The head-vs-centroid eval (#1130) existed to prove the 'frozen embedding +
trained head' spine; the operator accepted the tagging system and dropped the
harness. Removed per rule 22: TagEvalCard + store, /api/tag_eval blueprint,
tag_eval_run ml task, recover-stalled-tag-eval-runs sweep + beat entry,
TagEvalRun model + table (migration 0073), and its tests.
The eval's data loaders + metric helpers were NOT eval-specific — the nightly
heads trainer runs on them — so they moved verbatim to
services/ml/training_data.py (heads.py import updated; behavior unchanged).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CDgx8bQS5YrGRK76v8HUnM
An errored GPU job's stored reason is a suspicion; the file probe is the
verdict. A 15-min beat sweep (triage_gpu_errors) runs verify_integrity's own
probe (sha256 + decode) on each errored image ONCE and writes both verdicts:
ImageRecord.integrity_status and the new GpuJob.triage_status ('defect' |
'file_ok', migration 0072). Every classification logs at WARNING so it
surfaces in Logs/System Activity.
- 'defect' rows are excluded from /retry_errors (re-running a known-bad file
burns agent time re-minting the tombstone); response now reports
defects_kept and the GpuAgentCard toast says so.
- GET /api/gpu/errors: triage view — reason buckets (classify_reason),
probe verdicts, per-job detail. POST /errors/triage runs the sweep now.
- POST /api/gpu/errors/<id>/recover: reuses the Layer-2 refetch pattern —
delete the defective copy + record (full cascade takes the tombstones too)
and re-poll its subscription Source so a fresh copy re-imports and re-enters
the pipeline; 'no_source' when nothing pollable resolves.
- New 'Failed processing' card (GpuTriageCard) in Maintenance: verdict counts,
reason summary, probe-now, defect list with thumbnails + per-image Recover.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CDgx8bQS5YrGRK76v8HUnM
- ml-backfill-daily: the CPU tag_and_embed backfill raced the GPU agent's
daily embed backfill for the same NULL-embedding images at ~100x the cost
(B1 audit verdict, milestone #124). The backfill TASK stays — the manual
/api/ml/backfill button remains the deliberate CPU fallback pending B3.
- purge-legacy: one-time IR-migration cleanup, dry-run verified 0 targets on
the live library before removal (A2 audit, milestone #123). Fully retired
per rule 22: tile, store action, route, service fn, tests.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CDgx8bQS5YrGRK76v8HUnM
The hourly ccip backfill's skip-list lacked 'error' (and the daily
siglip/embed variants re-gated failures on their missing results), so every
permanently-bad file got a fresh doomed job each run — ~24 duplicate error
rows/day per file, the perpetual 'unprocessable' flood. An errored job is now
a TOMBSTONE: no backfill re-enqueues it; retry is deliberate-only via
/retry_errors (an errored back-catalogue needs one button press after a
model swap).
One shared set of dedupe DELETEs (services/ml/gpu_jobs.error_dedupe_statements)
runs before every backfill and inside /retry_errors: error rows made moot by a
later pending/leased/done row go first, then older duplicates (newest reason
survives) — so the error count reads as distinct failing files and a retry
can't fan one file out into duplicate pending jobs. /retry_errors now returns
{requeued, pruned} and the toast shows both.
Poison-loop guards (release and lease-expiry burn no attempts, so a job that
stalls its transfer or crashes the agent every time cycled forever —
operator-observed jobs 99044/125288/131594/143131):
- agent: 3 in-session transient bounces (fetch or submit) → fail with the real
reason instead of another release; strikes never count while stopping, and
clear on submit success. Agent build 2026-07-02.3.
- server: the 60s orphan sweep (statements shared between the beat task and
GpuJobService so they can't drift) converts expired leases with >=5 lease
grants and pending jobs with >=10 to 'error', preserving the last stored
failure reason. Backstops old agent builds.
Tests: tombstone rule across all three backfill variants, moot-row pruning,
poison conversions, and the extended /retry_errors dedupe contract.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CDgx8bQS5YrGRK76v8HUnM
After an agent-side fix (e.g. the short-video sampler), the errored jobs
(~2.8k) have exhausted their 3 attempts and stay parked: backfill skips
images that already have a job, and /reprocess is the nuclear option (it
resets the 179k DONE jobs too). There was no way to re-run just the errors.
POST /api/gpu/retry_errors resets every status='error' job (all task types)
to pending with attempts=0 and the stored error cleared — a small inline
UPDATE that returns {requeued: n} so the UI toast can show the count.
UI: a "Retry errored jobs" button on the GPU-agent card, right under the
queue tiles; disabled when errored==0. With the agent now logging ffmpeg's
stderr on failure, retrying also reveals which errors were real vs victims
of the fps-filter bug.
Test: retry_errors requeues the errored job (fresh attempts, error cleared)
and leaves done work untouched; asserts via column selects (Core-DML
gotcha), not ORM refresh.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The image library is on a CIFS/SMB share (mounted rsize=4 MiB, actimeo=1), and
Quart's FileBody streams in 8 KiB chunks — so serving one large original was
~19k network round-trips to the storage server, i.e. 30–58s per download
(operator-flagged). That's what starved the GPU agent (constant "curator
unreachable" backoff) AND slowed the browser: every byte is read off CIFS and
streamed through the Python app (no reverse-proxy sendfile), and only 2 hypercorn
workers meant the agent + the browser's thumbnail grid queued behind each other.
