The FetchReplies job reads a thread's FEP-171b contextHistory before its FEP-7888 context: the posts each Add (or
Forte's plain Create) brought in, read from their own servers. A thread collection read whole keeps its ETag when the
document itself changes with every post (it counts them, or holds them all with no further page); the next read sends
it as If-None-Match and a 304 ends the job. The ETag is kept as sent, since NodeBB's has no quotes and the typed header
drops it. A context naming a post we hold (Forte's first post) is not fetched.
Checked in the pasture: Mastodon 4.7.3 answers the second read 304; NodeBB 4.16's unquoted ETag is kept (23/23 in its
scenario); a Forte thread completes through its replies.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012CzABvBkbcFqoHdmi8b9WB
A root exports one of its personas (a job) as Mastodon's account archive, so other servers' importers read it: the
actor with its public key only, its own posts and boosts with their media, likes, bookmarks and Mastodon's CSV files,
plus PrivaPub's filters, followed hashtags, notification policy, pins, scheduled and located posts; nothing of its root,
its siblings, its keys or anyone's token. A ticket link downloads it, for a week.
An archive (PrivaPub's or Mastodon's) uploaded in pieces is imported into a persona in a job, the parts the root picks,
with progress and a stop. SafeArchive refuses links, escaping paths, duplicates, bombs and oversized items, and reads the
outbox one item at a time. Imported posts are delivered to no one, put in no home and notify nobody, yet show on the
profile, outbox, hashtags and search; back home a post keeps its id, from another actor it gets one of its date and
ImportedFromURI, so importing twice changes nothing. Relationships go through the existing services; located, scheduled
and likes only when asked; followers never.
tools/pasture/scenarios/persona-archive.sh imports mastouser's real Mastodon archive (156 posts, 24 pictures) into a
persona Mastodon follows: Mastodon receives none of it, and a second import changes nothing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
tools/pasture/scenarios/restore.sh backs the pasture up from the administrator's endpoint, then alice_restore deletes a
post and makes another, mastouser follows her, she blocks bob_restore and carol_restore is made; the restore, asked for
from the endpoint, keeps every protective act (19 checks). town.sh renew <peer> signs a peer's town accounts in again,
since a restore ends every session.
It found that a backup listed media files already missing when it was made, so verifying it failed and the restore was
refused: a manifest now lists only the files it holds (refusals are cut to five lines). And a local post made after the
backup now comes back as a deleted row, as a deletion leaves it, so it answers 410 and its id is never given again;
DeletedObject holds remote tombstones only.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
/clientapi/admin/backups (owner decision 2026-10-07, approving these endpoints in production): the list with what runs,
the restore waiting and the last restore's report; back up now; delete; a download for the password, through a ticket
good for ten minutes in the link's path, as one tar whose length is known first and which honours Range (BackupTar);
an upload in pieces of at most 32 MB, each at the offset already received or refused with it, so a broken upload
resumes, read into a backup only when it holds nothing but plain files under one backup's folder; and a restore for
the password and the host typed out. Every call checks the administrator against the database. nginx streams downloads
for an hour and takes upload pieces unbuffered, rate-limited.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
PrivaPub admin restore <id> (and soon the administrator's page) checks the backup (same host, a format and newest
migration this build reads, every hash) and writes restore.json; the running service sees it within seconds and stops,
and the next start restores it in MaintenanceGate, before migrations, indexes and hosted services: a pre-restore backup
taken once, every collection dropped and imported raw with its indexes, the media the live directory lacks brought
back, then the protective merge from the pre-restore backup. Followers and follows are the live ones; blocks, mutes,
domain blocks, reserved names, tombstones, reports, filters and OAuth applications are the union; deletions win;
accounts made since become tombstones and local posts made since answer 410; every session ends.
Each attempt redoes everything; one refused before any change is abandoned and recorded, one failed midway exits 1 for
systemd to retry, and after three it exits 75, which the unit no longer restarts. Commands wait (exit 75) while a
restore is pending. RestoreRecord tells what happened (admin restore --status).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
A backup is a directory <stamp>-<kind> under Backups:Root (/var/lib/privapub/backups, 2770, files 0640), written as
.partial and renamed once whole: a manifest (host, build, newest migration, each collection's count, size, sha256 and
indexes, what was left out and why, the media list), each collection as gzipped canonical Extended JSON read raw, and
hard links to the files of untrashed media rows (copies where a link can't be made). On a replica set every collection
is read in one snapshot session. Never in a backup: the statistics salt, jobs, recovery codes, sessions, the
maintenance lock and the configuration's copy with its SMTP password.
One backup or restore at a time (MaintenanceLock, a heartbeat document), and the janitor purges nothing meanwhile.
BackupScheduler backs up nightly at 03:30 UTC (or at once after missing a night); rotation keeps 7 daily, 4 weekly,
3 pre-deploy and 3 pre-restore backups. CLI: admin backup [--kind] [--db-only], admin backups, admin backup verify;
these run before migrations, so the deploy's own pre-deploy backup, which replaces mongodump, is of the database as the
live build left it. EntityMaps.Warm runs once under a lock, since test hosts now boot side by side.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Owner decision 2026-10-07: production's mongod becomes a one-member replica set (rs0), so that a backup can read every
collection at one instant (a snapshot read session), which a standalone mongod can't. The unit runs mongod with
--replSet rs0 and a 990 MB oplog; setup.sh converts it once, idempotently (PrivaPub stopped, mongod restarted,
rs.initiate, the primary awaited, PrivaPub started); every connection string says directConnection=true, which works
against the standalone too, so this code can deploy before the conversion. The CI's throwaway mongod and the pasture's
are replica sets as well, so the snapshot path is what the tests exercise; MongoTopology tells which a mongod is, and
TopologyTests holds the test mongod to it. The suite passes on it (906).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Anyone could mint signed proxy URLs (a remote account changes its icon, an anonymous lookup returns the URL), and each
anonymous request held up to 40 MB in memory; the cache grew without bound between hourly trims; two clients asking
for the same new file downloaded it twice and wrote over each other in place, so a reader could get half a file with a
7-day cache header; a file over the limit was downloaded twice on every request; a failed fetch, a 404, was cached by
browsers for a week; cached media of a server suspended later were still served, and RejectMedia skipped avatars,
emoji, covers, video variants, link cards and remote edits; /clientapi/group/members returned remote pictures raw.
