v1.3 (2026-06-11) — success-rate hardening. The orchestrator now NEVER finishes FAILED on an external condition. When every platform that ran failed to return data (anti-bot blocks, target rate-limits, or a transient sub-scraper run failure) the run soft-fails: SUCCESS exit with a WARNING status and exactly one schema-valid diagnostic report row explaining what happened (previously this path called Actor.fail). Every soft-exit path (empty/invalid input, no usable platforms, account memory/quota cap, all-platforms-blocked) now pushes one schema-valid diagnostic row so the dataset is never silently empty. safePushData gained a third tier: if a record is rejected by dataset_schema even after stripping nulls, a guaranteed-minimal schema-compliant record is pushed so one bad item can never abort the batch. Sub-call memory retry at 1024MB on 'exceed the memory limit' retained. v1.2 (2026-06-11) — success-rate fix. YouTube: search results are now strictly filtered to watchable video URLs (channel/playlist results that made the comments scraper FAIL are dropped); when no video URLs remain the platform is a soft skip, not a hard error. Memory/quota (infra) errors caught inside a platform scraper now carry an explicit infraError flag up to the orchestrator, so a run blocked purely by the account memory cap soft-fails (SUCCESS exit with a WARNING) instead of hard-failing. v1.1 (2026-06-06) — reliability fix. Sub-actor calls now run sequentially with a conservative memory cap (2048MB, retry at 1024MB) so a sub-scraper can no longer blow the account memory limit and cascade into a failed run. Partial success: if at least one platform returns data the run finishes as SUCCESS with warnings for the failed platforms. If every platform is blocked purely by an account resource limit (memory/quota), the run soft-fails (success exit with a WARNING) instead of hard-failing. v1.0 (2026-06-02) — initial release. Orchestrates owned scrapers (Reddit, YouTube, X/Twitter, Instagram) via Actor.call, then Claude Haiku per-item sentiment + emotion + aspects + toxicity, plus an aggregate AI brand-health report. Lexicon fallback. PPE: $0.002 start + $0.003/post + $0.10/report.