In-container fix, no new service:
- Raise FileBody.buffer_size 8 KiB → 4 MiB in create_app, matching the mount's
read size: one round-trip per read, ~500× fewer. buffer_size is the MAX read so
small thumbnails still read in one gulp, and Range/mime/ETag/conditional
handling lives on Response — all preserved. Guarded so a Quart-internal change
can't break boot.
- HYPERCORN_WORKERS default 2 → 4 so concurrent /images requests stop queuing.
Expected: large-file transfers drop from ~40s toward link speed (a few seconds)
for the agent and the browser. See issue #1223.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The gallery's newest/oldest sort keys off image_record.effective_date =
COALESCE(primary post's post_date, created_at). The primary post is often the
repost/download the file came from, so the grid led with download dates rather
than when content was first posted (operator-flagged).
Add a second materialized sort key, earliest_post_date = MIN(post_date) across
ALL of an image's provenance posts (every post it appears in), else created_at —
the original publish date. Mirrors the effective_date pattern so the sort stays a
forward index scan.
- alembic 0071: add earliest_post_date + index (DESC, id DESC); backfill
created_at baseline then MIN over image_provenance ⋈ post.
- importer: recompute earliest_post_date whenever a dated post is linked (MIN over
the image's provenance, which now includes the just-added row).
- gallery_service: new sorts posted_new / posted_old key off earliest_post_date;
cursor + year/month grouping follow the active column transparently.
- api: accept posted_new|posted_old; DEFAULT is now posted_new so the grid leads
with original publish date. newest/oldest (effective_date) still available.
- frontend: sort dropdown gains "Newest/Oldest post date" (default Newest post
date); existing effective-date sorts relabelled "Newest/Oldest added".
- tests: service test asserts posted_new/posted_old key off earliest_post_date;
frontend default-sort omission test updated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Operator wants the Explore "related" rail to span more — the #1188 diversifier
was tuned conservatively. Push all three knobs so it reaches further across
clusters instead of clumping near the anchor:
- MMR lam 0.55 → 0.40 — weight the diversity penalty harder (the main dial).
- candidate pool min(200, max(limit*5, 60)) → min(400, max(limit*8, 100)) — a
wider nearest-cosine pool so MMR has genuinely distinct neighbourhoods to pick
from, not just the near-dupes.
- pHash dup_threshold 6 → 8 — collapse more near-duplicate reposts/clones,
freeing rail slots for distinct picks.
Still deterministic (same set per image, just more spread) and relevance-anchored
via the lam*sim-to-anchor term. Backend-only; no migration.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The throughput bottleneck was curator-side, not the network. lease() claimed the
lowest-id pending/expired jobs with `... ORDER BY id LIMIT n`, but with only a
plain `status` index Postgres walked the primary key from id=1, skipping the
entire prefix of already done/error rows before reaching pending ones. As `done`
grew (69k+), every lease became an O(done) scan — leasing crawled, the DB
saturated, and even /status (the queue GROUP BY count) stalled the agent.
- Migration 0070 adds two partial indexes over just the live slice: pending rows
indexed by id (hot path), and leased rows by lease_expires_at (crash-recovery
+ orphan sweep). They stay tiny no matter how large the done/error history.
- lease() split into two phases so each uses a partial index: claim pending
first (id-ordered, O(batch)); reclaim expired leases only when pending can't
fill the batch. Same semantics (SKIP LOCKED, attempts++, expired reclaim).
- Model __table_args__ declares the indexes so ORM and schema agree.
- Test: a done-prefix at low ids must not stop the lease reaching pending.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
- Migration 0069: new installs default to SigLIP 2 (so400m, 512px, 1152-d drop-in)
— UPDATE applies ONLY where no image is embedded yet (fresh install), so an
existing library is NOT silently invalidated; it switches deliberately via the
dropdown → Re-embed → Retrain. Column server_defaults moved to SigLIP 2.
- GET /api/ml/embedder-models: server-authoritative supported list (SigLIP 2 512
recommended / 384 faster / SigLIP 1 384 original) so the UI never free-types.
- GpuAgentCard: the two name/version text fields → a single model dropdown;
Save sets name+version from the picked option (the current model is always
selectable even if off-list).
- embedder.py DEFAULT_MODEL_NAME unchanged (stays the baked local-dir SigLIP 1)
to avoid a local-dir/weights mismatch; SigLIP 2 loads by HF name, cached on the
ml-worker's persistent HF_HOME.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The siglip/ccip backfills skip images that already have current-version regions,
so adding crop detectors only affected NEW images — the back-catalogue would
never be re-cropped. Add a reprocess trigger that resets every done/error job of
a task back to pending, so the agent re-runs the FULL pipeline (figure detection
+ CCIP + concept/panel crops) over the whole library under the current detectors.
- reprocess_gpu_jobs(task='ccip') task + POST /api/gpu/reprocess.
- gpu store reprocess() + GpuAgentCard "Re-process library (re-detect + re-crop)"
button with a confirm (it's heavy).
- Test: a done job resets to pending (attempts cleared).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Better region PROPOSERS feeding the existing crop→SigLIP→max-over-bag heads (no
change to the learned-tagging approach; no per-tag cost — propose once, embed
each region, all heads in one matmul).
