deletes the studyset with its cards, each
blanked first.
- The create button floats in compact windows and slides away while
reading onwards; from medium windows on it sits in the app bar.
The old list and its title-only dialog are gone; the studyset screen's
tests create studysets through the sheet.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A studyset in the library: a bar in its colour, its title with the secondary
title as a kicker, what it holds, and beside the title the action it
invites inside a ring its mature cards fill, with a badge counting the due
cards (at most "99+", growing inwards so it never passes the card's edge).
The action is a review when reviews are due, studying when only new cards
are, and adding content otherwise, which opens the studyset: the card only
invites a session that has something to study. A long press stars it, as
does the star, which a hovering pointer reveals and which stays while
starred; the options edit, archive or unarchive, and delete it. An archived
studyset rests: dimmed, neutral, paused. Until its progress is known, a
muted bar stands in for "No content yet", which it might not be.
A starred star is amber, dark enough in the light theme to stand out from
the cards as a graphic must (3:1), a colour the theme extras now hold. It is
Material's filled star: Lucide has no filled one, and the fill is what says
the star is on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
What a library card shows about its studyset's cards: how many there are,
how many a session started now would study (the study queue's length, so the
card only invites a session that has something to study), how many of those
are reviews rather than new cards, and how many are mature.
A card is mature once FSRS's stability, the days it is remembered with a 90 %
chance, reaches 21: the three weeks after which spaced-repetition apps
traditionally call a card mature, measured by memory rather than by the
interval. A schedule now carries its stability.
Replaying every card's schedule from its reviews moves from the study
session's bloc into the review repository (schedulesOf), and the start of the
learner's local day into a function beside the study queue, so the session
and the library compute alike.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A studyset is now its content (StudysetContent, everything the synced payload
holds) and the columns that change on their own: where it sits (parent,
position), and when it was archived or starred. Like a card's content, the
payload's part is edited as one value; the columns are written one by one,
so starring or archiving on one device and editing on another all survive a
sync (ADR 0004).
The payload's schema 2 adds the secondary title, the colour (a palette name),
the licence (SPDX identifier or name, holder, year) and the languages of the
cards' two sides (BCP 47), each left out when not given. Schema 1 reads as
schema 2 without them, and a field of the wrong kind makes the payload
unreadable rather than half read. Older app versions skip schema 2 studysets,
which costs nothing before the first release; fields foreseeable for the
first release's features are in, so none of them needs a schema 3 later.
The repository stars, unstars, archives and unarchives (at clock time, in
UTC), writing only that column. A new sync test (run against a server)
checks that a star and an archive on one device and a rename on another all
arrive on both.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The pieces the library's cards and its create/edit sheet are built from, each
with its tests:
- TitleWithKicker: a title with an optional kicker above it (a studyset's
secondary title, uppercased in the primary colour), a trailing action lined
up with the title, and a reserved kicker line so cards side by side line up.
- ContentCountsLine: "3 studysets • 52 flashcards", leaving out what is
unknown or zero, with a placeholder when nothing is left; labels are plural
messages, so every language words its own counts.
- showAdaptiveSheet and SheetScaffold: a bottom sheet below expanded windows
and a dialog from there on (a bottom sheet is a touch surface; expanded
windows are mostly desktops). The title with its close button and the
footer stay in view, only the body scrolls; the close button asks the
route to close, so a sheet can ask before discarding changes.
- showUndoSnackBar: what just happened, with Undo for three seconds.
- ScrollHidingFab: a floating action button that slides away while reading
onwards and returns on scrolling up, and stays where nothing scrolls.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The app gets its look and navigation before more features land on them.
- Theme (core/theme/): hand-written light and dark Material 3 colour schemes around
cobalt and copper; ThemeExtras for the colours Material has no role for (states, the
four FSRS ratings); the 20 colours a person can give content. Tests hold text to WCAG
AA contrast on each palette colour and on the dark theme's primary colour and surface.
