The previous commit treated this package like the polyfill: a provider whose
source is off limits, with the unknowns raised as questions. That was the wrong
regime. The no-read rule attaches to a provider that publishes an engagement —
this one publishes nothing at all, and for that case the protocol has a bounded
reconstruction, made once at adoption time and pinned. So the two questions are
answered here rather than asked, and one of them turned out to be about the Pod
rather than about the providers, which is not our concern.
The observation is now the specification: contract_activitypods-providers,
stamped at the commit it was read from. A record's id is the resource URI and it
is the only identifier, so the seam's two collapse into one. update fetches,
merges and PUTs, so a partial write is safe and the seam declares this backend
as patching.
What the reading also surfaced matters more than either question. Every
container resolution runs against the session's own WebID — in getList, in
create, in a list subscription alike. So this backend can only ever show a user
their own records: no other user's event is reachable, and an event's
participant count cannot be computed from participations the app cannot read.
That is the product, not a detail, so it is need 1 rather than a friction, and
it is phrased as what it is: the capability sits one layer down, exposed as a
WebID parameter, just not reachable through the data provider.
Four frictions are recorded with the workarounds, including the auth provider's
timer and listener that construction installs and nothing removes — which the
runtime backend switch would accumulate, so the interval goes off and the app
checks status itself.
The Refine seam asks four things of a backend, and the ActivityPods README
answers two: how its providers are configured, and that a declared shape tree
gets its own Data Registration and container on the Pod. It is silent on the
record shape and on whether update replaces or patches — the two the seam
cannot default, and the two that fail silently when wrong.
So the declaration is written before adoption instead of after: the surface is
marked as declared rather than measured, the domain conditions are stated at
full strength, and the two unknowns sit in Needs as requests addressed to the
provider. Reading their source to settle them is the one thing this project
does not do.
This branch had no contract manifest at all, so it gets one. It carries no
consume entry, because neither targeted backend publishes an engagement to
pull; the declaration anchors on the README's commit identity, resolved with
ls-remote, which names a state without fetching any content.
The meeting point is the product's unit of value and had no shape at all: it
lived in React state, lost on reload. The shape comes from the sibling branch
rather than from my own reading of the domain -- event, host, title, optional
description/place/time, and the optional inbox its own backend will want.
Regenerating the bindings does what that branch documented it would: the
emitted names lose the Fp prefix, because they follow each shape's IRI and
those IRIs are the persisted RDF classes, not ours to rename. So the aliasing
happens at the four import sites, which keeps every downstream name unchanged
and stops the DOM's own Event from being shadowed. The generated files are
never edited; build:orm overwrites them.
I had called this blocked, as "a binding with no consumer". That was wrong and
worth correcting: the shape and the bindings are the PREREQUISITE every
NextGraph-flavoured backend needs, and only the RecordBinding waits for one.
The RecordBindings slot stays optional until a backend can fill it.
Still local-only in the data context: persisting a meeting point needs a
backend to write it through, and the registry is empty.
The shared-wallet shim's fan-out in the data context fed the ORM subscriptions
this provider used to open: readGraphs, writeGraphs, protectedScope,
protectedGraph and the effect that filled them from storeRegistry. Every entity
reads through a Refine backend now, which takes no graph list, so none of it was
reachable. Same for findNg, mapUser and mapParticipation -- reading a record
into a domain type belongs to a backend's binding, not here. storeRegistry
itself stays: the @data multistore scenarios drive it directly.
This is debt the migration created, not debt it inherited, which is why it goes
rather than waiting for a tidy-up.
What the removal COSTS is written where it happened: placement by scope no
longer narrows what the app reads. The backend's visibility declaration is
where that belongs and nothing honours it yet; applyIsolation is the only
narrowing left.
And setLike.delete is now marked as the defect it is. It is a no-op lifted from
a branch where the fixture seed is off, so nothing needs deleting before it --
false here, where the step that empties the wallet therefore does nothing.
Fixing it needs a decision the comment states, because the app exposes no
deleteEvent.
Every Participation is readable by everyone — only the link to a person is
protected — so an event's count is obtained by counting its participations. No
privileged view, nothing to compensate for.