- detectors.py: lazy ultralytics YOLO wrapper, each proposer independently
optional + guarded (a bad weight spec / inference error self-disables that one,
logged, never breaks the worker). Weights resolve from an ultralytics name |
http(s) URL | "hf_repo::file", cached under HF_HOME. NMS merge so a figure two
detectors both find collapses to one crop.
- worker: figure boxes = imgutils detect_person ∪ general COCO person (merged)
→ CCIP + concept (anime + Western/realistic coverage); booru_yolo anatomy
components (head/cat-head/anatomy/…) → concept crops; comic panels → kind=
'panel' concept crops. Capped per frame (MAX_COMPONENTS/MAX_PANELS).
- config + compose: PERSON_WEIGHTS (default yolo11n.pt, works OOB),
ANATOMY_WEIGHTS + PANEL_WEIGHTS (operator sets booru_yolo URL + mosesb panel
hf::file; empty = off). ultralytics added to requirements.
- backend: image_region 'kind' doc notes 'panel'; no migration (free String,
and the bag scorer keys on a non-null siglip_embedding, not the kind, so any
SigLIP region joins the bag automatically).
Agent is outside CI — py-compiled here; operator tests on the GPU and checks
Western-vs-anime crop quality via /api/ccip observability.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Fast, read-only, indexed aggregates shaped for ANALYSIS (not the paged UI
directory, which is alphabetical + builds previews and timed out at 10 min on a
full count sweep).
- GET /api/tags/top — top tags by image count, desc. ?kind, ?limit (cap 500),
?min_count, ?source=all|human|manual|accepted|auto (human=manual+ml_accepted,
auto=head_auto+ccip_auto+ml_auto). One GROUP BY over image_tag (indexed on
tag_id).
- GET /api/tags/<id>/stats — per-tag dataset health: total + per-source counts
(manual/accepted/head_auto/ccip_auto), human vs auto rollups, rejection count,
and whether a trained head exists. Backs concept-readiness + source-split
analysis.
Plain-HTTP homelab posture, no auth change. Tests cover ranking, source filter,
min_count, the source breakdown, and 404.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Hygiene follow-up to the Camie retirement (#1189) — these were left inert to
bound that change; nothing reads them now. Migration 0068 drops:
- ml_settings: tagger_store_floor, tagger_model_version, suggestion_threshold_
character/general (already dead pre-retirement — scoring uses per-head
thresholds), video_min_tag_frames (only the deleted video-prediction
aggregator used it).
- image_record: tagger_model_version (no writer), centroid_scores (dead JSON
cache, no reader).
Also: ml_admin _EDITABLE/GET/_validate pruned (dropped the store-floor invariant
+ video_min_tag_frames check); MLThresholdSliders trimmed to a video-embedding
card (interval + max frames only); importer no longer resets the dropped cols;
download_models drops the Camie fetch; stale CASCADE comments in cleanup_service
no longer name the removed tables. Tests updated.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Heads + CCIP are the tag source and head auto-apply is the earned propagation.
The Camie tagger ran only to feed the allowlist bulk-apply (its ImagePrediction
rows had no other consumer), and the allowlist was a SECOND, un-earned auto-apply
path firing in parallel with heads on every accept — exactly the un-earned spray
the v2 pivot replaced. Retire both.
Behavior change: accepting a suggestion now applies the tag to THAT image only
(source='ml_accepted', a head-training positive) — it no longer allowlists +
fans the tag across the library via Camie. Propagation is heads' earned
auto-apply. (Loses instant cold-start propagation for booru-vocab tags; that was
un-earned and bypassed the precision gate.)
- tag_and_embed is now EMBED-ONLY (no Camie load/infer, no ImagePrediction
writes); backfill enqueues it for images with no embedding.
- Removed: services/ml/tagger.py, apply_allowlist_tags + helpers + daily beat +
every enqueue caller (accept/alias/merge/per-image), api/allowlist.py +
blueprint, ImagePrediction + TagAllowlist models/tables (migration 0067),
AllowlistTable.vue + allowlist store, the accept coverage-projection payload.
- AllowlistService gutted to accept/dismiss/undismiss/reject (the rejection store
the rail still needs); accept returns nothing, API returns {accepted, tag_id}.
- tag merge no longer repoints/triggers the allowlist; _keep_as_alias now keys on
ML-applied image_tag sources (incl. head_auto) instead of the allowlist.
- UI: MLBackfillCard relabelled to embedding-only; accept toast simplified;
MaintenancePanel drops the allowlist tile.
Left for a follow-up hygiene pass (now-inert, harmless): the dead settings
columns (tagger_store_floor, tagger_model_version, suggestion_threshold_*,
video_min_tag_frames), image_record.tagger_model_version, MLThresholdSliders
trim, and the Camie model download in download_models.py.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The v2 pivot replaced per-tag SigLIP centroids with learned heads + CCIP.
Centroids were still recomputed (on every tag merge + a daily beat) but NOTHING
read them — suggestions come from heads+CCIP and apply_allowlist_tags applies
via Camie predictions, not centroids. Pure dead wiring; remove it.
Removed: CentroidService, recompute_centroid/recompute_centroids tasks, the
daily beat, POST /api/ml/recompute-centroids, the recompute-on-merge trigger,
the tag_reference_embedding table + model, the centroid_similarity_threshold +
min_reference_images settings (migration 0066), the CentroidRecomputeCard +
its store action + MaintenancePanel tile, and the centroid slider in
MLThresholdSliders. _keep_as_alias drops its vestigial has-centroid branch (the
allowlist branch already covers "could re-emit"); tag merge no longer clears a
table that no longer exists.