- Layout and shared widgets: Material 3's window-size classes (core/layout/);
SectionCard, AppLoadingIndicator, confirmDestructiveAction and FoundById
(core/widget/). Lucide icons.
- Navigation with go_router: four tabs (Home, Library, Explore, Profile), each keeping
its own stack of screens, as a navigation rail with the app's icon from 600 dp and as
a navigation bar below; choosing the open tab again returns to its first screen. A
studyset's screen sits in the library; studying, cards and settings cover the tabs.
- Routes carry ids: FoundById finds the item in the repository's live list, so a
studyset deleted on another device, or an old link, says it no longer exists. A
screen that deletes its own item keeps showing it while it closes, rather than
flashing that message.
- Every person gets an app of their own (PersonApp, lib/src/app/): a
MaterialApp.router with its own GoRouter and PersonScope around the navigator, built
by SessionGate inside the person's keyed subtree. Screens read the person's
repositories wherever they are built, so the static route() helpers and the listener
that closed every pushed screen on a change of person are gone: a change of person
now disposes the whole router and starts again at home.
- The account dialog becomes the Profile tab, with Settings (sign out, delete the
account, About) behind a gear. A waiting GitHub sign-in now keeps waiting while
another tab shows; a change of person still cancels it.
- The new screens' strings come from Flutter's gen-l10n (English for now). The
generated files are committed, left out of analysis and formatting, and written
anew by `flutter pub get`, so CI's check of generated code also catches stale ones.
Widget tests now start the whole app (test/support/pump_app.dart) at the screen's path
and move through it as a person would, by tabs and taps.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
studyset gains archivedAt (archived studysets rest in their own section and are left out
of studying, with everything below them) and starredAt (kept at the top of the
library), both nullable. Columns rather than payload fields, for the same reason as the
parent: starring on one device and renaming on another must not overwrite each other.
The branch's migration wasn't applied anywhere yet (tests build fresh databases, and
the branch isn't merged), so it is recreated with the two columns instead of adding a
second one: 20260930213716558-tree-tags-assets replaces 20260930211154224, and differs
from it only by the two ALTER TABLE lines and its version.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The schema of the features planned for the first release, as one migration: every
migration of a synced table locks older app versions out of syncing (ADR 0002, rule 3),
which costs nothing before that release. New card types need no migration; they are the
next version of the revision payload.
- studyset: `parent` (the tree; cycles and orphans show at the top level) and `position`
among siblings (fractional index; existing studysets at 'a0').
- tag (its name in the payload) and card_tag, one row per assignment, so concurrent
adds and removes merge.
- review_log.mode: review, reverse (a schedule of its own) or cram (never moves a
schedule); existing reviews are 'review'.
- person_settings: study goals now, preferences later; random ids, since an id derived
from the person could be claimed by someone else and break the owner's sync.
- asset: a file's hash, type, size and name; the file goes into Serverpod's private
cloud storage, served through an owner-checked endpoint (to come).
Migration 20260930211154224-tree-tags-assets only adds: four tables with their space
indexes and foreign keys, three columns with defaults. On devices SQLite rebuilds
studyset and review_log to add foreign keys, which is safe only with foreign keys off,
as Serverpod's runner does; a new test replays every client migration on a database
with a row in each of our synced tables and checks that none is lost (with foreign keys
on, it fails as it should).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Two of Serverpod's own conventions, as its v4 project template and roles have them.
