Method · scope · provenance

What the evidence says—and where it stops.

Crapwatch is designed around a simple rule: a public record may establish an event, a reported zero, or a gap. Those states are different, and the interface must never blur them.

The source corpus

The current analysis starts with the operator’s published 2024–25 Annual Environmental Reviews (AERs) for Tasmania’s 77 EPA-regulated Level 2 sewage treatment plants. These are public disclosure PDFs, captured with their source URL, retrieval time and SHA-256 hash.

What the corpus does not cover

It is not a complete sewer-network or pump-station event ledger, a live telemetry feed, an independent sampling program, or a statewide public-health assessment. No missing category is inferred into existence.

Bypass and treatment performance answer different questions

Bypass evidence records occasions when a reported flow did not follow the usual treatment path at a Level 2 plant. Treated-effluent evidence reports sampled discharge statistics against applicable limits. Reuse evidence concerns a separate reuse stream and is kept out of the treated-effluent result.

A zero in the bypass section is therefore not a clean bill of health. In the current analysis, 61 plants reported zero bypass events; 53 of them still had at least one applicable treated-effluent statistic below 100%.

Itemised records are not count-only reports

The AERs provide 200 itemised bypass-event records, with event-level fields such as time, duration or volume where published. Rosebery reports 94 events and Ti Tree Bend 139 only as aggregate counts, with dates available on request. Those 233 events have lower temporal resolution.

The evidence-state vocabulary

These labels describe what the captured source establishes—not what probably happened in the world.

▮ EVENT_OBSERVED: The source reports an event and supplies an itemised record.
— ZERO_EVENTS_REPORTED: The AER explicitly reports no plant bypass events for the period. This does not establish no network spills or no treatment-performance issue.
╱ EVENT_TIMING_UNAVAILABLE: The source gives an aggregate event count but does not publish event dates.
? UNKNOWN / NOT_PUBLISHED: The public corpus does not establish the answer, or says the detail is not published. Neither state means zero, safe or clean.

Geometry carries its own certainty

Authoritative geometry reproduces a location or shape supplied by an identified source. Derived geometry is constructed analytically—for example, a relationship or pathway inferred from separate sources—and must be labelled as derived rather than allowed to inherit authority from nearby data.

The geospatial observatory exists, but public map data is withheld while source-geometry redistribution rights are being verified. The public site includes no geometry or map payload.

Every result should be reproducible

Extraction retains document, page, raw source text and typed status. Unrecognised values are marked rather than guessed. Dependencies and input hashes are pinned; committed outputs are checked against a declared analytical baseline; mismatches fail closed.

Where to inspect the audit trail

The repository contains PROVENANCE.md, BUILD_MANIFEST.json, MANIFEST.sha256, DATA_QUALITY.md, CLAIMS_LEDGER.csv and VERIFICATION_RECEIPTS.md. The repository record is authoritative if this plain-language summary ever diverges.

Why the historical 55 became 53

The historical wall result counted Brighton and Penna among zero-bypass plants with a below-100% statistic. Later scope repair identified their relevant rows as reuse-only, not treated-effluent discharge statistics. Keeping the streams separate removes those two plants from this particular conjunction: 55 became the current canonical 53.

This is a scope correction, not a claim that either plant became cleaner. The pinned wall remains a reproducible historical artefact; the current figure is used in the surrounding public shell.