NOT touched (still live, parallel to heads): the Camie tagger, ImagePrediction,
and the allowlist bulk-apply — accepting a suggestion still allowlists + applies
it across the library. The tag-eval "centroid" baseline metric is unrelated
(in-memory) and stays. (image_record.centroid_scores JSON column also remains —
separate legacy field, its own micro-cleanup.)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Make the SigLIP embedder an operator choice (drop-in to SigLIP 2:
google/siglip2-so400m-patch16-512 is a verified 1152-d model at 512px → no
schema change, better small-cue fidelity). A swap = set model + re-embed +
retrain, all operator-driven; the GPU agent does the re-embed so it's fast.
- settings: embedder_model_name is now a setting (migration 0065) alongside the
existing embedder_model_version; both editable + validated (non-empty) in the
ml admin API. The server embedder loads by HF name (AutoImageProcessor/Model,
model-agnostic), preferring the pre-downloaded local dir for the default so
existing deploys don't re-download; rebuilds on a name change.
- agent: new 'embed' job = whole-image SigLIP embedding (mean-pool video frames)
under the lease-announced model → POST /jobs/submit_embedding writes
image_record.siglip_embedding + siglip_model_version. The lease now announces
the model FROM THE SETTING (not a constant).
- re-embed routing: enqueue_gpu_backfill('embed') selects unembedded + stale-
version images; 'siglip' now re-embeds concept crops whose version != current
(so a swap re-triggers crops, not just the never-embedded back-catalogue). The
CPU ml-worker backfill no longer re-embeds on a version mismatch (it can't
churn the library at 512px) — the GPU agent owns version re-embeds. Daily
'embed' + 'siglip' beats self-heal.
- scoring: score_image only bags embeddings in the CURRENT model's space (whole-
image gated by siglip_model_version, concept regions by embedding_version) so a
mid-swap stale vector isn't scored by new-space heads; legacy NULL = current.
- UI: GpuAgentCard "Embedding model (advanced)" — edit name/version, Save, and
"Re-embed library (GPU)" (queues embed + siglip); points at SigLIP 2.
Tests: lease announces model + submit_embedding round-trip; enqueue 'embed'
selects stale/unembedded; stale-version excluded from scoring; embedder model
settable + empty rejected; siglip gate updated to current-version concept.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Pure nearest-cosine piled near-identical images into the neighbour grid — a
reposted banner filled all 24 slots, and once you wandered into a B&W /
comic-panel cluster every neighbour was more of the same with no way back to
colour without the Random button (operator-reported, with screenshot).
similar() now over-fetches a wide candidate pool (5x the requested limit, cap
200), then diversifies down to `limit`:
- pHash near-duplicate collapse: drop candidates within 6 Hamming bits of the
anchor or an already-kept candidate, so a repost (and the anchor's own clones)
appears at most once.
- MMR re-rank: greedily pick for closeness-to-anchor minus similarity-to-already
-picked (lambda 0.55), so the result SPANS clusters instead of returning 40
variations of one image. Falls back to nearest-order on any failure / small
pool, so existing nearest-first behaviour is unchanged when there's nothing to
diversify.
Frontend forwardTarget drops the now-redundant skip-nearest-third hack (the list
is already diversified server-side) — plain random-over-unvisited gives the
variance now.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Lift recall on small/local concepts (glasses, cum, stomach-bulge, xray,
lactation) that the whole-image SigLIP vector washes out: the GPU agent now
embeds figure crops with SigLIP too, stored as kind='concept' regions, and the
suggestion rail scores each image as a BAG (whole-image + every concept crop),
taking each head's MAX over the bag. The whole-image vector is always in the
bag, so this can never score lower than before.
Model-agnostic by construction: the server ANNOUNCES the embedding model
(HF name + version) in the lease, so the agent loads whatever the heads were
trained in and stays in lock-step — a model swap is a server setting + a
re-embed migration, never an agent change.
- agent: model-agnostic CropEmbedder (torch/transformers get_image_features,
fp16 on CUDA, inference-locked); worker branches on job.task — 'ccip' emits
figure(CCIP)+concept(SigLIP) in one pass, 'siglip' emits concept-only so the
back-catalogue backfill never churns figure/CCIP regions; torch cu124 +
transformers in the image.
- server: lease announces embed_model_name/embed_version; score_image is
max-over-bag (version-filtered region embeddings); enqueue_gpu_backfill
'siglip' gates on a missing concept region (drains the back-catalogue,
retries failures, no double-enqueue); daily siglip-backfill beat; UI button;
/api/ccip/overview reports images_with_concept_siglip.
- v1 scope: suggestion rail only — auto-apply stays whole-image (conservative;
heads' thresholds were calibrated on whole-image). Bulk-apply bag = follow-up.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Works through the optional CCIP ideas + the "keep moving even if I forget" ask:
AUTOMATION (no button needed):
- Hourly beat auto-enqueues CCIP backfill — new images get embedded (and errored
ones retried) on their own; the queue never goes idle waiting for a click.
- CCIP auto-apply: a daily sweep tags confident matches (source='ccip_auto') so
identity tags keep flowing. ON by default (opt-out, like head auto-apply);
ml_settings.ccip_auto_apply_enabled + _threshold (0.92, above the suggest cut),
migration 0064. Vectorized (one matmul + reduceat per image), reversible, skips
already-applied/rejected. Switch + threshold in the GPU agent card; GET/PATCH
/api/ml/settings; auto_applied count in /api/ccip/overview.
