Puppeteer Table Row Change Monitor

Monitor one selected public HTML table per URL with a unique key column. Returns added/deleted rows, cell-level updates with before/after values, column headers and row order. Changed key becomes delete/add; changed price preserves row ID. First run exports baseline rows; unchanged returns empty changes. This is a structured data contract rather than RAG text chunks or a generic CSV wrapper.

Use examples/basic.json. Up to3 URLs, one matched table, single TH header, zero-based keyColumn. Reuse scopeKey/selector/keyColumn/settle time. ParentSnapshotId must match your downstream snapshot; serial runs required. Changing settings or resetting baseline requires explicit consumer full resync. Header drift returns INCOMPLETE/SCHEMA_CHANGED and preserves old baseline.

Coverage is the selected main-frame HTML table at capture time, not the whole site or all remote records. No pagination, scrolling, virtual grids, frames, merged cells, nested tables or repeated/empty redacted keys. Known DataTables paging is rejected; arbitrary site pagination or anti-bot pages cannot always be detected. Empty table is incomplete to avoid mass deletion.

Limits:200 data rows,30 columns,500 chars/cell,32k raw chars,10k DOM/text-node visits,256KiB dataset item;60s capture,150 connections/16MiB network cutoff. Cutoff counters checked after shutdown; buffered bytes can overshoot, so this is not a strict browser-memory or exact wire-byte ceiling. Over-limit capture is incomplete/free.

Public text is untrusted data. Heuristic URL/email/credential/opaque-number redaction occurs before identity/storage; it is not complete PII detection and may hide useful identifiers. Editable controls, script/hidden text excluded; only safe page origin/path hash in metadata. No logins, cookies, arbitrary JS, user proxies, downloads or external index writes.

Proposed standard-tier USD0.012 per complete table check, including unchanged. Persist dataset then baseline then charge, not atomic across services. Post-persistence failure fails run; consumer must reconcile failed output. No concurrent same-scope runs. Store raw source URL mapping in your own pipeline. Cloud PPE needs actual pricing configuration and verification.

Local: Node22.12+,npm ci,npm test,apify validate-schema. Read PROJECT_STATUS for current gates. No revenue, completeness, legal evidence or anti-bot bypass promise.