CI now runs `serverpod generate` (CLI pinned to the server's Serverpod version, like
the template's workflow) and fails if the committed generated code differs from its
output, so a model or endpoint changed without regenerating can't reach main. Checked in
a clean clone: today's code passes, a committed stale generated file fails.
redeploy.sh now applies the migrations in a run of their own, with Serverpod's
maintenance role (--role=maintenance --apply-migrations), after the dump and before
switching to the new server. That run exits with 0 only if the migrations worked, and
each migration is one transaction, so a failing one stops the deploy while the old
server keeps serving; before, the new server applied them at start and would have
crash-looped. Tested on the local stack: a deploy through the new step, and a failing
run (wrong database password) that exits with 1 after six attempts. AGENTS.md says
migrations must keep the old server working until the switch.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The first CI run failed at the server tests: they need the test run mode's passwords,
which passwords.yaml holds on a developer's machine and which is never committed. Without
them Serverpod's test tools exit with code 1 and print nothing. Reproduced in a clean
clone; with the three passwords in the environment (SERVERPOD_DATABASE_PASSWORD,
SERVERPOD_PASSWORD_serverSideSessionKeyHashPepper, SERVERPOD_PASSWORD_serviceSecret) all
checks pass. The job now draws new random ones every time: they only guard the test
database it creates in its own checkout.
Also switch off Flutter's and Dart's usage reports on every job, not only when installing
the SDK: the cancelled first run left an SDK behind before getting that far.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Installing the runner on sahaquiel, `useradd --system` gave `runner` user ID 999: the ID
under which PostgreSQL and the server run inside their containers, which Docker doesn't
remap. To the kernel, those processes were the runner's own, so a job could have read the
server's environment (database password, session pepper, GitHub secret) and, through
/proc, PostgreSQL's files and its local socket, which the official image trusts. The
sandbox's other walls don't help there: ProtectProc hides only other users' processes,
IPAddressDeny doesn't cover Unix sockets. Only our own check ran as 999; `runner` now has
user and group ID 61000.
The check job now starts with a Sandbox step that fails unless only the runner's own
processes are visible, all in one process namespace, /home/deploy is invisible, the
Docker socket is unusable and Docker's bridge network unreachable. It would have caught
this; outside a sandbox it fails as it should. The unit and AGENTS.md say how to create
the user.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Pushes to main were deployed by hand. Now Forgejo Actions runs tool/check.sh on every
push, and a green main deploys learn.halkyon.eu by itself.
The runner (forgejo-runner v13.1.0) runs on sahaquiel as the unprivileged user `runner`,
with jobs directly on the host: containers would take Docker access, which is root on the
production server, and Forgejo warns that host jobs have no isolation from the runner's
user. So its systemd unit (deploy/runner/forgejo-runner.service) provides it: /home and
the Docker socket invisible, Docker's container networks unreachable, no capabilities,
two cores and 3 GB at most, lowest priority. It is registered to this repository only,
with the UUID and token of Forgejo's "Create new runner" (the old `register` is
deprecated); the token reaches it as a systemd credential. The configuration
(deploy/runner/config.yml) was dry-run against v13.1.0, the unit rates 1.8 in
`systemd-analyze security`. CI runs on pushes only: a pull request from a fork would run
its code on the production server.
The deploy job calls deploy/redeploy.sh as `deploy`, through a key that authorized_keys
pins to that script and to connections from sahaquiel itself. The script deploys the
commit CI checked while it is still main's tip, as a fast-forward, only if every new
commit is signed by a key in deploy/allowed-signers (as the deployed checkout lists
them), after dumping the database (the last ten are kept), and fails unless the server
answers afterwards; a failure puts the checkout back. Tested against the local stack:
refusals of a malformed argument, a stale commit, a forged unsigned commit and a
non-fast-forward, a real deploy with its dump, a repeat, and a failed build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The app took the editor from our fork of AppFlowy Editor in a folder next to the
repository, which nobody else had, and which CI can't reach either. The fork now lives at
https://code.halkyon.app/project-halkyon/appflowy-editor (public; dual-licensed AGPL-3.0
or MPL-2.0 like upstream), on its branch halkyon/win32-v6, and the app depends on it by
git URL, pinned to the signed commit 0dac8a0. That commit adds `publish_to: none` to the
fork: it is a git dependency only, and the package name belongs to AppFlowy on pub.dev.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The image build wrote a workspace root of its own, containing only the server package, and without the root's dependency overrides. Since the offline sync package came in (step 7), it pins Serverpod v4.0.0 exactly while the server uses v4.0.3, so `dart pub get` failed with "version solving failed": main could no longer be deployed. The deployed instance predates step 7 and was never built this way.
