EventPulse

EventPulse — trustworthy San Francisco events for SceneShift

Collect attributed public event listings for a live SF discovery map. SceneShift helps residents and newcomers explore local culture and gives organizers visibility beyond their followers.

Release status

This repository now contains the source retrieved from owner Actor version 0.0 and a proposed 0.1 quality update. The existing live build is 0.0.13. Do not claim this update is cloud-verified until its build/run evidence is recorded below. Local source includes tests and a reproducible npm lockfile.

Sources

• Eventbrite through the Store Actor (aitorsm/eventbrite): source-local times require a supplied timezone; inconsistent free/paid evidence becomes unknown with a review warning.
• Luma through aitorsm/luma-events: retains sourced coordinates and dates. Ticket availability is raw evidence, not a verified application claim.
• DoTheBay: custom Crawlee/Cheerio adapter, Pacific calendar days, bounded/rate-limited crawling, ongoing-event overlap, optional exact-address geocoding.
• Partiful through heliostech/partiful-event-scraper: optional adapter; records without a public SF address go to diagnostics. Current evidence does not establish usable Partiful coverage. Do not infer hidden locations.
• Funcheap: not supported in this release. The old implementation contained an unused/unvalidated scraper; it has been removed rather than presented as working coverage.

Input example

{"enabledSources":["eventbrite","luma","dothebay"],"maxEventbriteEvents":2,"maxLumaEvents":2,"maxDoTheBayEvents":2,"daysAhead":14,"resolveCoordinates":true,"maxGeocodeRequests":2}

Defaults: 5 events per source, SF city slugs, 14-day overlap window, geocoding off by default; up to 3 requests only when explicitly enabled. Source caps 1–1000; DoTheBay alone retains legacy 0 meaning no event cap, bounded by days/requests. maxGeocodeRequests is 0–10. Individual source failures produce diagnostics without fabricating success. Enabled sources are explicit; [] permits an offline Actor initialization smoke check. No proxy settings are added.

Output and uncertainty

Flat records preserve source, title, shortDescription, startDateTime/endDateTime (UTC), venue/address, priceStatus/priceText, canonical url, collectedAt, categories, tags, coordinates, and extra. Missing ends stay null; a labeled three-hour fallback is used only to decide window overlap. Invalid or ambiguous required dates are excluded with diagnostics. Repeated DST times are rejected instead of guessing. In-run deduplication uses URL + occurrence start; downstream consumers must also preserve distinct recurring occurrences when a source reuses URLs.

Categories use documented keyword rules over title, description and venue; they are classification, not new event facts. Unmatched events have an empty category list. Coordinates come from the listing, or a bounded exact house-number/road/San Francisco geocoder match. Fuzzy matches, wrong cities, ambiguous results and unavailable geocoding remain null. No city-center fallback. Preserve attribution to © OpenStreetMap contributors if using derived coordinates.

Nominatim requests identify this Actor, run sequentially at least 1.1 seconds apart, and successful lookups are cached for 30 days in sceneshift-eventpulse-geocodes. Run only one geocoding Actor at a time across this account to respect the public service's global 1 request/second limit. Lookup is off by default. Enable only for coordinated single-run prototype use; use a dedicated geocoding service for public concurrent workloads. Public service availability is not guaranteed.

DIAGNOSTICS in the run's default key-value store records source counts/errors, excluded URLs with reasons, accepted count, and geocoder request count. Output excludes guest/attendance statistics and personal attendee information. An event absent from a later crawl is never inferred cancelled. Explicit cancellation reconciliation is still a gap in these source adapters and must be added before claiming comprehensive lifecycle coverage.

Convex handoff

This Actor only supplies data. It has no Convex credentials and never approves or publishes records. The protected importer validates/upserts records as pending and unverified, preserves provenance, and calculates mapEndsAt. Availability booleans must not become sold-out labels without explicit source evidence and the actual check time. Source quality warnings need human review.

Development and costs

npm ci
npm test
npm run build
apify validate-schema
apify run --user-agent apify-codex-plugin/apify-actor-development

Put input in storage/key_value_stores/default/INPUT.json. Local data never becomes cloud run evidence. Live source tests require authenticated access for nested Store Actors and incur their fees plus compute/storage. Use small source caps and maxGeocodeRequests. A parent run budget does not guarantee a total cap on nested Actor fees. Never hardcode or log APIFY_TOKEN. The platform supplies it to the Actor.

Evidence

Existing 0.0.13 run ieoWQtqh2UjhguSoG succeeded and produced three records imported into development Convex; repeat import inserted zero duplicates. The six-record earlier dataset exposed the issues this update addresses. See ../../docs/APIFY_INTEGRATION.md for IDs and review evidence. This 0.1 update's local/cloud validation results must be recorded separately.