Methodology

Show the change.
Keep the context.

A transparent account of what ContractDelta calculates, what it deliberately leaves alone, and where a researcher must look next.

01

Source fact

A value, date, party, or category reported in a public AusTender OCDS release.

02

Calculated fact

A deterministic before-and-after difference, percentage, or calendar-day count.

03

Context

Notice sequence, agency, supplier, category, timestamps, and primary record links.

04

Interpretation

Reserved for researchers after reviewing source documents and surrounding evidence.

Collection

Evidence begins at the source

Scheduled runs query the public AusTender OCDS API by latest-modified date. Every opaque next-page link is checked for HTTPS, the official host, and the expected path before it is followed.

A date-range response can contain an amendment without its parent release. In that case ContractDelta fetches the full Contract Notice history before comparing values. This prevents an amendment from being mistaken for a newly published contract.

Each interval identifies explicit amendment contracts, ranks them by current reported value with deterministic source-time and contract-ID tie-breakers, and hydrates only the highest-ranked bounded set. The data page publishes the discovered, selected, and skipped counts for the latest interval. This is a reproducible coverage boundary, not an exhaustive census of every amendment.

Normalisation

Compare meaning, not formatting

  • Currency values use exact two-decimal arithmetic.
  • Contract timestamps are converted to the Canberra calendar date.
  • Source modification timestamps remain in UTC.
  • Whitespace, empty values, and HTML entities are normalised.
  • Explicit `-A#` identifiers define amendment order.
  • Contact details, ABNs, and addresses are discarded.

Comparison

One event per changed field

Value changeabsolute = current − previouspercentage = absolute ÷ previous × 100No percentage is calculated when the previous value is zero or absent.

Each event records its before and after release IDs, normalised SHA-256 hashes, source timestamp, official record URL, OCDS JSON URL, and exact old and new values. Formatting-only changes produce no event. Null-to-value transitions do.

Quality controls

Fail closed when confidence falls

Publication stops when validation failures exceed 2%, all releases fail, pagination exceeds its bound, or a pagination link crosses the official source boundary.

Publication also stops if the source publisher or data licence changes unexpectedly. Temporary source failures are retried, and a temporarily missing record never deletes preserved history.

Each successful run publishes a manifest with counts, failures, source URLs, parser version, and dataset hashes. CSV text is neutralised when it could be interpreted as a spreadsheet formula; JSON retains the exact normalised value.

Every published source path is validated before a request. The weekly update availability-checks a deterministic sample of up to 100 recent destinations. Definite missing-record responses stop the run; redirects, access controls, rate limits, server errors, and network failures are reported as indeterminate for manual review.

Limitations

What the data cannot prove

  • Cause. A modification can be a correction or legitimate variation.
  • Expenditure. Reported contract value is not annual or realised spending.
  • Identity. A supplier name can change without a legal-entity change.
  • Complete scope. The primary UNSPSC code does not describe every deliverable.
  • Coverage. Per-interval selection is capped and prioritises higher current reported values.
  • Misconduct. No change, count, or magnitude is evidence of wrongdoing.
  • Privacy. A supplier name can identify an individual even though contact details and addresses are excluded.

Responsible use

Investigate carefully

ContractDelta identifies changes in publicly reported procurement records. A change is not evidence of misconduct. Contract amendments can result from legitimate changes in scope, timing, pricing, reporting, or data correction.

Inspect primary records, seek surrounding documents, publish uncertainty, and give affected organisations a fair opportunity to provide context.