REFERENCE QUALITY (the over-fire root cause):
- character_references now draws ONLY from single-character images — on a
multi-character image the tag is image-level, so every figure would otherwise
pollute each character's prototypes (a 2-char image tagged 'Velma' made
Daphne's figure a Velma reference). This is the contamination behind residual
over-firing.
- Cached on a cheap signature (char-tag count + ccip-region count/max-id) so the
reference load isn't redone on every modal open.
Tests: multi-character image not used as a reference; auto-apply tags a confident
match as ccip_auto.
NEXT (not done, confirmed): comic-panel cropping + SigLIP concept crops ("spot
interesting content").
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Live data showed the v1 flat 0.75 cosine over-fired — ~64% of matched images got
3-10 character guesses dominated by the most-referenced characters (a 27-ref
character clears a low bar on many images). A sweep showed 0.85 collapses the
noise (noisy multi-matches 47→3) while keeping the confident single-character
matches.
- ml_settings.ccip_match_threshold (migration 0063, default 0.85); match_image
reads it (override still accepted). DEFAULT_SIM_THRESHOLD fallback 0.75→0.85.
- Exposed in GET/PATCH /api/ml/settings (validated 0.5–0.999).
- Slider in the GPU agent card ("Character-match strictness") — tune live, no
redeploy, same observe-and-tune loop as auto-apply.
Test: a ~0.9-cosine figure matches at 0.85, dropped at 0.95.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
So work an agent orphaned gets picked back up quickly, three layers:
- GpuJobService.release(): a graceful agent stop hands its still-leased jobs back
to pending instantly (POST /api/gpu/jobs/release), no waiting out the lease.
- GpuJobService.recover_orphaned() + recover_orphaned_gpu_jobs Celery task on a
60s beat: resets expired leases (a hard-crashed agent) to pending and keeps the
queue counts honest even when nothing is leasing.
- Lease TTL 300→180s: still well above any single job (a capped-frame video embed
is tens of seconds, and a live worker heartbeats), but a hard crash recovers
faster once the sweep fires.
Tests: release returns-to-pending (token-scoped), recover_orphaned resets only
expired leases, release API round-trip.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
So the work can be checked through an API as the agent fills in vectors (same
pattern as /api/heads/metrics):
- GET /api/ccip/overview: regions by kind, images with figure CCIP vectors, the
per-character reference counts (which characters have enough examples to match
on), and the embedding versions present.
- GET /api/ccip/images/<id>: that image's stored regions (bbox, frame_time,
has_ccip/has_siglip, versions) + the CCIP character matches it would get — for
spot-checking detector + matcher output.
Read-only, no GPU. (Queue depth is already at /api/gpu/status.)
Tests: overview coverage counts + per-character refs; per-image regions + matches.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
SuggestionService.for_image now merges CCIP character matches with the SigLIP
head suggestions — they're complementary, not exclusive: CCIP is the identity-
specialized signal but needs a detected figure; the heads work whole-image but
conflate identity with style. Merged by tag: 'both' when they corroborate
(higher score wins), 'ccip' / 'head' otherwise. Cheap when no CCIP vectors exist
yet (match_image returns early without a figure vector), so it's a no-op until
the agent runs. Suggestion.source is now 'head' | 'ccip' | 'both'.
Test: a character with a CCIP reference figure surfaces (source='ccip') on a new
image whose figure matches.
NEXT: the agent container (real CCIP/detector models, hands-on) that produces the
vectors this consumes.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The server-side brain that turns stored CCIP vectors into character suggestions
— no GPU. character_references() gathers each character tag's prototype vectors
(figure/face-region CCIP embeddings on images carrying that tag); match_image()
cosine-matches an image's figure vectors against every character (multi-
prototype: best over a character's examples), surfacing those above a tunable
threshold as {tag_id, name, category:'character', score, source:'ccip'},
excluding already-applied characters. v1 = cosine on raw CCIP vectors; the exact
CCIP metric/threshold gets validated against the model in the hands-on eval.
Tests (synthetic vectors): same-character match across images, no-match for an
orthogonal figure, already-applied exclusion, no-figure-vectors empty.
NEXT: merge CCIP character suggestions into the rail; the agent container that
actually produces the vectors (hands-on, GPU — not CI-verifiable).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The thin HTTP surface over the queue so the desktop agent stays HTTP-only:
- Agent endpoints (Authorization: Bearer <token>): POST /api/gpu/jobs/lease
(returns jobs + image_url + mime + video frame cadence), /submit (stores
regions via RegionService + closes the job; 409 on a stale lease), /heartbeat,
/fail. Token validated against AppSetting (mirrors the extension-key pattern,
constant-time compare).
- Admin (browser): GET/POST /api/gpu/token[/rotate] (generate + show the agent
token), GET /api/gpu/status (queue counts), POST /api/gpu/backfill → dispatches
enqueue_gpu_backfill.
- enqueue_gpu_backfill(task): one INSERT…SELECT enqueues a job per image lacking
one for the task (scales to the full library; idempotent).
Agent flow: lease over HTTP → fetch pixels via the normal FC image URL → compute
on the GPU → submit. Redis/Postgres never exposed.
Tests: bearer required (+ wrong-token 401), lease→submit round-trip (region+CCIP
vector stored, job done via /status), stale-lease 409, backfill enqueue +
idempotency.
