How this is built

What is counted

The unit is an obligation: something a government undertook to answer, owed to a body with standing to demand it, by a date that can be established from the record. A register is admitted to this site only when all three exist — a stated obligation, a claimant, and a date. That is what allows this site to report rather than argue.

The Senate is the first claimant, not the only one. It has required a response within three months of the report being tabled since 14 March 1973. A report leaves the register when a response is tabled, whatever that response says — the register measures whether an answer was given, never whether it was a good one.

Planned registers: Western Australian parliamentary committee reports. Each will be published as its own register with its own rule, not merged into this one.

Sources

Six, all public:

• The President of the Senate’s report on the status of government responses to parliamentary committee reports, tabled twice a year. It lists every committee report that requires a response and records, against each, whether a response has been received. The authoritative statement of what is owed.
• The Speaker’s schedule of the status of government responses to committee reports, presented to the House at roughly six-monthly intervals. The same instrument for the other chamber, and the basis of the House register.
• The government’s own status report on responses to committee reports, published for each chamber alongside the presiding officer’s and covering the same date. It is where the sentence “the Government’s response is being considered” is actually written. It is read to say what the government has said about a report — never to add a report to a register or remove one.
• The Senate’s live tracker of government responses, carried only for reports tabled after the President’s report — the window he has not yet covered.
• The Tabled Documents database (OTD), through its public API, for committee reports and for the text of every government response.
• The Senate’s own register of responses, 2000 to date, which supplies the compliance record and the dates behind the history chart.

The register follows the President. His report is fetched from the Tabled Documents database, not typed in by hand, and its status column decides every row: a date means the response has been received and the report leaves the register; a dash means nothing has been received; “Interim” means an interim response; and “Interim*” means — in the definition his own report gives — that the government has said only that “the Government’s response is being considered”. Reports the President records as answered are not published here as outstanding, and reports he lists are not filtered out by us.

The House register works the same way, from the Speaker’s schedule rather than the President’s report. Its status column reads slightly differently — a date, “no response to date”, “partial response received” or “fully responded”. Two further differences matter. The Speaker states whether each report is past its deadline, and we take his word for it rather than counting six months off a calendar, because the 2010 resolution excludes any period when the House was dissolved. And his schedule is rolling: a report is dropped from later schedules once its response arrives, so a report still listed against an older response date is still owed something, and is published here as outstanding — as his own tally counts it. That tally, printed on the schedule’s fourth page, is what our reading of the document is checked against before the register is written.

The tracker cannot add to, subtract from or overrule any of that. It is carried for one purpose: reports tabled after the President’s as-at date, which he has not yet reported on. Those rows are marked as the tracker’s in the register’s source column, and on many builds there are none. Where a response has been tabled since the President’s date, the report is removed and the response named. The build refuses to write a file if the President’s report comes back short, so a failed fetch cannot quietly shrink the register.

Currently 84 reports are outstanding, and 1,486 responses tabled between 2000 and 2026 can be measured against the deadline. A further 81 cannot: for those responses — almost all of them tabled in 2000, 2001 and 2004 — the register does not record when the report was tabled, so no deadline can be computed. They are excluded rather than counted as failures. The practical effect is that the first two years of the register are barely represented in any compliance figure on this site.

The response corpus

Separately from the Ledger, this site examines the text of every government response tabled since mid-2022 — 623 documents — to ask what the response actually did.

The database category Government response is broader than that question. It also holds documents that answer somebody other than a parliamentary committee, or that are not responses at all. 35 documents are excluded by identifier, each with a recorded reason:

All but one of these would otherwise have been counted as substantive responses, so the exclusions raise the reported failure rate rather than lowering it. The list is published in full, with reasons, in the dataset.

How responses are classified

One test is applied automatically, and it is a narrow one. A response is a form-letter closure when it disposes of the committee’s recommendations with a repeated template sentence and addresses none of them. Of 623 responses, 260 are form-letter closures (41.7%), 5 use the template for some recommendations but not all, and 358 do something else.

A stratified sample of 56 responses was checked by hand against the source documents, deliberately weighted towards the classes most likely to be wrong. That review is what produced the scope exclusions above, and the list of undetected forms below.

What this measure does not catch

A form-letter closure is only the most conspicuous way of answering without answering. There are at least three others. None is detected automatically, and each is named here with a document anyone can download and check.

They are undetected for one reason, and it is the same reason in all three cases. A form-letter closure gives itself away because the template repeats: the same sentence appears once per recommendation, and a machine can see that without knowing anything about the report. The other three forms are written in ordinary prose that varies from document to document, and nothing in the text marks them out. Read on its own, a response that says “the Government notes the recommendations of the Committee” and then describes a policy is indistinguishable from a real answer — unless you already know what the committee recommended, and can check the policy against it.