The stored counter is gone from the write path: joining writes a participation
and nothing else, leaving writes nothing else. Writing the participation IS
moving the count, so there is no second write to keep in step and no counter to
race. That race was measured, not supposed: the count sat on whatever the seed
had written, with no error anywhere, because each increment derived from a read
of the previous one.
Derived in BOTH providers, so demo and connected modes cannot disagree about
what a count means. fp:participantCount stays in the shape and the fixtures,
kept in step, but nothing reads it.
The fixtures become the source of truth and had to say something true: they
advertised 24, 12, 45, 16, 30 against five participation rows. Seventeen rows
now, counts aligned to them. Marie's three are untouched on purpose — the home
screen lists "my events" from exactly those, and the @ui tests render against
it. Small on purpose too: a plausible demo, not a hundred rows reproducing
figures nobody reads.
us-7-inscription-evenement.feature still sets a starting count and expects it
±1. Those steps encode the superseded counter and will need rewriting around
creating participations; they are @data, which is parked while no backend is
registered.
The previous commit wrote a requirement into types.ts that is false: that the
count needs an inbox, and that a CRUD-only backend therefore cannot carry the
model. That came from misreading the brief's "no filtering" as "the readable
participations are not all of them".
Every Participation is readable by everyone. What is protected is the link to a
person — only a connection can read the profile one points at and recognise who
it is. So the count is obtained by counting an event's participations, which is
ordinary CRUD and asks nothing of a backend.
Left standing, because the measurements did establish it: the count must not be
an incremented field. Reading a stored counter out of a list, adding one and
writing it back derives each write from a read of the previous one, which races
itself — and that mutated counter is the one the project's brief supersedes.
The ng-org backend stays withdrawn, on the reason that holds on its own: it is
not the target, the NextGraph+polyfill provider is.
The participant count is the number of participants INCLUDING those this session
cannot see, so it cannot be a count of the participations we are allowed to
read. The project derives it: whoever signs up deposits a marker into the
event's inbox, and the event OWNER recomputes the field from those deposits.
That needs an inbox. @ng-org/refine-providers is CRUD over SHEX shapes and has
none, so it cannot carry the model — and the counter it could carry, incremented
from a value read back out of a list, races itself and is the one the project's
brief declares superseded. Measured, not assumed: optimistic mutations moved one
assertion from the seeded value to the written one and left the rest.
So the wiring goes and the structure stays. The contract, the registry, the
runtime switch and useBackend are untouched; what is gone is the single entry,
its three ng-org-specific files and the dependency. What the next backend must
answer, the derived count included, is written in types.ts, and the withdrawn
implementation stays in this branch's history as the worked example.
Two refusals rather than quiet degradations. With no backend the data context
falls back to demo data and SAYS so, instead of mounting a connected path that
would hand every screen empty lists. And the @data harness refuses to publish
its bridge at all: on fixtures those scenarios would assert happily and report
green while touching no wallet, which is worse than red.
Profiles and participations now read through useList and write through
useCreate/useUpdate/useDelete, so nothing in the data context talks to the ORM
any more. The seed receives injected writers instead of ORM sets, and decides on
the count the caller has already read -- taken before any await, which is what
the previous version got wrong in the other direction.
The test bridge had to follow. It exposed the raw ORM sets, which no longer hold
what the app holds: a step observing one while the app decides on the other
races itself, and an explicit load seeded a second time because the set still
read empty. The bridge now reads the app's own views through a ref, because a
captured snapshot goes stale after a seed -- the sibling branch had met this and
said so. Its mutations delegate to the data context too; a bridge mutating an
ORM set was testing the bridge.
Those mutations are awaitable now, which is right on its own terms but did NOT
fix what I expected: the counts stay at their seeded values with no error
anywhere, identical before and after. The app's own caveat says the count needs
one connection more than the write, and main only passed because it mutated the
set in place, locally. Those three assertions encode an immediacy this path does
not offer, and whether a test waits for the push is not mine to decide.
STILL FAILING, deliberately left: the three counter assertions above, an
unexplained double seed, and the human wallet-import flow that fails on main
too. setLike.delete is a no-op copied from a branch where nothing needs deleting
before the seed -- true there, false here.