The build now copies the repository's root pubspec and only removes the two members it can't or needn't build (the client, and the Flutter app, which would need Flutter). The overrides stay in one place, and a check fails the build loudly if the member list changes shape.
Rehearsed locally on a database with the old schema, as on sahaquiel: the image builds, the studysets, cards and reviews migrations apply at start, GET / answers 200 through Caddy, instance.info works, the log shows no errors, and the pruning job is scheduled once, also after a restart.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Signing in with GitHub on Linux ended in "Unable to connect" at 127.0.0.1:43823 after authorizing. flutter_web_auth_2 (v5.1.0, unchanged on its master) ends every pending sign-in when the app is resumed. That suits mobile, where coming back without finishing means giving up, but on the desktop the window regaining focus resumes the app, and the package's loopback server closed before GitHub sent the browser back. FluffyChat's issue #3341 and the package's #171 report the same.
The app now puts its own Linux and Windows implementation behind the package's platform interface (LoopbackWebAuth): the system browser and a loopback address, as RFC 8252 recommends, waiting until GitHub answers, the person cancels, or five minutes pass. It listens on 127.0.0.1 only and answers only the callback's path. Serverpod's PKCE and state check stay in use. There, the account dialog shows "Continue in your browser." with a Cancel button, and closing the dialog cancels the sign-in.
Serverpod (v4.0.3) reports a cancelled browser round trip as an unknown error, so the gateway knows a cancel by having asked for it. A cancel on GitHub's own page, or leaving the browser on Android, still reads as a failure; AGENTS.md says so.
The regression test drives the package's front through a change of focus. With the package's behaviour it fails with the same refused connection.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Rewording two commit messages with a rebase, while serverpod start was running, left two generated files at an intermediate state: its watcher regenerates code whenever a model changes, and the rebase checked out every commit in turn. The commits were unaffected; the files were restored from the tip. AGENTS.md now says to stop serverpod start first, or to work in a separate worktree.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Run against serverpod start, two assumptions of the sync tests turned out wrong, and all eight tests pass with the fixes.
The SQL checks connected to the development database over TCP, but the embedded PostgreSQL that serverpod start runs listens on a Unix socket on Linux and macOS and trusts local connections. queryServer now connects through the socket, by a short link in /tmp because socket paths are limited to about 100 bytes, and keeps TCP with the configured password for other platforms.
The forgery test expected the server to refuse Alice's write into Bob's space and log it. In fact an honest device only syncs the spaces the server announces, so the forged row never leaves Alice's device: no error, no log entry, nothing on the server. The test now claims what it proves, that the row never reaches Bob and is not on the server. The server's own check against a modified client (_assertCanMergeInboundSpace) stays untested; reaching it would mean speaking the sync protocol by hand. AGENTS.md says so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The account dialog offers "Delete account", which asks first and then starts anew with an anonymous account. Signing out of an anonymous account deletes it, and so does switching away from one to GitHub, as decided: nobody could ever sign into it again, so nothing should wait on the server. The confirmations say so.
When a person's data opens, PersonFolders wipes every other person's folder: one person uses a device at a time, so their sessions ended here, whether they signed out, switched, were revoked or deleted. Opening waits for the previous person's database to close, so the files can go even on Windows; if closing hangs, opening goes on after ten seconds and the folder goes next time.