That is the missing ingredient: the response alone is not enough evidence. You cannot tell from its own words whether it disposed of five recommendations or forty, or whether what it describes answers them. Only the report says what was owed. So these are not gaps waiting for a cleverer pattern — a longer list of template phrases would only chase the wording of the excuse, and there would be a fifth form next year. They are the same gap, and it closes when the reports are parsed as well as the responses.

1. The passage-of-time closuredetected

   The response notes each recommendation and declines to address it because the report is old.

   Current and future impacts of climate change on housing, buildings and infrastructure — 65 recommendations, the same sentence 65 times, no position stated on any of them.

2. Superseded by eventsnot detected

   The recommendations are noted as a group because the thing they concerned has been withdrawn or replaced.

   Nature Positive (Environment Protection Australia) Bill 2024 — 40 recommendations noted together; the bills were withdrawn in February 2025.

3. Answered by substitutionnot detected

   The recommendations are noted as a group, and the response describes a different package of measures instead.

   ‘You win some, you lose more’: online gambling — 31 recommendations noted as a group; not one addressed individually.

4. Enumerated without a positionnot detected

   Each section lists the recommendations it addresses, then describes existing policy without accepting, rejecting or qualifying any of them.

   Senate Select Committee on Adopting Artificial Intelligence — All 13 final report and all 5 interim recommendations mapped to a section; a position stated on none.

This is why the published claim is stated narrowly. It is not that 260 responses answered nothing; it is that 260 used a form-letter closure that addressed no recommendation. The true figure for “answered nothing” is higher, and the way to reach it is to ask a better question.

That question is coverage: for each recommendation a committee made, did the response state a position on it? It is countable, it is neutral about whether the position was right, and all four forms above score zero against it. Building it requires parsing the reports for their recommendations as well as the responses. It is the next substantial piece of work on this project, and it is not published yet.

Known limits and open questions

Which reports attract the three-month rule. The President’s report sets out what it leaves out: reports on bills whose recommendations could only be dealt with during debate on the bill, estimates and annual-report scrutiny, the Senate’s domestic committees, Selection of Bills, and — usually — Public Works. Some reports the President nonetheless lists appear to fall inside those categories. The two readings cannot both be right. We have written to the Senate Procedure Office asking which is correct and will publish the answer here. Until then the register follows the President’s report, because that is the Senate’s own statement of what is owed.

Sitting days. A response can only be tabled when the chamber is sitting, and three months can contain a long adjournment. That is a fair objection to the compliance figure and we have not tested it against the sitting calendar. It can be bounded instead: allow twice the deadline and 78% of responses still failed; allow a full year and 55% still failed. An adjournment can explain part of the gap between three and six months. It cannot explain a median of 420 days.

Compliance counts only responses that arrived. Reports still waiting are not in the denominator, so a report that is never answered never counts as a failure. The compliance rate therefore overstates performance, as does every other figure here.

Attributing a late response to a government can be done two ways — by who tabled the response, or by who was in office when the report was tabled — and they give different orderings. Both are published on the Senate register, side by side, because choosing one would be choosing a result.

Interim reports are excluded from the outstanding count. Governments ordinarily respond to a final report, and counting an interim report as separately owed would overstate the backlog.

The report index is Senate-weighted. The harvest of committee reports is drawn from the Tabled Documents database and covers Senate committees far more completely than House committees. This does not affect the Senate register. It does mean that where a response answers a House report, this site can identify the response but often not the report behind it — a gap to close before the House register is published.

History before this site existed is reconstructed. Points on the history chart marked reconstructed are recomputed from report and response dates each time the dataset is rebuilt, so they can move if a source record is corrected. Points marked observed are what the register showed on the day and are never recomputed.

Refresh. The dataset is rebuilt by hand and every page shows the date of the data it was built from. Automating the harvest is in progress; until it is running, this site does not claim a fixed refresh interval.

Corrections

If a row here is wrong, we want to know, and the fastest way to change it is to tell us. Write to corrections@toohardbasket.org.au with the row and the source. Corrections are made within 24 hours of being confirmed and listed in the corrections log, alongside the dataset’s full change history, so the record of what this site said, and when it changed, stays public.

The full dataset and the code that builds it are in the public repository — data under CC BY 4.0, code under the MIT licence. Every figure on this site is produced by that code from those files. Nothing is typed in by hand.