The README is now purely the public Apify Store page: build, test and schema notes moved
to DEVELOPMENT.md, and field documentation points to the Store's Output tab instead of
files in the (private) repository.
Earthquakes are stored and charged as one batch with
Actor.pushData(items, 'earthquake-record'): the SDK trims the batch to what the
spending limit covers, stores it in one request, then charges the stored count in one
more. 0.2.1 made two API calls per earthquake, so a 63-earthquake cloud run took 16-93 s
and one capped run stalled on a single call until it timed out; batched, the same run
takes about 5 s. Billing is unchanged: every stored earthquake is charged, nothing else,
and a capped run keeps the most recent earthquakes that fit (#7).
The actor's Console default run options are now 256 MB and a 300 s timeout (they were
4096 MB and 3600 s). Events-only pricing leaves platform costs with the developer, so
the defaults stay small.
[0.2.1] - 2026-09-24
Fixed
A run stopped by the user's spending limit no longer bills an earthquake it does not
deliver. Earthquakes were charged before they were pushed, and the run stopped as soon as
the SDK reported the limit. The SDK reports the limit after a successful charge, so the
last paid earthquake was never stored. Earthquakes now go through
Actor.pushData(item, 'earthquake-record'), which stores each item before charging it
(#5).
A source the spending limit cannot cover is now skipped before it is fetched.
Changed
Run starts are billed by Apify's synthetic apify-actor-start event instead of the
custom actor-start event. Apify charges it and covers the first 5 s of compute. The
code no longer charges actor-start, so that event must not be configured in Console.
SDK wiring moved from main.ts to src/platform.ts (createPlatformDependencies), so a
test can drive the real SDK charging in its local pay-per-event mode.
[0.2.0] - 2026-09-23
Added
Apify output schema (.actor/output_schema.json) so the actor is Store-ready, pointing
agents at the earthquakes and weather dataset outputs with descriptions written for
AI consumers.
Dataset schemas (.actor/earthquakes_dataset_schema.json,
.actor/weather_dataset_schema.json) with per-field title/description/example and
an "Overview" view each. Their fields are JSON Schema draft-07 (the draft Apify
validates dataset items against), generated from the zod schemas via z.toJSONSchema
(pnpm run schemas:generate), with tests that fail on drift and that replay Apify's
.actor schema validators and AJV item validation offline.
weatherSnapshotSchema and the weather alias dataset for the nested per-run snapshot.
Changed
Split output into two datasets: PHIVOLCS earthquakes stay in the default dataset (one
item per earthquake), while PAGASA now writes exactly one nested snapshot per run to the
weather alias dataset (Actor.openDataset({ alias: 'weather' })). The snapshot keeps
source, fetchedAt and sourceUrl once at the top level and nests the bulletin
status, tropical cyclones, forecasts, wind/coastal rows, temperature/humidity and
tides/astronomy instead of repeating provenance per row.
Extracted the run orchestration into runActor (dependency-injected) so dataset routing
is unit-tested without the network.
Input, the pay-per-event names (actor-start, weather-snapshot, earthquake-record)
and their charge points, limited permissions and no-Standby behaviour are unchanged.