Tests: the repository's sign-out and deletion, the session bloc starting anew after a deletion, the dialog (cancelling keeps the account), and PersonFolders on a real temporary folder. AGENTS.md records deletion, pruning and the folders.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
deletePerson deletes a person with everything they have, in one transaction and in ADR 0001 rule 6's order: their sessions first, since a device with a valid session would re-upload everything; then their memberships and personal space, whose foreign keys cascade into every synced table and the sync engine's bookkeeping; then the person, whose identities cascade. AccountEndpoint.delete deletes the caller and tells their devices that their sessions ended.
pruneAbandonedAnonymousPeople deletes, 500 at a time, anonymous people created and unused for 90 days: their sessions have expired and nobody can ever return to them. Identified people are never pruned. PruningFutureCall runs it daily at 03:17; the server schedules it at start, cancelling the previous entry by its identifier first, since scheduling always adds one.
Tests, refusal first: deleting when not signed in; deleting removes the caller, their sessions, their anonymous identity and the content of their space, and nobody else's; pruning takes the unused and the never-signed-in, and spares the recently used, newcomers and identified people.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A studyset's card list gains "Study". The study screen shows the front, then the answer and the four ratings, each labelled with when the card would come back; every rating is recorded with how long answering took, and the queue is worked out anew after each.
Schedules are never stored: scheduleOf replays a card's reviews in time order with FSRS-6 (fsrs v2.0.1, pinned exactly so every device on one app version computes the same). The package's fuzz can only be seeded through an API meant for tests, and due dates don't feed back into a card's state, so the replay runs without fuzz and fuzzedDays spreads the final due date over FSRS's range, picked by a stable FNV-1a fraction of the card's id and review count. studyQueue takes due cards earliest first, then new cards in studyset order, 20 per studyset and local day, and brings a learning card due within 20 minutes early when nothing else is due. Deleting a card deletes its reviews.
Tests: the replay (learning, growing intervals, the same schedule whatever order reviews arrive in, previews matching what a rating yields), the fuzz range, the queue (order, the daily limit counted per local day, learn-ahead), the repository, the session bloc under a controlled clock, the screen, and a sync test for ratings on two devices. AGENTS.md records the review log and scheduling, and how plugin lint rules are ignored.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
ReviewLog keeps every rating of a card: the card, the rating (FSRS's 1 again to 4 easy), when (UTC) and how long answering took, as decided. It is never changed; each device derives a card's schedule by replaying its reviews, so reviews from different devices never conflict. No device id: the sync engine knows each row's device.
The migration (by CLI) only adds: the table, an index on spaceId for space deletion, one for a card's reviews in time order, and the foreign keys (from the space with ON DELETE CASCADE; from the card with ON DELETE CASCADE, DEFERRABLE INITIALLY DEFERRED).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The account dialog opens Flutter's about dialog: the licences (AGPL-3.0-or-later for the app, SSPL-1.0 for the server), the address of the source the instance names, as selectable text, and "View licenses" with every package's notice, which the BSD and MIT packages require in a distributed app. The address comes through a small about feature (gateway, repository, bloc) that says so when the instance can't be reached. Tests cover the bloc both ways and the way from the account dialog; AGENTS.md records the obligation.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Both licences (AGPL for the app, SSPL for the server) require offering the source of what an instance runs to the people who use it. instance.info() now carries sourceCodeUrl: the operator's SOURCE_URL, which they must set when running modified code, and otherwise this project's repository. Serverpod validates its configuration files strictly, so the value comes from the environment; deploy/compose.yaml passes it on from deploy/.env, whose example explains it. A test covers the default.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
What Halkyon Learn is and its three packages with their licences; how to run it locally (the editor fork next to the repository, the git-ignored password file and its keys, serverpod start without analytics, the server URL); the checks and the sync tests; running an instance with deploy/; and where the rules, decisions and licences are.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The screen that says "Couldn't open this device's data." had no button, so the person was stuck until they restarted the app. It now offers "Try again", which opens the person's data anew; a test covers a first attempt that fails and a second that works.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Records step 10's decisions with their reasons and the alternatives weighed: a list of parents instead of ADR 0001's single parentRevisionId (a revision with two parents resolves a conflict), a three-way merge per side against the newest common ancestor (no silent merge inside a side), and an explicit fractional position for the order. ADR 0001 gets an "Amended by" line, and AGENTS.md points to the new ADR.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
A studyset's screen lists its cards (reorderable; a marker on cards changed on two devices) and opens a card's screen with an editor for each side. Saving appends a revision whose parents are all current heads. mergeRevisions merges several heads per side against their newest common ancestor, as decided: different sides merge automatically, the same side changed differently is a conflict. The card screen then shows every version of that side with "Keep this", and saving is possible once each conflicted side is settled, which resolves the conflict for every device. Nothing is merged silently inside a side.