NEXT: the agent container + control UI, then the CCIP detector/embedder + matcher.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Answers "how are videos/all media handled by the GPU worker": a job is per ITEM,
but the agent fans a VIDEO into per-frame instances (ffmpeg in the agent, the
existing cadence), each stored with a timestamp — so a video becomes a BAG of
frame embeddings (fixes the mean-embedding muddle) instead of one washed-out
vector. Stills → frame_time NULL; animated GIF/WebP treated like short video.
- image_region.frame_time (migration 0061, not yet deployed so folded in): the
source frame's seconds for video/animated media; NULL for stills. RegionService
passes it through. A whole frame is just kind='frame'.
- gpu_job + GpuJobService (migration 0062): the durable work list that keeps the
desktop agent HTTP-only — enqueue (dedupes (image,task)) / lease (FOR UPDATE
SKIP LOCKED, re-claims expired leases so the queue self-heals) / heartbeat /
complete / fail (re-queues until MAX_ATTEMPTS then 'error'). The server enqueues;
the agent leases+submits over the web API; Redis/Postgres stay private.
Tests: enqueue dedupe, lease-then-skip-when-held, expired-lease reclaim, scoped
heartbeat, complete, fail-requeue-then-error. region test now covers frame_time.
NEXT: the thin HTTP API (lease/submit/heartbeat) + bearer-token auth, then the
agent container + control UI.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The storage backbone both crop jobs write to and read from. image_region =
normalized bbox (rx/ry/rw/rh) + kind ('face'/'figure' → CCIP character id;
'concept' → SigLIP head bag) + the crop's embedding (nullable Vector(768) CCIP /
Vector(1152) SigLIP, one per kind) + version stamps for compute-once gating. The
bbox doubles as grounded-tag provenance. Migration 0061.
RegionService.replace_regions (scoped BY KIND so the figure + concept pipelines
don't clobber each other) + get_regions — the GPU agent's results endpoint will
call the writer; the character matcher + bag scorer read. Server-side, no GPU.
Tests: replace/get round-trip, kind-scoped replacement, CCIP vector round-trip.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The trunk of both crop jobs — CCIP figure-crops and SigLIP concept-crops call
the SAME crop_region(): normalized-bbox crop with optional context padding,
edge-clamping, and the lower-bound size floor (max of a fraction-of-short-side
and an absolute pixel floor) below which a region is too small to embed and
returns None. Only the proposer (where) and embedder (what) differ; the crop is
shared. Pure Pillow — importable + testable anywhere (the GPU agent imports it
for the crop step). Unit-lane tests (no DB): region pixels, floor rejection,
edge clamp, pad expansion, out-size resize.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
dict(session.execute(...)) on a bare Result invokes the mapping protocol (a
Result has .keys() = column names) and subscripts it → "CursorResult is not
subscriptable". Materialize with .all() so dict() consumes rows as key-value
pairs. The API path already did this; the snapshot task missed it. Caught by
test_snapshot_records_timeseries_point (run 1628).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Auto-apply is now ON by default (operator-asked: opt-OUT, not opt-in) — migration
0059 + model default flipped. The support (>=30) + measured-precision gates keep
it safe and every auto-tag is reversible.
Observability so the operator can tune from real data:
- MISFIRE = an auto-applied (source='head_auto') tag the operator later removes.
UNDER-FIRE = a tag with a head the operator adds by hand (the head missed it).
Both captured at correction time in TagService.add_to_image/remove_from_image
(source is lost on delete) into durable per-tag counters (head_metric), keyed
by tag so they survive head retrain/prune.
- Daily snapshot_head_metrics writes a per-concept time-series point
(head_metrics_snapshot): auto-applied volume + cumulative misfires/under-fires
+ head quality; 180-day retention; daily beat.
- GET /api/heads/metrics: per-concept current counts + realized misfire rate +
head quality, plus the snapshot time-series — the report to tune the precision
target + support floor.
Migration 0060. Tests: misfire/under-fire counting (and the negatives — manual
removal isn't a misfire, headless manual add isn't an under-fire), snapshot
time-series, metrics API.
What's the autofire threshold? There's no single number — each graduated head
derives its OWN probability cutoff from its PR curve: the operating point that
holds precision >= head_auto_apply_precision (0.97) at max recall. The global
knobs are that target + the >=30 support floor.
NEXT (slice 3): UI — enable toggle, dry-run preview, per-concept trends.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Graduated heads can now apply their tag without a human — gated so it's safe:
- FIRING GATE: a head fires only when the master switch (head_auto_apply_enabled,
default OFF) is on AND it has >= head_auto_apply_min_positives (default 30)
clean labels. A precise-looking but under-supported low-N head can't spray tags.
- auto_apply_sweep (heads.py): streams every embedded image in chunks, scores
against the eligible heads (numpy, no sklearn), applies each head's tag where
score >= its auto_apply_threshold and the tag isn't already applied/rejected,
with source='head_auto' (distinguishable + reversible). dry_run counts only.
- HeadAutoApplyRun (migration 0059) tracks each sweep / preview; apply_head_tags
task (ml queue) + scheduled_apply_head_tags daily beat (no-op unless enabled)
+ recovery sweep + retention(20).
- API: POST /api/heads/auto-apply {dry_run} (202 / 409 running / 400 disabled),
GET /api/heads/auto-apply (recent runs + per-concept report). Settings
head_auto_apply_enabled + min_positives via /api/ml/settings.