The @data suite writes into one wallet that nothing empties, so every run
leaves more behind. Past some volume the broker stops answering in time and
scenarios die on their step timeout, SILENTLY -- nothing in the console. That
failure looks exactly like a bug in whatever was last edited and gets diagnosed
as one: it happened twice in this session before anybody measured the profile,
which had reached 94MB.
Ported from the sibling branch's validate.ts, which already had the remedy and
the threshold. Placed just before ensureAuth rather than at the top of
BeforeAll: that is the branch which actually opens the profile, and the mock
fallback would pay a broker sign-up for nothing.
Two things the comment records rather than hides. Here the rotation resets the
DATA, not just browser state -- nothing serves a shared wallet file in the @data
path, so the wallet is the broker-side one reached through this profile; the
sibling branch documents the opposite, correctly for itself. And it is not free:
a new wallet comes from driving the real broker's sign-up, so each rotation
abandons one there. FESTIPOD_PROFILE_BLOAT_MB=0 opts out.
MEASURED AND UNRESOLVED: one run leaves 54MB, so a 50MB threshold fires every
time. Either it rises or systematic rotation is accepted.
The contract carried a single event binding, which only worked while Events was
the one vertical on Refine. It now carries one binding per persisted entity,
and each binding says which resource name it lives under and who must be able
to read it.
Visibility is DECLARED, not yet enforced: Festipod's confidentiality is
per-entity while Refine has no notion of it, and the NextGraph provider reads
across every graph it can see. Naming it marks where each backend has to close
that gap rather than pretending it is closed.
meetingPoint is optional for now because its SHEX shape does not exist on this
branch; adding it means regenerating the ORM bindings, which drops the Fp prefix
and is its own step. Friendship is absent on purpose — it has no shape and no
persistence anywhere in this project.
A backend now answers a contract, and each of its four points is there because
the Events vertical ran into it: how its providers are configured (SHEX shape
types here, resources+jsonContext for ActivityPods), what its records look like,
which key Refine addresses a record by versus which id the app references, and
whether its update replaces or patches. Nothing in it is anticipated.
Switching happens at runtime, so nothing resolves at module scope: the current
backend is React state held by RefineProvider, read through useBackend(), and
<Refine> carries its name as key so switching remounts it and drops the cache
of the backend being left. Leaving one loses access to its data, accepted.
Not done, and not guessed: no activitypods backend. Its configuration is
published, its record shape and update semantics are not, and it declares
resources in ActivityStreams while Festipod persists fp: shapes -- as:Event
exists, a meeting point does not. And AuthGate still gates on the NextGraph
session, so a second backend needs its own barrier to be reachable at all.
The @data budgets in connexion.steps.ts came from ng-eventually, which had
already met this: the timeout was in waitForFunction's SECOND slot, where it is
taken as the predicate's argument, so the wait silently ran on the 30s default.
ng-eventually lives with one document per entity already, so it had met this
cost and settled it in two lines: await the seed, and give the wait a budget
that matches what opening five documents takes. The 10 s was calibrated when
seeding meant adding rows to an already-open graph.
This replaces a rewrite of the step, begun before looking there, that would
have had each scenario create its own event. @data is now 14 passed / 1 failed,
the same as main, and the remaining failure is main's too.
Doctrine prescribes a connected seed that writes only what is needed, because
it is linear in documents. Verified safe before applying: no step asserts on a
seeded participation -- a scenario SETS its starting participant count on the
event and reads it back -- and no profile is looked up by name. So three
profiles instead of fourteen, and no participations.
The speed argument for it was REFUTED by measurement: 24 documents took 13.3s,
eight took 20.2s. Main writes 24 in 1.4s on the same accumulated wallet, so
neither document count nor accumulation is what dominates. Main adds into one
already-open graph; this branch opens a document and a fresh subscription per
event, which is what the Refine provider's create does. The cost is in opening
documents, not in writing triples.
Kept anyway, on the justification that survives: it writes less into a wallet
nothing ever empties.