CardRepository watches a studyset's cards, creates at the end and moves by fractional index (fractional_indexing_dart, checked against the reference algorithm's test vectors), and deletes by blanking every revision first (ADR 0002, rule 2). Deleting a studyset now deletes its cards the same way before the studyset goes. Every person gets a CardRepository whose revisions name them as author. RichContent gains plainText for the list's previews.
Tests: the merge (chains, different and same sides, conflicts, resolving, three heads, missing parents, arrival order), the payload, the repository against a fake gateway, both blocs, and the widgets, including that cancelling a delete keeps the card. The sync tests, which need the running server, cover merging and conflicts across two devices and blank revisions on the server; their device helper is now shared in test/sync/support. AGENTS.md describes the revision pattern, the fractional order, and two tooling traps.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Cards are the second feature that stores content as a versioned UTF-8 JSON payload, so the shared part moves to core/content/payload.dart: encoding with a schemaVersion, decoding that refuses garbage and newer schemas, and the blanked redactedPayload (formerly redactedStudysetPayload). The studyset codec now builds on it and behaves as before; its tests pass unchanged apart from the rename.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
Card belongs to a studyset and carries only its position, a fractional index (last writer wins), so cards can be reordered and bulk imports keep their order. Its content lives in CardRevision rows, which are never updated except to blank them (ADR 0001, rule 2): parents lists the revisions a revision replaces (decided: a list, so a revision with two parents resolves a conflict), authorId the writer, reportId the report it answers once reports exist, and the envelope holds the content. Deciding the list of parents and the position now matters: changing a synced table later locks out older app versions.
The migration (created by CLI) only adds: the two tables, an index on spaceId for each (space deletion cascades into them), one for a studyset's cards in order and one for a card's revisions, and the foreign keys, from the space with ON DELETE CASCADE and from the studyset and the card with ON DELETE CASCADE, DEFERRABLE INITIALLY DEFERRED, since synced rows can arrive in any order. The client migration creates the same tables on each device.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
The account dialog offers "Sign in with GitHub" to anonymous people where the instance has GitHub sign-in. It links the GitHub account to the person, who keeps all their studysets. If the account already belongs to another person (typically their account from another device), the app switches to that person right away when this device holds nothing, and otherwise asks first: switch and leave this device's studysets behind, or stay anonymous.
ServerpodAuthGateway uses Serverpod's GitHubSignInService for the browser part (PKCE), with the system browser and a loopback address on Linux and Windows (http://127.0.0.1:43823/github-callback) rather than an embedded webview, and app.halkyon.learn://github-callback elsewhere; Android gets the callback activity. After linking, the device's copy of the session gets the identified scope. A switch signs in through a client without a session, swaps sessions directly (never through "signed out", which would start another anonymous account), then ends the anonymous session left behind.
GitHubSignInBloc holds the flow; tests cover every branch (GitHub sign-in off, linked, cancelled on GitHub, switching right away, asking with staying anonymous as the refusal, switching once confirmed, a failure), and the widget tests the dialog. StudysetRepository gains all(), a one-off read: Stream.first on its stream never completes under the widget tests' fake clock, because the broadcast subscription's cancel future belongs to the root zone; AGENTS.md records it along with the GitHub pattern.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Signed-off-by: Tizian 「ティツィアーン」 <raisondetredev@pm.me>