Tests: sweep applies above threshold, dry-run writes nothing, skips under-
supported + ungraduated heads; API disabled/dry-run/conflict guards.
NEXT (slice 2): the observability the operator asked for — per-concept misfire
(auto-applied-then-removed) + under-fire tracking, time-series snapshots, and a
reporting API to tune. Slice 3: the UI (enable, preview, trends).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Two cadences for keeping heads in sync with your tagging:
- PASSIVE: a nightly `scheduled_train_heads` beat (skips if a run is already
in flight; creates+commits the run row before dispatching train_heads so the
ml worker always finds it). Folds the day's accepts/rejects + newly-eligible
concepts into the heads without anyone clicking.
- ACTIVE: a "Retrain heads" button in the Explore trail bar — bank the +/-
feedback you just gave while walking content, without a trip to Settings.
Shared logic in a new useHeadTraining composable (trigger + poll + start/finish
toasts), used by the Explore button; reflects an already-running run (incl. the
nightly one) on mount.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The rail's Suggestions now come from the trained per-concept heads. SuggestionService.for_image scores the image's frozen SigLIP embedding against
every head (heads.score_image) and surfaces concepts above each head's own
suggest threshold; the typed-dropdown's min=0 "show everything" mode maps to a
flat floor so any head-scored concept can still be picked. Already-applied tags
drop; rejected tags stay flagged + reversible (unchanged).
REMOVED from the suggestion path (rule 22, no fallback): the Camie
ImagePrediction candidate/alias/merge pipeline and the per-tag centroid
augmentation, plus the now-dead SuggestionService internals (_load_predictions,
_threshold_for, _settings, self.aliases, self.centroids). Head suggestions are
always canonical tags, so raw_name/via_alias are null/false and the rail's
alias kebab is inert by data (its removal + the Camie ingest-tagger rip are the
flagged follow-up). for_selection (bulk consensus) now aggregates head
suggestions unchanged.
Tests rewritten to the head path: test_ml_suggestions (surfaces/applied/
rejected-reversible/override/no-embedding/no-heads), test_suggestions_bulk
(consensus), test_api_suggestions (get + dropped the Camie-alias roundtrip),
and test_ml_artist_retired (artist not head-eligible via _HEAD_KINDS).
DEPLOY NOTE: after this lands, the rail is empty until you run Train heads
(Settings → Tagging → Concept heads) — deploy, train, then the rail populates.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
Alphabetize HeadTrainingRun in models/__init__ + maintenance imports (H before
I), and drop the inline comment that split heads.py's import block. Pure import
ordering — no behavior change. (run 1601 lint)
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
The eval (#1130) proved the frozen-embedding + trained-head spine; this lands
its production form (the first of three slices that make heads the suggestion
source, replacing Camie + centroid).
- tag_head: one logistic-regression head per general/character concept with
enough labelled positives. Weights (pgvector), honest CV-derived suggest
threshold + earned-auto-apply point, and per-concept quality metrics.
- head_training_run: persisted batch lifecycle (mirrors tag_eval_run) so the
admin card shows live + historical status across navigation.
- services/ml/heads.py: TRAIN (sync, ml worker, reuses tag_eval's proven data
loaders + metric math so production heads match measured eval numbers) and
SCORE (async, API worker — numpy via pgvector, no scikit-learn): score one
image's embedding against all heads → the rail's suggestions, cached on
(count, max trained_at) so a retrain invalidates without per-request loads.
- tasks.ml.train_heads (ml queue, commits per head so a kill leaves progress)
+ recover_stalled_head_training_runs sweep + retention(20) + 5-min beat
(rule 89).
- api/heads.py: POST /api/heads/train (one run at a time, 409 guard) + GET
/api/heads (count, graduated, last-trained, running, per-concept table,
recent runs).
- ml_settings: head_min_positives + head_auto_apply_precision, tunable via
/api/ml/settings.
Scoring isn't wired into the rail yet (slice C) and the admin UI is slice B —
this slice makes training + scoring exist and CI-verifiable. 'precision' column
stored as precision_cv (SQL reserved word). Migration 0058.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
A red-✗ dismissal no longer makes the suggestion vanish. The rejected tag
stays in the rail — dimmed, struck-through, with a "rejected" pill and a
one-click undo (↶) in place of the ✗ — so a misclick is recoverable and the
operator can see what they've said no to (operator-asked 2026-06-27).
Backend: SuggestionService.for_image now KEEPS rejected tags, flagged
rejected=True, sorted to the bottom of their category, instead of dropping
them. New AllowlistService.undismiss + POST /suggestions/undismiss clears the
TagSuggestionRejection. Rejected items are still excluded from bulk consensus
(for_selection) and the type-to-add dropdown, whose jobs are unchanged.
Frontend: store.dismiss flags in place (canonical tags) rather than dropping;
new store.undismiss reverts. SuggestionItem renders the rejected state and
swaps ✗→↶; ✓ still accepts (which clears the rejection server-side).
Tests: rejected-surfaced-flagged-then-reversible (service) + undismiss
endpoint idempotency (API).