The recovered comment claimed a dev server picks the flag up from the
environment it was built with. It does not: dev and start both serve from src/,
so build.ts's define only reaches the staging bundle the @e2e harness builds.
OFF, the default and the reason the gate exists, needs no plumbing; opting IN
is what is out of reach, and closing that gap means making the app bootstrap
async, which nothing here needs yet.
bootstrapWallet consulted ngUsers.size, which the await for the subscription
lets the broker re-populate — a caller that has just emptied the wallet then
never seeds again. The count now comes from the caller, as the event count
already did: deciding whether to seed reads what the caller sees, writing
needs the resolved set.
Recovered from ng-eventually, which had answers this session went looking for:
autoSeed.ts, whose gate keeps the dev seed from bloating a persistent wallet
over repeated runs (its fixtureSeedEnabled master switch is deliberately NOT
enforced here — adopting it stands down the seeding scenarios and is a product
decision); overlay-local-checkout.ts, which holds the single-@ng-org invariant;
ng: NG in place of typeof NG, and the harness reading the app's own
currentUserId with optional chaining, which together take typecheck from 3
errors to 1.
bun install does not rewrite what is already in node_modules when a version
goes DOWN, which silently left @ng-org/orm and @ng-org/web on versions that
disagree about orm_start_graph's arity — the callback then lands in the config
slot and the SDK rejects it as unserializable. Every dependency measurement
taken this way was meaningless.
Carried over from ng-eventually with it: cucumber invoked through the package
entry rather than the .bin shim, which node --import tsx/esm cannot load under
pnpm; the build:orm paths, which still pointed at a src/shapes that moved to
src/shared/shapes; and the tsconfig fix that DROPS baseUrl instead of silencing
its deprecation — reported as a config error, it made tsc exit without checking
a single file.
The placeholder useShapeSet hands back while a subscription loads refuses
writes, and bootstrapWallet writes its users straight into that set — so the
seed threw on its very first add and the wallet reported nothing persisted.
The refusal is right; issuing the write that early was not.
The previous orm version handed back a writable set immediately for a defined
scope, so a caller never had to wait. Since alpha.21 the set does not exist
until the subscription resolves, so loadTestData now awaits the ORM's own
promise and seeds into the set it resolves with — re-reading ngSet after the
await would hand over the placeholder this closure captured. The dev auto-seed
gates on the same readiness.
@ng-org/orm stopped returning the DeepSignalSet from useShape and now returns
a result object carrying it under data. The bump is forced: the providers
import RdfOrmSubscription, which alpha.18 published as OrmSubscription.
useShapeSet unwraps it in one place and hands back a set, so the call sites
keep the contract they were written against. While the subscription is
loading it yields an empty set that refuses writes rather than dropping
them, matching what the previous version documented.
The harness builds its bundles with a classic script tag, so an unwrapped
bundle publishes every top-level declaration as a global. Bundling Refine
brought in @tanstack/query-core, whose exported MutationObserver class then
shadowed the DOM one for Playwright's injected script, and every
waitForFunction died on defaultMutationOptions. --format=iife keeps the
bundle's declarations to itself.
The harness also mounted the data context outside <Refine> and called the
previous bootstrapWallet signature; it now mirrors the app's provider order
and seeds through the context, so tests and app write events by one path.
Mounts Refine over @ng-org/refine-providers and routes the Events vertical
through it. Known incomplete: the @data harness still mounts the data context
outside <Refine> and calls the old bootstrapWallet signature.
Stopgap staging multi-user sur wallet partagé (cf. brief_2026-06-15_shared-wallet-shim).
Distribution / import du wallet :
- AccessGateScreen : barrière d'accès ON PAR DÉFAUT (désactivable via
globalThis.__FESTIPOD_ACCESS_GATE_DISABLED__ pour tests/dev). Fournit le FICHIER
.ngw + le mot de passe + un guide en 3 étapes (import assisté sur nextgraph.eu —
le broker hébergé n'autorise pas l'import inline pendant l'auth web-app).
- sharedWallet.ts + build.ts : fichier copié en /shared-wallet.ngw, mot de passe gravé.