Now a download is shared by everyone asking at once, streamed into a .part file and renamed into place
(FederationHttp.DownloadMedia copies bounded, never into memory), at most eight at a time; a file too big to cache is
remembered for an hour and only streamed, a failure for five minutes; the cache's size is counted as it grows and
trimmed as soon as it passes the cap; a cached file is opened before it is answered; browsers may cache only a
success; nothing of a suspended server, or of one whose media are rejected, is proxied (everything remote a client
sees goes through the proxy, so that covers every kind), and blocking one purges its cache; the proxy has its own rate
limit per client address; group members' pictures are proxied; the proxy's key is loaded once, the oldest if two
were made. This changes what PrivaPub serves its clients, not what it sends to other servers.
Tests: clients asking at once share one download, a failure isn't cached by browsers, an over-limit file is fetched
three times for two requests instead of four, a blocked server's media are refused and its cache purged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
An upload went straight to libvips: whatever loader recognised the bytes ran (an SVG sent as image/png was rasterised),
nothing bounded how many pixels it would decode to (a small PNG could decode to gigabytes, three times over), a GIF
was loaded frame by frame and never resized, and all of it ran inside the request with nothing limiting how many at
once. A GIF was typed gifv but stayed a .gif, which a gifv player can't play; its metadata was kept; colours lost
their ICC profile without being converted; HEIC was advertised but the bundled libvips can't decode it.
Now:
- only libvips' JPEG, PNG, GIF, WebP and HEIF loaders ever run on an upload (every other loader is blocked);
- the header alone says how big an image would decode, refused above Media:MaxPixels (40 MP) or MaxFrames;
- a still image is shrunk on load, turned by its orientation and brought into sRGB (thumbnail), then written without
metadata, a profile picture the same way;
- an animated GIF becomes a looping silent H.264 mp4 typed gifv, as on Mastodon (PostMedia.Kind keeps it a gifv),
and a remote GIF is an image;
- processing runs Media:Concurrency at a time, and uploads have their own rate limit per credential;
- HEIC and HEIF are no longer offered.
Tests: only the upload formats load, the header tells the size, an SVG posing as a PNG and an image too large are
refused before decoding, an animated GIF becomes a gifv and a still one an image, HEIC isn't advertised. The media
scenarios against the pasture (GoToSocial, Mastodon, Misskey, Akkoma, Pixelfed, Smithereen, Vernissage, Castopod,
PeerTube) pass: 329 checks.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Deleting a post, editing media out, replacing an avatar or a header, and removing a whole root deleted no file: every
one stayed on disk and publicly served from /media/files with a year-long immutable cache, its row orphaned. Now every
upload is a row, profile pictures too (Kind avatar or header, ProfileOfAvatarId), and each of those acts trashes what
it held, as does an upload never posted for a day and a dropped scheduled post. A trashed row is marked in one
conditional update (an upload attached meanwhile is left alone), its files move into media-trash, beside the media
root and outside what /media/files serves, and the janitor deletes them a day later. Nothing is deleted for looking
unused.
Along the way: a profile picture that isn't an image, or can't be read, answers 422 instead of being silently ignored
with a 200; a removed root's scheduled posts are dropped, so nothing of it publishes later; media rows get indexes
(they had none), and the janitor's first pass comes five minutes after boot instead of an hour.
`PrivaPub admin media audit [--fix]` compares the disk with the database. With --fix (as www-data) it gives the
pictures personas show today a row, and trashes media of deleted posts or personas, rows whose files are missing, and
files nothing holds: the leftovers of every deletion until now. MediaLifecycleTests covers each act, that the trash is
never served, and the audit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
float.TryParse takes "NaN" and "Infinity", and Math.Clamp keeps a NaN, so `focus=NaN,NaN` on an upload or a media PUT
was stored as given. Once the media was posted, every status and timeline response holding the post, and the Note
and its delivery, failed to serialise, for every viewer. FocalPoint.Parse takes finite numbers only (anything else is
ignored, as an unreadable focus was); the mapper and the renderer leave out a focus that isn't sound, and migration
_016 removes the ones stored before, from the uploads and from the copy each post keeps. What PrivaPub sends is
unchanged for any focus that could be sent before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
A lease lasted two minutes and was never renewed, so the reaper gave any longer job to a second worker while the first
still ran it, and both finished it. The worker now renews the lease every third of its length while the handler runs;
a lease found taken (reaped and leased again) cancels the handler. Each lease carries its own owner stamp, since every
worker of a process shares one name, and Finish only counts for the lease it was given. Media processing and persona
archives will run longer than two minutes. JobQueueTests: a job three times its lease runs once with the reaper finding
nothing, and a stolen lease stops its handler and drops its outcome.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Lemmy takes a report only from a person or a service, about one post or comment, addressed to its community, and it
answered PrivaPub's Flag (the instance actor's, an Application, with no `to` and the account and posts as its object)
400. A report of a post or comment in a community on a server whose NodeInfo names Lemmy now leaves from
`privapub_reports`, a Service with its own key that names nobody: one Flag per post, `to` the community (its own
audience, else its thread's), with the persona's words, or the category, in `summary` and `content`, sent to the
community's inbox. This is the second exception to "a server's software is for display" (owner decision 2026-10-06,
`ReportService.ServiceReportTakers`). Every other server keeps the instance actor's report. An account alone is not
reported to Lemmy, which takes no such report, and `forwarded` now says whether anything left.