Completes #1134's reversible-rejection half. Heads-as-suggestion-source is
the remaining piece.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ttrj5P7upUTueSfoJcxEqa
"Keep" on a doubted positive was a no-op, so the same confirmed-correct images
came back in "head doubts" every run (operator-flagged: reinforcement keeps
surfacing the same images). Add tag_positive_confirmation (mirror of
tag_suggestion_rejection): keep → POST /images/<id>/tags/<tag_id>/confirm, and
the eval excludes confirmed positives from the doubts list — exactly as rejected
items already drop out of the suggest list. The tag stays a positive either way
(confirmation is a "reviewed" marker, not a training change).
- model TagPositiveConfirmation + migration 0057; confirm endpoint (idempotent).
- tag_eval: _confirmed_ids + exclude from head_doubts_positive examples.
- store.confirmTag + card "keep" calls it.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
"head would suggest" drew from the whole negative pool, which INCLUDES the
images the operator rejected. A rejected near-miss (e.g. an orc under "goblin")
is a hard negative that still scores high, so it kept resurfacing as a fresh
suggestion every run (operator-flagged: "same items keep appearing"). Exclude
already-rejected ids from the suggest list — once you've said no, it's gone.
(head doubts = lowest-scoring positives is unchanged; genuinely-hard true
positives legitimately recur there.)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Two additions driven by "what's the commit threshold?" + "find more tags":
1. High-precision operating point (Bar 4). Per concept, report the threshold that
maximizes recall while holding precision >= a target (default 0.97, configurable
via `precision_target`) — i.e. "could this fire without a human, and how much
would it catch?" `head.auto_apply` = {target, threshold, precision, recall} or
null if the target is unreachable. Surfaced on the card.
2. Server-side concept auto-discovery. `auto_top_n` param unions the explicit
concept list with the N most-tagged general tags (one fast DB query) so the
eval can broaden itself without hand-listing — replaces the slow HTTP directory
paging. Card gains "+ auto-add top-N" and precision-target inputs.
No migration; numpy/sklearn stay lazy. Existing _normalize_params test still
holds (new keys additive; None still falls back to DEFAULT_CONCEPTS).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Frontend for #1130. A maintenance tile in Settings → Tagging:
- Editable concept list + "Run eval" → POST /api/tag-eval (one running at a time).
- Rehydrates on mount via the persisted run (getRun by latest id) and polls while
running — so the report SURVIVES navigation (operator-flagged); the task runs
backend-side regardless and the card reconnects to its row.
- Renders the saved report: per-concept head-vs-centroid metrics table (AP/F1/
precision/recall) with Δ AP, the learning curve (AP @ N positives), and
thumbnail galleries (head-would-suggest / head-doubts-positive) for eyeballing.
Backend: _examples now stores thumbnail_urls (not just ids) so the report is a
self-contained artifact that renders without per-id lookups on reload.
No new top-level surface — slots into the existing maintenance area.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Slice 1 of milestone #114 (tagging v2). Proves the frozen-embedding + trained-
head spine on the operator's own data, reusing the SigLIP embeddings already
stored on image_record — no re-embedding, no GPU.
Per concept: train a logistic-regression HEAD (positives + negatives = explicit
rejections + sampled unlabeled) vs the old single-CENTROID baseline; report
cross-validated precision/recall/AP for both, a LEARNING CURVE (AP/F1 as tagged
positives grow 10→30→100→300), and example image ids (head-would-suggest /
head-doubts-positive) to eyeball.
Persisted so the report SURVIVES navigation (operator-flagged): the run + full
report live in a new tag_eval_run row (mirrors library_audit_run); the admin
card will rehydrate from GET on mount, not transient state.
- models.TagEvalRun + migration 0056; runs on the ml queue (only worker with
numpy/sklearn) — numpy/sklearn lazy-imported so the API can still enqueue.
- services/ml/tag_eval (compute + start helper, one-running guard), tasks.ml
.tag_eval_run, api/tag-eval (POST create, GET history light / detail w/ report).
- recover_stalled_tag_eval_runs sweep + retention (keep last 20) + 5-min beat
(rule 89). scikit-learn added to requirements-ml.
- tests: param normalization + the rehydrate read-path + create/conflict.
Frontend admin card (trigger + render persisted report) follows next.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The cluster tag-gap feature's only UI (Explore's TagGapPanel) was removed in the
3-pane rework, leaving the backend that fed it with no caller. Surgical removal:
- drop the POST /api/images/cluster/tag-gaps route (cluster_tag_gaps)
- drop BulkTagService.tag_gaps (+ the now-unused `import math`)
- drop the tag_gaps tests (test_bulk_tag_service, test_api_bulk_tags)
BulkTagService's common_tags / bulk_add / bulk_remove stay — they still back the
gallery bulk editor. Pure deletion, no behaviour change.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The directory card count regressed to a globally-inflated number (~every
card showed the same ~469): the fandom leg used a doubly-nested correlated
subquery — image_tag.tag_id IN (SELECT member.id WHERE member.fandom_id ==
Tag.id) — whose inner predicate did not correlate the outer Tag, so it
matched EVERY character that has any fandom and counted all their images for
every tag. The gallery scope and cleanup count were unaffected (they pass a
literal tag id, a single-level subquery), which is why only the card diverged
from the gallery.
Rewrite the count as a single-level correlated scalar subquery: join `member`
(the tag applied to the image) and match image_tag.tag_id == Tag.id (direct)
OR member.fandom_id == Tag.id (a character of this fandom). Strengthen the
directory test with a second unrelated fandom/character so a non-correlating
fandom leg fails (count would read 4 instead of 3).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
