spb/websensor

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

- Deep discovery (robots → sitemaps → feeds incl. official sub-domains → navigation-classified pages → status providers → GitHub org → EDGAR → Hugging Face → OpenAPI → posture), every candidate validated, wildcard DNS / anti-bot detection, content de-duplication, expected change frequency (median gap)
- Factory: coverage universes → factory_seeds → scored candidates (explainable) → promotion caps → SHADOW sensors (polled, never published) → evidence-based acceptance/rejection; separate websensor-factory process with Redis heartbeat; CLI expand/factory seed|run|evaluate|stats|export|requeue; YAML export for graduation
- Migration 0007 (sources.origin/sector, factory_seeds, candidate lifecycle columns, factory_daily); pipeline keeps SHADOW status and skips events for shadow sensors
- Coverage: sector universes schema + validator (--dns), registrableDomain(), GET /api/v1/coverage[/:sector], pages /coverage and /coverage/[sector], homepage rail, stats exclude shadow sensors
- Admin API /api/v1/admin/factory* + /ops Source Factory panel (funnel, sectors, candidates, accept/reject)
- Docs: COVERAGE.md, CLAUDE.md, deploy README, ARCHITECTURE, README; manifest gains websensor-factory + factory seed hook; version 0.3.0

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Registry: 7 depth fragments (40–46: AI frontier, cloud infrastructure, cybersecurity,
finance & markets, governments, science & health, transport/telecom/sports/news) →
3 167 organizations / 6 320 validated sensors (was 2 681 / 4 052); seeds carry
country, language and first_party; sync infers country/language by TLD and a
0–3 scheduling priority.

Core: semantic diff (noise classes vs pricing/policy/product/personnel, field-level
before → after with % deltas), WebSensor Signal Score with explainable reasons,
impact / velocity / anomaly scores, breaking state, entity rank, ~40 new event
classes (zero_day, active_exploitation, supply_chain_attack, guidance, buyback,
sanction, legislation, outbreak, grounding, service_shutdown…), search syntax
(entity: type: after:7d silent: importance:>70 country:), shared registry schema.

Engine: noise never becomes an event, routine firehose batches damped, fingerprint
idempotency, ambiguous-alias rule for entity resolution (+ relink-entities CLI),
silent-change bar (first-party · eligible type · importance ≥ 45), clusters with
propagation timeline, first-party/external counts, velocity, lead time and
breaking/developing/confirmed states, server-side alerts with HMAC-signed
webhooks, raw-body retention with reference counting, priority-aware scheduler
with per-domain circuit breaker and Redis heartbeat, UTF-16 BOM transcoding.

API: new filters (signal_min, group, first_party, confirmed, country, language,
change_class, q syntax, order=signal), /breaking /pulse /radar /clusters/:slug
/entities/rank /countries /categories/:channel /sensors/:id/snapshots, owner
routes (bookmarks, saved views, notifications, custom URL monitors with SSRF
checks), admin routes behind WS_ADMIN_TOKEN (ops, sensor/source actions,
connector test, YAML/JSON bulk import), TTL cache, WebSocket protocol 2 with
stream ids and replay.

Web: terminal-style design system (badges, density modes, skeletons, sparklines,
heatmaps), live system strip, ⌘K command palette, intelligence drawer, live feed
v2 (URL-synced filters, pause, "↑ N new events", replay), pages /live /pulse
/radar /entity/[id] /cluster/[slug] /country /bookmarks /monitors /ops, redesigned
breaking / silent / explore / event / source / sensor / search / alerts /
watchlists / health / api docs, OG image per event, sitemap, loading skeletons.

Migrations 0005 (additive schema) and 0006 (backfills). Docs updated.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>