The reporter is read unsigned in SecureMode and answers WebFinger like the instance actor. Nobody follows or mentions
it, the Mastodon API has no account for it, and a migration reserves its name. Checked live: Lemmy 1.0 and 0.19 keep the
reports of a thread and of a comment, with alice's words, from "Reports from privapub.test", and none names her (69
checks). A sweep of every scenario with this and the next commit: 876 checks pass; Ghost's Network feed listed alice's
post too late once, and Ghost passes alone.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Mastodon 4.3's policy was a stub that accepted everything. It is now kept per persona: notifications from accounts it
does not follow, accounts that do not follow it (or only for three days), accounts newer than 30 days, private mentions
it did not ask for and silenced accounts are accepted, filtered or dropped. Filtered ones stay out of every list and
count unless asked for, gathered in a request per account; letting a request in lets that account in for good,
dismissing it deletes them. Everything is accepted until the persona chooses.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
FEP-521a and FEP-8b32. Every persona has an Ed25519 key of its own (Avatar.SigningKey; migration 014 gives the earlier
ones theirs), named in its actor's assertionMethod as a Multikey, the terms defined in the actor's own context. A
persona's activity going to a relay carries an eddsa-jcs-2022 proof (JSON canonicalised by RFC 8785, Jcs), so what
Activity-Relay forwards reaches Mastodon, which verifies it with its own code. Nothing else carries one: Mitra takes a
proof over the HTTP signature and refuses one by a key it has not read, without reading the actor again. Received: an
actor's own Multikeys are kept, and a forwarded activity whose proof one of them verifies is taken as it came instead of
being read again from its origin.
Discovery: WebFinger for the server's origin links its instance actor (FEP-d556), NodeInfo links it as the application
actor (FEP-2677), and actors name RFC 9421 under implements (FEP-844e).
Checked live: relay 16 (Activity-Relay's forward of alice's post reaches Mastodon), Mitra, GoToSocial and Mastodon
unchanged (165 in all).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Owner decision of 2026-10-06 (G-0009, FEP-400e). A persona's actor names its wall (…/graffiti, sm:wall) and, in
Smithereen's privacySettings, that its followers may write on it. A public post that is not a reply, sent with the wall
as its target by an account following the persona, is hosted: the persona is notified, it reaches the persona's and its
followers' homes, and the followers' servers and the author's are told with Add{Note}. The persona deletes it with
DELETE /api/v1/statuses/:id, which sends Remove{Note}; Smithereen deletes the post then. The wall lists the persona's
public posts that start a thread and what was written on it.
Elsewhere: an account's Add{Note} on its own wall shows the post to its followers here as on that wall (privapub.wall on
the status), and its Remove takes it away. An Add or Remove naming a collection of the account's own server PrivaPub
does not know has its document read again first, at most hourly: accounts kept before walls were read had none.
Checked live: Smithereen 40 checks (G-0009 closed); GoToSocial and Mastodon unchanged (121).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Federation:Relays names relays by their actor (or inbox) address; the instance actor follows Public at each, as Mastodon
subscribes, a minute after start and every six hours (asked again a day after no answer or a refusal, undone when a
relay is no longer named). What an accepted relay passes on comes to the federated timeline and nobody's home: a public
post it forwards (Activity-Relay), read again from its origin like any forwarded post, and a post it announces
(aode-relay), kept as its author's and never as the relay's boost. Nothing of a persona's is sent to a relay; sending
public posts there waits for the owner.
The pasture gains both relays (peers/relay.sh, peers/aoderelay.sh) and scenarios/relay.sh, 13 checks. The village
backlog is clean: 2574 checks pass, one known gap (Misskey's).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
An account elsewhere that pins or unpins one of its posts (Add or Remove on its `featured`) now shows those pins on its
profile here (`pinned=true`), in its order and its public posts only; a community's announce of a moderator's Add does
the same for the community. The `featured` collection itself is read with the account's counts, at most once a day, so
pins made before PrivaPub ever saw an account show too. Any other target (Smithereen's wall, a community's moderators)
is dropped. A persona's pin and unpin go to the post's audience as Add and Remove on /trophies, as Mastodon sends them.
Checked live against Mastodon (scenarios/pins.sh, 8 checks). The town's checker learnt three peer rules from the
village: Misskey and Sharkey keep a forwarded reply only with its author's LD signature (only Mastodon signs), they
count no renote by a bot, and Mastodon never sees a Lemmy vote on a post in a community. The village of 2026-10-05
checks clean, 2454 of 2454.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
tools/pasture/flood/flood.cs answers as twenty fake servers (flood1..20.test)
and sends signed Creates, Likes and Follows at a set rate; load.sh measures
the answers, the queue's wait and processing times, its drain and a persona's
home timeline meanwhile (docs/LOAD.md has the method and the runs).
What the runs found:
- Every unique index was partial on $type: "string", which MongoDB never uses
for an equality lookup, so every post by ObjectURI, actor by ActorURI,
deleted object, domain block, remote instance and the rest was a
collection scan (280 ms a post lookup at 30 000 posts). They are partial on
$gt: "" now, which an equality on a string implies; MongoDB.Entities
rebuilds them in place at the next start.
