# Changelog of Australia NDIS Provider Intelligence (`fryingken/ndis-provider-intelligence`) Actor

- **URL**: https://apify.com/fryingken/ndis-provider-intelligence/changelog.md
- **Full Actor documentation**: https://apify.com/fryingken/ndis-provider-intelligence.md

## Changelog

### Unreleased — approved synthetic PPE

- Adopt one synthetic Dataset-item event at $0.002/provider; retain default synthetic
  startup pricing, no custom/base fee and no separate usage surcharge.
- Bound delivery by SDK affordability and maxResults; skip acquisition when exhausted,
  report actual saved counts and chargeLimitReached. Retain no-replay Cloud writes.
- Set repository default memory to 2048 MB; Console requires 1200s/LIMITED\_PERMISSIONS.
- Add paid-budget/restart/failure tests and real SDK local PPE smoke. Private Cloud
  billing verification is required before main or publication.

### Unreleased — measured pricing proposal

- Record all four valid corrected benchmarks, combined means and variability.
- Recommend 2048 MB/1200s and document three unapproved PPE price points plus a
  startup-component alternative. Exclude the invalid 6485-item run from economics.
- No runtime, resource configuration or monetization changes.

### Unreleased — Dataset count propagation

- Correct immediate metadata verification: Apify item counts can lag successful writes.
  Poll reads every second within 15s; never replay appends. Excess counts fail immediately.
- Add verification-stage diagnostics and delayed-count, excess-count and timeout regressions.

### Unreleased — Dataset result-cap safeguards

- Record six user-reported Cloud benchmarks; exclude the 6485-item / maxResults=5000
  case from pricing. Historical cause remains unconfirmed without run exports.
- Validate output cap/unique IDs, reject used storage or a previous write attempt,
  disable automatic Cloud append retries and verify actual Dataset count before success.
  Persist write intent; uncertain delivery requires a fresh run. No PPE changes.
- Add exact-cap, byte-boundary, lost-acknowledgement and restart regression tests.

### Checkpoint 2 readiness — 2026-09-17 (unreleased)

- Recorded Checkpoint 1 PASS from the user's successful private Cloud run:
  LIMITED\_PERMISSIONS, 4096 MB, five results, 496 seconds, USD 0.111 per run.
- Prepared fixed-build 100/1000/5000-result benchmarks at 2048/4096 MB, an evidence
  ledger, fixed-versus-marginal cost procedure and gated Store release checklist.
- Made the Actor-directory Docker context and provisional 4096 MB default explicit.
- Added a schema-validated fictional output example. Existing stage timing, HTTPX
  navigation headers and all parsing/matching behavior remain unchanged.
- PPE implementation/pricing, production main-branch switch and publication remain
  gated on benchmark evidence, production verification and user approval as applicable.

### 0.1.0 — 2026-09-16 (unreleased)

- Added NDIS CSV acquisition, normalization, filters, conservative compliance matching,
  partial-source failure reporting, Apify schemas, Dataset/OUTPUT integration and offline tests.
- Mapped current provider address/group fields; retained multiple registrations under
  one ABN without combining unrelated filter or name/location evidence.
- Added regression tests for outlet merging and conflicting provider identifiers.
- At that initial local milestone, no Cloud deployment, Store publication or billing configuration.