scripts · pub-search · Tangled

zzstoatzz.io / pub-search

search for standard sites pub-search.waow.tech

search zig blog atproto

pub-search scripts

1 folder · 56 files

zzstoatzz · judge: use Haiku 5.5 with corrected authorship evaluation · 11h ago

judge: use Haiku 5.5 with corrected authorship evaluation

apply-corpus-reconciliation

feat(reconcile): add bounded corpus apply pipeline

atlas_context.py

Ground Atlas summaries in related notes and preserve snapshot history

atlas_history.py

atlas: load the map from a columnar view of atlas.json.gz

atlas_labels.py

fix(labeler): count unique bulk publishers

atlas_rewrite_summaries.py

Preserve sampling metadata and expose rewrite validation failures

atlas_summaries.py

atlas: load topic summaries from a small sidecar and tidy the topic panel lists

atlas: load the map from a columnar view of atlas.json.gz

fix(audit): paginate resumable snapshot export

feat(backend): safety bundle — async backfill, bridgy guard on the side door, quick_check boot

backfill-content-type

feat(site): /recommended endpoint + leaderboard page

backfill-cover-images

fix: sync cover_image to local replica, cycling theme icon, clickable platform badges

backfill-embeddings

rename: leaflet-search → pub-search (repo self-references)

backfill-pckt-platform

chore(scripts): identify backfill clients with a User-Agent

feat: add whitewind support to backfill-pds script + fix dep URLs

backfill-publications

chore(scripts): identify backfill clients with a User-Agent

backfill-recommends

chore(scripts): identify backfill clients with a User-Agent

backfill-subscriptions

chore(scripts): User-Agent on the remaining backfill/reconcile clients

fix: increase VM memory and keep machine running

atlas: use Haiku 5.5 for cluster labels

wrapped: wheel-tick fact transitions + visible links on facts

add vector similarity search and tombstone tracking

add vector similarity search and tombstone tracking

classify-bulk-mirror

feat(labeler): pub-search bulk-mirror labeler in the backend

cleanup-republish-ghosts

scripts: add cleanup-republish-ghosts (PDS vs turso vs tpuf per-DID audit)

cleanup-vector-index

fix: improve semantic search quality + revert mode toggle UI

constellation_pacing.py

reconcile-recommends: daily, single-worker, recency-scoped; fail on persistent 429

daily-discovery

fix(daily-discovery): dedup over the full window, not 20h

add terminal dashboard script with rich + plotext

rename: leaflet-search → pub-search (repo self-references)

fixup-subscription-created-at

fix(subscriptions): derive created_at from rkey TID when createdAt absent

ingestion-watchdog

fix(watchdog): stop paging on publication volume, not ingest health

judge: use Haiku 5.5 with corrected authorship evaluation

feat(labeler): rename label machine-generated → bulk-generated

atlas: ship atlas.json gzipped to stay under pages' 25MiB file limit

fix: dotenv priority in scripts, add connection lines to constellation

feat(mcp)!: curate the tool surface instead of mirroring endpoints

measure-distances

fix: raise semantic distance threshold from 0.5 to 0.6

offline-replica-catchup

feat(backend): drivepatents ban + snapshot adoption + span use-after-free fix

populate-vector-index

chore: add vector index and bridgy-fed purge scripts

purge-banned-turso

scripts: purge-banned-turso takes --batch and --timeout

purge-bridgyfed

fix(ingester): drop bridgy fed at the source — sig_only re-opened a door tap had sealed

purge-bridgyfed-local

fix(ingester): drop bridgy fed at the source — sig_only re-opened a door tap had sealed

purge-bridgyfed-vectors

refactor(bans): single source of truth for banned DIDs (banned-dids.txt)

purge-republish-duplicates

scripts: one-time purge of accumulated republish duplicates

rebuild-documents-table

fix: sync MCP stats model and scripts with voyage-4-lite

perf(fts): key documents_fts deletes by rowid; stop re-embedding on base_path change

rebuild-pub-fts

feat: search documents by publication subdomain (base_path)

rebuild-vector-index

fix: deduplicate semantic search results + clean up rebuild script

reconcile-corpus

docs(reconcile): describe separate guarded apply step

reconcile-recommends

reconcile-recommends: daily, single-worker, recency-scoped; fail on persistent 429

scripts(smoke): drop empty-query probe — the API rejects it by design and the homepage never sends it

spike-columnar-atlas

atlas: spike columnar/dictionary-encoded format (v2)

add vector similarity search and tombstone tracking

visibility-baseline

Merge: showInDiscover enforcement via a complete policy set

wait-and-create-index

chore: add vector index and bridgy-fed purge scripts

watchdog_checks.py

fix(watchdog): derive the snapshot bound from the adoption window