- Two inbox workers capped intake near 110 activities a second:
Federation:InboxConcurrency and DeliveryConcurrency (default 8) set them.
- The indexes the plan listed as missing: a post's boosts and replies, a
persona's boosts, who follows an actor, timeline rows by author, a post's
likes and pins.
At 300 activities a second (200 let through, the rest 429 by the per-origin
limit) the queue wait went from 29 s to 6 ms at p50; with the limits lifted
PrivaPub processes about 900 a second, each in under 10 ms, and the home
timeline stays under 20 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Two owner decisions of 2026-10-05, recorded in the roadmap.
Replies passed on ("the fediverse is broken without"): a public or unlisted
reply from another server to a persona's public, unlisted or followers-only
post goes on to the persona's followers as its author's server sent it, as
Mastodon forwards it, never to the replier's own server, never for a
local-only or group post; its edit and deletion follow. Only an activity its
own actor delivered is passed on (Arrival.Raw), so nothing forwarded is
forwarded again. The town checks it as relay.reply cells (specs/relay-five:
882 checks pass); Mastodon takes a passed-on activity only with an LD
signature, which GoToSocial and Akkoma don't add, and the checker knows it.
Events: a persona joins another server's event with a Join and leaves it with
a Leave, both to the organiser only, through
POST /api/privapub/v1/statuses/:id/join|leave; the organiser's Accept or
Reject is routed by our join id and shows as privapub.event.participation.
Events by invitation or taken on another site are refused before anything
is sent. Mobilizon's scenario joins and leaves an event (28 checks) and keeps
one for decePubClient's e2e.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Mobilizon sends its NodeInfo as `application/json; profile=http://…#` with the URL unquoted, which .NET cannot parse,
so the document was refused for its content type and the server never described; and it names its software
"Mobilizon" where NodeInfo wants lower case. The media type is now read from the raw header when the parsed one is
missing, and software names are lowercased, so one software is counted under one name.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Friendica's activity ids are uniqid(): a short prefix and the microsecond. Two of its processes answering two follows at
once gave both Accepts one id, and PrivaPub, queueing each inbox activity once per id, dropped the second as a copy:
that follow stayed pending on our side while Friendica counted the persona as a follower (seen in the town's Friendica
pair). An id that comes back carrying another type, actor or object is now queued apart; a true copy is still dropped.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
A thread's server passes on what happens in it, signed with its own key: Mastodon forwards the replies to its
accounts' posts and their deletions, Friendica every activity in its threads. PrivaPub answered them 401, which also
tells a sender its signature failed. Now they get 202 and nothing in them is believed: a forwarded Create or Update is
taken as its object reads at the actor's origin, a Delete of a public or unlisted copy once that origin answers 404 or
410 (RemoteActorService.IsGone; FederationHttp remembers the status of a refusal), anything else is let go, and our own
activities coming back are ignored. A forwarded copy has its own dedupe key, so one that failed never hides the
author's own delivery.
A reply in the thread of someone followed here is kept, as Mastodon keeps them. Mastodon delivers a reply to the
followers of the account it answers; PrivaPub dropped those as unaddressed, which the pasture showed: the outsider's
reply its Mastodon scenario said was never delivered had been, and was thrown away.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
A delivery signed with RFC 9421 was counted under "other": its algorithm
(rsa-v1_5-sha256) has an underscore, which a signature name could not
hold. It is now counted as rfc9421:rsa-v1_5-sha256, as WordPress's are.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
/api/v1/conversations answered one page, never unread, its read endpoint
did nothing, and DELETE was missing; each conversation cost a query per
member. Now each conversation keeps its newest post (DmGroup.LastPostId,
set as posts arrive, learnt once by migration _013) and pages by it as
Mastodon does, and each persona's ConversationState holds what it read and
what it took off its list:
- unread when someone else wrote last, after what the persona read;
- read marks it so, and writing in a conversation reads it;
- DELETE takes it off the list until a newer message brings it back.
The list reads its states, newest posts, members and accounts in a few
queries per page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
The context of a remote post showed only what PrivaPub happened to hold:
replies from servers nobody here follows were never seen, and only the
ancestors were ever fetched. Now a persona opening a public remote thread
queues FetchReplies for the post and its root, at most once an hour each.
The job reads the thread's own collection first (FEP-7888 `context`, which
Mastodon 4.5+ serves with every reply at any depth; posts or, as FEP-f228
allows, the activities that made them), and otherwise the post's `replies`
(PeerTube's `comments`) and the replies' own, two levels down. At most 5
pages and 100 posts a job, signed by the instance actor, never a persona;
each post is fetched from its own origin and stored through StoreContext,
so only public and unlisted ones are kept.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Two ledger flushes that upsert the same new day (or touch the same new
instance) at once made one of them fail on the unique key, losing its
counts; MongoDB retries that only for single-document updates. The loser
now tries again and adds to the document the winner made. Seen as a rare
failure of LedgerOverHttpTests in full runs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
/api/v2/search now searches statuses by words for a signed-in persona, in
Mastodon's scope: what it wrote, boosted, favourited, bookmarked or was
named in, and public posts of authors who let themselves be indexed
(indexable, off by default). Newest first, only what the persona may see,
through a text index over the posts' words in every language alike.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
tags/:name shows the tag's last seven days in public posts PrivaPub holds
(uses and authors, as Mastodon gives them) and whether the persona follows
it; follow, unfollow and followed_tags replace the stubs. A public post that
is neither a boost nor a reply comes, as it arrives, to the homes of those
following one of its tags. Nothing is fetched for a followed tag and no
other server hears of it. Posts are indexed by tag.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Mastodon's v2 filters replace the empty stubs: a filter's title, contexts,
action (warn, hide, blur) and expiry, its keywords (whole words or not,
taken as JSON objects, listed or numbered form fields, with id and _destroy
on update) and its statuses, each with their own endpoints; the v1 API is
the same filters seen keyword by keyword. Every status a persona reads
carries the filters it matches in `filtered` (a boost as what it boosts),
matched as Mastodon matches: warning, title, text, poll options and media
descriptions. Clients apply context and action. Filters never federate, and
go with a deleted persona.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Mastodon's lists replace the empty stubs: CRUD, members (only accounts the
persona follows; a follow that ends takes its memberships with it),
accounts/:id/lists, and timelines/list/:id from the persona's home entries
with the replies policy (followed, list, none; self-replies and replies to
the persona always). An exclusive list's members stay out of home. Lists
never federate, and go with a deleted persona.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
Inbound Block was dropped as an unknown type (the pasture's last
Mastodon expected failure, G-0002). Now, as Mastodon does it: the
follows between the blocker and the persona end here too (nothing is
sent back; the blocker already ended its side), the blocker's posts and
notifications are hidden from the persona and kept out of its home, the
persona's posts are no longer addressed to the blocker by mention or
reply, and the relationship says blocked_by. Undo{Block} lifts it. The
block is kept in BlockedBy, unique per persona and blocker.