- Ancien LoginScreen (/login) retiré ; atterrissage post-login -> /home.
- NextGraphContext : dé-piégeage de l'état "connecting" au retour (pageshow/bfcache).
Couche multistore stopgap : storeRegistry, isolation, AccountContext, FestipodDataContext.
Tests e2e multi-navigateur :
- browserPool + world.openBrowser : contextes frais isolés, 2 axes orthogonaux
(nb de navigateurs × modèle de wallet own/shared).
- @humain : parcours humain complet (télécharge -> importe le fichier sur
nextgraph.eu -> Entrer -> pseudo -> accueil).
- Bypass de la barrière pour @e2e via context.addInitScript.
- Convention @wip exclue via cucumber.json.
Docs (concepts) : nextgraph-platform (knowledge_broker-import-constraint,
decision_2026-06-17_assisted-wallet-import), bdd-testing (knowledge_multibrowser-harness).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Couche compte/login + isolation livrées et vérifiées ; couche multi-document
(storeRegistry) livrée derrière FESTIPOD_MULTISTORE/FESTIPOD_STAGING (OFF par
défaut, mono-store reste le défaut), runtime NG à valider sur broker.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Capture the multi-user design exploration as project knowledge + briefs:
- knowledge: NextGraph store types/permissions (+ inbox at protocol, SDK
exposure, local repo path); integration model (iframe, where the verifier
runs, generic JS plumbing, ngd stateful, build-time broker target)
- briefs: multi-store refactor; authorization matrix + query inventory +
derived store partitions; temporary fork to expose the inbox (3 layers:
SDK fork, Coolify self-hosting, Festipod integration; libs via build:ng)
- fix stale @ng-org versions (alpha.11 -> alpha.13) and a broken
decision-record link in data-layer.md
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The @ng-org/* SDK alpha versions are now published on npm. Switch the
package.json dependencies from .ng-tarballs/*.tgz paths to npm
versions, removing the need for a sibling nextgraph-rs build to
install the project. Makes containerized deploys (Coolify) trivial.
- @ng-org/alien-deepsignals: 0.1.2-alpha.11 (unchanged)
- @ng-org/orm: 0.1.2-alpha.15 → 0.1.2-alpha.18
- @ng-org/shex-orm: 0.1.2-alpha.7 → 0.1.2-alpha.8
- @ng-org/web: 0.1.2-alpha.11 → 0.1.2-alpha.13
scripts/build-ng-packages.sh and the .ng-tarballs/ workflow remain
available for local development against an unreleased nextgraph-rs
build (`.gitignore` keeps the directory ignored).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Port modern clean theme (DM Sans, orange accent, app-* CSS classes)
and screen redesigns from festipod-mockups; replace sketchy Ubuntu
theme. New shared components: BottomNav, EventCover, EventMeetingPoints,
Toast, AvatarStack, Tag, RelevanceIcon.
- Restructure from prototyping shell to real mobile web app:
path-based routing (History API), Gallery/DemoMode/PhoneFrame removed,
Storybook setup for screen/component browsing.
- ConnectScreen ported from mockup (QR-based user connection); routed
at /profile/connect, wired from FriendsListScreen.
- Dev-only auto-seed of NG wallet when empty
(gated on NODE_ENV !== 'production'); bootstrapWallet already
self-checks for non-empty ngSet so safe even in race conditions.
- Render-based @ui test infrastructure: happy-dom + LocalDataProvider +
RouterProvider via src/shared/test-harness/renderHelper.tsx, exposed
on the world as renderedDoc. world.hasText/hasField/hasElement prefer
the rendered DOM and fall back to source for backward compatibility.
- Migrate 25 brittle @ui assertions from regex-on-source to DOM
queries; delete implementation-detail tests (showDuplicateWarning,
importableEvents, importedFrom — anti-patterns per the new contract).
Update feature files where the UI changed: "Mes amis" → "Mon réseau",
"Mes événements à venir" → "À venir" on home, Thématique removed
from create-event wizard, etc.
- Path-based @e2e steps (pushState + popstate dispatch) replacing the
legacy "#/demo/…" hash routing tied to the deleted Gallery.