The Mastodon scenario checks blocked_by instead of expecting a failure.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
PrivaPub now finds CDNs three ways, best first: the address ranges the
CDNs publish (Cloudflare, Fastly, Amazon CloudFront, Bunny, Gcore,
Imperva), downloaded daily by CdnUpdater and kept in CdnRangeSet; the
CDN's fingerprint in the responses it already gets from a server
(EdgeHintsHandler on the federation client); and the networks that carry
only a CDN. The fixed ASN list is gone; ASNs shared with plain hosting
(AWS, DataPacket) no longer hide a server. A server's Geo records the
CDN, its domain and how it was found, and weekly snapshots now keep the
city and coordinates too.
Servers through time (ServerPlaces): /instances/:host/history lists a
server's weekly snapshots, a CDN-fronted server's geo names the CDN's
domain and where the server was before it (before_cdn), and
/api/privapub/v1/cdns and /cdns/:domain group servers by CDN with week
by week who joined and who left. Owner decisions recorded in ROADMAP.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
The public projection of a server's place showed only the country for
servers reporting fewer than 10 users. The owner never decided that
threshold: every located server now shows its city, coordinates (0.1°)
and network, and a CDN-fronted one still shows only its CDN, since the
address reached is the CDN's edge. Statistics:PublicCityMinUsers is
gone; the ROADMAP decision on server locations, CLAUDE.md and the
/stargazing explainer are corrected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
/api/privapub/v1/instances/:host gains `geo`, the public projection of a
server's place already decided for public server locations
(PublicGeo.Project): city, coordinates and network for servers reporting
at least 10 users and not behind a CDN, the country otherwise, the CDN's
name for a CDN-fronted one, with DB-IP's attribution. `?host[]=`
answers up to 40 servers at once, and this server describes itself:
its host's address located once a day (SelfLocation), or
Statistics:Geo:Self when the owner sets it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LsXgEaXee4GCU1hwYgPJXw
- A community's announces resolve, as FEDERATION.md and ROADMAP always said Announce ids do. GroupDistributor keeps
each one it sends (GroupAnnouncement, unique by id). /grunts serves it while the post it carries is shown and 410
after, and also serves the announce-{postId} ids the group outbox lists, which now keep a stable `published`
instead of the time of the fetch.
- A persona's boost points at the boosted post: its `url` in the Mastodon API is the boosted post's page, and a browser
following the boost's id is redirected there instead of getting JSON.
- /tags/{tag}, where every Hashtag link we send points, is now a public page of this server's public posts with that
tag, with the same strict CSP and noindex as the profile pages.
- Grouped notifications (/api/v2/notifications, its unread count, a group, its accounts and dismiss). We advertise
api_versions.mastodon = 7 so clients show quotes, and clients that trust it call these; they answered 404. Likes and
boosts of one post group together, as do follows within an hour. The version string stays 4.2.0 until streaming
and Web Push exist.
- A remote account's follower, following and post counts are what its server publishes. AccountCountsJob reads its
collections' totalItems from its own origin, signed by the instance actor, at most daily and only after the account
was fetched, never when someone looks. They used to be 0.
- The other ids that do not resolve are documented as such: Update, Delete, EmojiReact, QuoteRequest and its answers,
Flag, Ignore and poll votes.
676 tests pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
Owner decision 2026-10-04: fix the mismatches and every other mismatch of the same kind.
- One counting rule (Domain/Privacy/Counted), Mastodon's. It is used for a persona's statuses_count, its outbox
totalItems, NodeInfo localPosts and the instance status_count, which used to count four different things. It
counts every post that is neither deleted nor a DM, boosts included, and circle and located posts too (owner
decision). A group's count includes its remote members' posts.
- Users. Personas of banned or deleted roots no longer count, and are not found in search. NodeInfo now gives
activeMonth and activeHalfyear, and the v2 instance gives active_month instead of a constant 0.
- replies_count counts only public and unlisted replies, so it no longer tells anyone that a private reply exists.
Migration _012 recounts it.
- A remote account that deletes itself takes everything out of every count (GoneActors): its likes, downvotes,
reactions and poll votes go and their counters come back, as do its boosts', replies' and quotes' counts, and its
notifications. Lookups, account lists, search and favourited_by no longer show it. Migration _012 applies this to
accounts already gone.
- Deleting a post also deletes its pins and the local boosts of it.
- /stalking gives the same total as following_count. Members are still never listed, and hide_collections is now
always true, since the setting never did anything.
- Joining a community by invitation is following it, so /flock and /groupies agree; leaving unfollows.
- Search. Anyone may search, as on Mastodon; resolve and offset need a sign-in, offset pages, and deleted accounts
are never found.
- notifications/unread_count counts what the list shows, and the owner's follower and following lists page with
Link.
- The instance API advertises what is enforced:
- max_characters, now enforced with a 422;
- max_pinned_statuses = MaxPins;
- the media types and limits MediaService and MediaOptions accept;
- PollService's limits;
- the configured languages;
- no streaming URL until streaming exists.
domain_count counts the servers we have exchanged with; which ones stays unpublished (peers is empty).
- Routes Mastodon answers now answer instead of 404:
- directory, tags/{name}, timelines/link and identity_proofs;
- instance/languages, translation_languages, domain_blocks and privacy_policy;
- the v1 and v2 notification policy, and notification requests.
Also, from phase 3: a recovered password ends /clientapi sessions through a per-root SessionStamp claim instead of
comparing the JWT's whole-second nbf with the change time. That comparison let a token issued in the same second
survive, which made a test flaky.
671 tests pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
Owner decision 2026-10-04: fix the account privacy findings.
- Sign-in. Every failure answers "That username and password do not match." after the same work: an unknown login
is hashed against a decoy, and the comparison is constant-time. "Banned" is told only to someone who gave the right
password. This covers /clientapi/user/login, /invitation/login and /oauth/login.
- Recovery.
- Every request answers the same sentence and queues a SendRecovery job, whether or not the account exists or has an
email. The lookup, the code and SMTP move to RecoveryJob, so neither the answer nor its timing says anything.
- Codes are kept only as a SHA-256 hash, for one hour. Migration _011 drops the plaintext ones, which never expired.
- A recovered password ends every session of the root. RootSessions sets CredentialsChangedAt, which JwtEvents
checks against the JWT's issue time, now stamped as nbf, and revokes each persona's OAuth tokens and authorizations.
- Deleting a root (RootRemoval: the admin route, or the restored self-delete at /clientapi/user/delete, which asks for
the password).
- Its sessions end.
- Each persona and each group it owns sends Delete{Actor} to its followers, its members and the accounts it follows.
- The personas' posts are emptied.
- /peasants/{name} answers 410 with a Tombstone (formerType Person or Group), as do its inbox and WebFinger, through
LocalActorService.Gone. The names stay reserved.
- The root keeps only a unique `deleted-{id}` name; the second deletion on an instance used to collide on
"Deleted user".
Also, from phase 2's pasture: GoToSocial files a circle post like a DM and shows it only to accounts it mentions. Each
member's copy, and a member's refetch, now also mentions that member silently. The GoToSocial scenario checks circle
posts in conversations, like DMs, and they pass there now, as on Mastodon.
657 tests pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
Owner decisions (2026-10-04, recorded in docs/ROADMAP.md): production runs everything that is built, and nothing waits
on a person running a command.
- Geolocation updates itself. GeoUpdater, a hosted service, checks daily whether each DB-IP Lite database was built this
month. If not, it fetches this month's, or last month's early in the month. It installs a file only once it opens as
the right kind of database, then swaps it in atomically, and the locator reloads at once. Lookups now run under the
lock, so a reload can no longer dispose a reader mid-lookup. The systemd timer, its script and their setup.sh lines
are gone: the root step they needed never happened, and none is needed now. /stargazing names the database in use.
- The admin CLI runs after the app is built, with every service and nothing started.
- `create-root <login> [--admin]` takes the password on stdin; it is how the first login is made while sign-up is
closed.
- `smoke <persona>` keeps the root `deploy-smoke` and an undiscoverable persona, and gives the root a new password
on every run.
- The deploy signs in as @thepra. It runs the CLI, gets a token through the real OAuth flow (tools/smoke/oauth.sh,
moved out of the pasture's privapub_token, which now uses it), checks the signed-in API and that @thepra is
undiscoverable, then revokes the token. PRIVAPUB_SMOKE_TOKEN is gone.
- The deploy also fails when:
- NodeInfo and the instance API disagree about registrations;
- /stargazing does not say the crawler is on;
- the geolocation databases are missing or more than 40 days old.
- The crawler is on in production, seeded with ten large servers of different kinds. FEDERATION.md now describes it
and how to opt out.
- One registrations switch (Registrations:Mode, default Invitations; Open in tests and the pasture). It is read by
open sign-up (403 when closed), NodeInfo `openRegistrations`, and v1 and v2 of the instance API, so they can no longer
disagree. Before, NodeInfo said open and the instance API said closed. Group invitations always work, so
invites_enabled is true.
- A persona edit through /clientapi no longer resets what the Mastodon API set (discoverable, locked, quote policy…):
the theme is merged into the settings instead of replacing them.
650 tests pass. The deploy's smoke step was rehearsed against the pasture's PrivaPub.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
Sharkey 2025.4.7 is the Misskey peer under another name, image, database, Redis db and
home. Its scenario is Misskey's plus edits both ways and the FEP-e232 quote tag: 40 checks
pass.
Akkoma 3.20.1 publishes no image, so tools/pasture/images/akkoma installs its OTP
release, pinned by checksum, and appends Caddy's CA to the CA bundles the release ships.
Its scenario has 44 checks, which pass three runs in a row:
- follows, posts, CW, followers-only posts and the published time;
- replies, likes, boosts and their undos;
- EmojiReact both ways and withdrawn;
- DMs, polls, quotes, media, edits with history and deletes, both ways;
- unfollow, block, unblock and statistics.