- Add .project/knowledge/test-layer-contracts.md defining the role of
each test layer (@ui = display with seed data + DOM, @data = mutations
through NG broker, @e2e = critical user journeys) with anti-patterns
and migration consequences.
Test status: 75 passed / 71 skipped (explicit "non implémenté")
/ 2 failed (pre-existing @wip on ngSet.delete() NG ORM limitation).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Refactor FestipodDataContext to use only the useShape ORM API:
- Remove ng.sparql_update, sessionPromise, ensureGraphNuri imports
- Use privateNuri (useShape scope) directly as @graph for adds
- Create/join are now synchronous (no async wrapper needed)
- Leave uses ngSet.delete() — known limitation: doesn't persist (@wip)
Add event lifecycle e2e scenarios (cycle-de-vie-evenement.feature):
- Create event via form and verify on home screen
- Created event persists after reconnexion
- Consult event detail from home
- Join an event
- Modify event location and verify
- Leave + persistence tagged @wip (ngSet.delete doesn't persist)
Fix DemoMode external navigation: sync initialScreenId prop changes
to internal state via useEffect (was ignored after first mount).
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
ngSet.delete() updates the local reactive set but does not persist to the
broker. Use ng.sparql_update() with SPARQL DELETE WHERE to remove RDF triples
directly — the broker sends back a GraphOrmUpdate that reactively removes the
item from the ORM set. The two methods must not be combined as they conflict
in the CRDT.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Writes (doc_create, orm_frontend_update) failed with RepoNotFound because
useShape with did:ng:i scope doesn't open individual repos in the verifier's
cache. Switched to did🆖${session.private_store_id} as both scope and
@graph, matching the expense-tracker-rdf pattern. This opens the private
store repo via orm_start_graph, making it available for subsequent writes.
Also adds wallet login step to ensureAuth so the verifier bootstraps repos
from the remote broker into localStorage on first run.
Key changes:
- useShapeWithDefaults accepts storeNuri param (private store NURI)
- FestipodDataContext.useNgData() passes private store scope
- ensureGraphNuri() simplified: reuse existing @graph or private_store_id
- ngBootstrap uses ensureGraphNuri + flushAndWait between ORM adds
- harness-ng.tsx uses private store scope for test bridge shapes
- hooks.ts: wallet creation logs in to bootstrap verifier repos
- E2e steps for data loading and persistence verification
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add @e2e test layer: real app in broker iframe via Playwright
- Fix broker redirect: conditional auto-init only when inside iframe
- Fix seed data flash: empty data during 'connecting' phase
- Fix Gallery button in iframe: explicit navigate instead of history.back
- Add auth e2e feature scenarios and step definitions
- Update docs: bdd-testing, data-layer-testing, data-layer, AGENTS.md
- Add decision record for conditional NG init approach
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Rename steps/frontend/ to steps/ui/ across all modules and shared
- Add data-layer test harness (mock + real broker modes) with Playwright
- Add inscription data-layer steps (@data scenarios)
- Add test auth setup script and browser debug script
- Update docs (architecture, BDD testing, data-layer testing)
- Add ADR for headless wallet creation decision
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add NextGraph data layer with @ng-org/orm, SHEX shapes (Event, UserProfile,
Participation), session management, and FestipodDataContext with dual-mode
operation (connected via NextGraph or local seed data)
- Add BrokerBanner and NgStatus components showing connection status
- Refactor to feature-based architecture: organize code by business domain
(event, user, home, auth, workshop, meeting, notification) instead of
technical layer. Modules only import from shared/, never from each other
- Collocate BDD features and step definitions with their modules: event-specific
steps in event/steps/, user steps in user/steps/, shared generic steps remain
in shared/steps/
- Set up multi-layer BDD structure (frontend/backend/e2e steps per module)
- Add project documentation (AGENTS.md, .project/knowledge/)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- EventDetailScreen: person who shared the event is now "Relayé par"
- ProfileScreen/UserProfileScreen: remove role badges from event lists
- FriendsListScreen: Organisateur/Organisatrice → Relayeur/Relayeuse
- Update event data to transition-themed examples
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Sylvain Duchesne