The two bugs, both fixed:
- Followers-only posts arrived as DMs on Pleroma and Akkoma. They call a post private
only if an address in `to` contains "/followers" or its cc is not empty. Ours is
/groupies, and a post mentioning nobody had an empty cc. The followers collection is now
named in cc as well, which tells nobody anything new.
- Akkoma's open polls showed as ended and refused votes. Akkoma carries an open poll's
end in `closed` and sends no `endTime`. A `closed` in the future is now read as the end.
Neither of these is a PrivaPub bug:
- Akkoma's Linkify never takes @user@host.test for a mention, so its DM addresses alice
with to[].
- Its API never reports a remote blocker as blocked_by, so the block is read from its
database.
Also in this commit:
- The rate limits are configuration (RateLimits: AccountsPerMinute, InboxBurst,
InboxPerTenSeconds), with the old values as defaults. The pasture raises the accounts
limit, which back-to-back runs from one address had hit.
- A new Lemmy never sends what it queued for a server before it started that server's
send worker, so the scenario waits for the worker before its first follow.
All six peers in one clean pass: GoToSocial 54 (+1 expected), Mastodon 49 (+2), Misskey
35, Sharkey 40, Akkoma 44, and Lemmy 20 (+3) once the worker wait was added. 642 tests
pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
Off by default (Statistics:Crawler:Enabled), as the owner decided. When it is on:
- CrawlPlan runs hourly, from StatisticsSchedule. It inserts the configured seeds and
queues up to HostsPerHour servers whose last visit is older than RevisitDays, are not
paused by the breaker and are not domain-blocked, spread across the hour.
- CrawlInstance visits one server at a time as
"PrivaPub-Stargazer/<ref> (+<base>/stargazing)":
- it reads robots.txt (RFC 9309: its own group first, then PrivaPub, then *; longest
rule wins; a 4xx allows everything; a 5xx or no answer keeps it out);
- it describes servers that only crawling ever found, through
InstanceDescriber.Describe with robots.txt as the path filter;
- it reads /api/v1/instance/peers through the new GetStringArray, which keeps what it
read from the first 1 MB instead of refusing a large list;
- it adds the names a server could ever be reached at as "crawled": DNS only,
punycode, no addresses, ports or hidden services, and the reserved test names only on
a test network. Never more than MaxNewHostsPerCrawl per visit or MaxHosts in all, and
never over a touched server.
- It reads nothing but robots.txt, NodeInfo, the instance API and the peers list.
IFederationHttp.GetText serves robots.txt, and HttpScope.Crawl carries the
User-Agent.
- /stargazing explains all this and how to keep the crawler out, says whether it is on,
and credits DB-IP. GET /clientapi/admin/statistics/crawler shows the frontier.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
- Touches: the ledger marks a server as touched when it sends us a verified activity, when
we exchange activities with it, or when we read its actors, keys, objects or WebFinger.
It upserts RemoteInstance.Seen, FirstSeenAt and LastSeenAt at most hourly per server, and
queues one DescribeInstance a week with the same dedupe key ObjectRecords uses. Suspended
servers and pages behind link previews are never described. Migration _010 marks the
servers already known as touched, with their dates.
- InstanceDescriber.Describe(host, crawled, allowed) reads:
- NodeInfo 2.2/2.1/2.0, now with its published user counts, posts, comments,
description, languages and schema version;
- for software with a Mastodon API, /api/v2/instance falling back to v1: title,
languages, registration mode, character limit, API version, source URL.
It never keeps a contact as a field; the raw document is kept for the admin only. It
locates the server from the address our connection reached (DB-IP Lite city and ASN, the
CDN named when fronted) and writes a RemoteInstanceSnapshot per ISO week, unreachable
weeks included. A crawled server is upserted as crawled only on insert, so it never
downgrades a touched one, and robots.txt can deny any path.
- PublicGeo.Project is the only public form of a location: a CDN-fronted server shows its
CDN only, a server reporting at least ten users shows its city, coordinates and network,
any other only its country.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
- IConnectedAddresses remembers which address each host's last connection reached, set in
SafeHttpHandlerFactory's connect callback. That is once per pooled connection, with no
second DNS lookup, and the address is never stored.
- IGeoLocator / DbIpLocator reads the offline DB-IP Lite city and ASN databases (MaxMind
.mmdb, via MaxMind.Db). It maps memory, swaps to new files within ten minutes, rounds
coordinates to one decimal, never looks up a private address, and answers nothing
when the files are missing. CdnNetworks names the CDNs whose edge addresses say nothing
about where a server is.
- deploy/max/geo-update.sh fetches this or last month's databases, checks them and swaps
them in atomically. The privapub-geo timer runs it monthly as www-data, and setup.sh
installs the directory, the script, the units and a first download.
Describing servers will use these in M8.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
A RollupDay job folds each finished day's InteractionEvents into one InstanceDay per server:
- Counters for the admin, keyed channel:activity:object:outcome:reason, plus signature
schemes, audiences, local kinds and features;
- PublicCounters, everything a public page may ever read:
- inbound and outbound activities from an allowlist, on public, unlisted or unaddressed
traffic only, with the outcome collapsed (accepted or dropped, delivered or failed);
- no reasons, no Flag or Block;
- features of public objects;
- health:ok or health:failed from reachability (a 4xx means the server answered);
- latency, wait and byte histograms, and the number of distinct accounts from that day's
hashes.
Re-running a day replaces it and keeps the live Reads counters. The day's salt is then
deleted, so its hashes can never be recomputed, and the next day is queued.
StatisticsSchedule plans today's rollup every hour and catches up any of the last seven
days that have events but no rollup. Waits round up into their bucket.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
- A fetched ObjectRecord no longer wears the signature, key, signed headers and @context
of the activity that caused the fetch, and its ReceivedAt is the fetch's own time. Its
activity fields now name the trigger. Provenance answers signature null and
fetchedBy "instance-actor". Migration _008 clears the old fetched records.
- ObjectRecords store their ObjectFeatures extensions and context namespaces (migration
_009 fills older records in batches), so statistics can count them. Provenance reads the
stored lists.
- ForeignAvatar.Features records what an actor document uses (ActorFeatures): featured,
shared inbox, FEP-521a, FEP-8b32, identity proofs, Misskey fields, indexable,
undiscoverable, locked, key size, and more.
- RemoteEdits.Apply and RemoteDeletes.Remove are shared by the Update, Delete and Announce
handlers. A community-wrapped Update is now an edit only when newer, refreshes polls
and media otherwise, revises the ObjectRecord and re-resolves quotes. A
community-wrapped Delete now tombstones the object, removes its ObjectRecord, reblogs
and timeline rows, and lowers the reply count. Any deleted quoting post lowers the
quoted post's quote count.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
- HttpScope (AsyncLocal) tags each outbound request with a purpose and a trigger:
- purpose is set by the caller: actor, key, object, webfinger, context, nodeinfo;
- trigger is set by the job kind, by "verify" during inbox verification, or defaults
to "request".
- FederationHttp records every JSON, media and stream fetch: status, time, bytes, hops,
and an outcome of ok, refused or failed, with a reason: disallowed, remembered,
bad-redirect, too-many-redirects, content-type, too-large, bad-json, private-address,
timeout, network, or the status. A fetch a reader caused (trigger "request") is only
counted per server per day.
- Link previews record a 'preview' event: card, no-card or failed.
- The media proxy counts cache hits.
- TrafficMeter counts the client API per endpoint group, method and status class. It
counts our served documents (actor, outbox, collection, object, activity, licence,
webfinger, nodeinfo) by kind, status and whether signed, per day and never per server,
and never names a circle's collections.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
Arrival carries a verdict that handlers set with one line before an existing return
(Arrival.Drop, Reject, Accept, About), so no handler signature changes. InboxProcessor
records it as an 'in' event, with the time taken, the wait since the inbox accepted it, the
attempt, the audience (from the post's visibility, or else the activity's addressing), the
local actor kind, the object's age for updates, deletes and reactions, and the FEP features
of a delivered object. It also records types nothing handles (unknown-type), actors that
cannot be loaded (deferred) and handler failures.
Drop reasons: fetch-failed, unparseable, misattributed, duplicate, deleted, not-addressed,
not-followed, not-public, not-visible, not-deleted, unknown-object, unknown-recipient,
cross-origin, unsupported. Accepted sub-reasons: stored, poll-vote, edit, refresh,
actor-refresh, actor-delete, removed, tombstone-only, auto-accepted, pending,
follow-answer, quote-answer, quote-granted, undone, reaction, reported. Rejected: blocked,
ignored, quote-refused.
A circle's traffic is private and kindless, and a stranger posting into one is just
not-addressed, so the event store cannot reveal a circle.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
InboxReceiver records each answer once, in a finally, with a reason: too-large, not-json,
not-activity, missing-type-or-actor, no-signature, the signature check's own codes
(headers-unsigned, digest-mismatch, header-unreadable, date-skew, expired,
algorithm-unsupported), signature-invalid, actor-not-key-owner, key-unavailable,
id-cross-origin, undo-foreign, misattributed, unknown-recipient, and for 202s queued,
duplicate, suspended or self-delete-unknown-key. Each event carries the activity and object
type, the inbox, the signature scheme, the bytes and the time taken. A 404 for an unknown
/mouth and a rate-limited inbox (429, from OnRejected) are recorded too.
Until the signature verifies, the host is only claimed, so it is kept only if the server is
already known. A suspended server is recorded under its own name.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
InteractionEvent records one interaction with a remote server: its channel (recv, in, out,
http, preview, crawl), activity and object type, outcome and reason, status, latency, wait,
bytes, attempt, audience, local actor kind, inbox, signature, features and the object's age.
IInteractionLedger.Record never blocks and never throws: events go into a bounded channel
of 10k, a full channel drops and counts, and a hosted service writes batches of up to 1000
every two seconds.
Privacy, as decided by the owner:
- no persona, root, group or activity id, inbox URL, actor URI or sender IP is stored;
- distinct accounts are counted with an HMAC keyed by a per-day salt (InteractionSalt,
upserted so restarts agree, never created for a past day);
- the local actor kind survives only on public and unlisted traffic;
- a host claimed by an unverified sender is kept only if it is already known.
Traffic caused by reading is only counted per day (InstanceDay.Reads, ServerDay). Indexes:
a 90-day TTL on events, unique day rows, and a TTL safety net on salts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
- JobQueue takes an optional scope, so a test's worker leases and reaps only its own jobs.
- The dead-host delivery test runs alone (Exclusive), on its own jobs, and cleans up the
breaker rows it trips; the breaker has tests of its own on unique hosts.
- Index and migration tests run alone: they drop indexes and rewrite every post.
- DomainBlocks.Load replaces reflection and a database-wide block in tests.
- Harness.Outgoing sees only deliveries queued since the harness started: Peer ports are
reused within a run, which made the circle test flaky.
- Two pure-logic tests leave Mongo-gated classes, so CI runs them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01ELjqpznMFMNrJoJUj6K5p2
thepraandClaude Opus 5.5