Who governs the organisations, and what does that connect
Across Australia, who sits on the boards of Aboriginal community-controlled organisations working in justice, what else do those people govern, and does the sector connect through shared leadership or is it genuinely locally held?
Where this is up to · updated 2026-08-21
3 of 6 passed · now: Adversarial checkGate 3 done: run at 99.6% coverage, prediction recorded before the run and supported
What it found
- The sector is locally governed, and now that is measured rather than assumed 91.9% of people governing established Aboriginal community-controlled organisations hold no other board seat in this dataset. An identical-looking result was produced this morning at 1.8% coverage and discarded; the difference between a finding and an artefact was the coverage, not the number.
- Where it connects, it connects on Country The multi-board clusters are regional. Broome Aboriginal Media with Kimberley Aboriginal Media and Gogolanyngor RNTBC. Dhimurru with Gumatj and the Aboriginal Sea Company. Dugalunji with Indjalandji-Dhidhanu RNTBC and Bulugudu around Mount Isa. Organisational families in one place, not a national elite.
- One link runs local to national A single person holds roles across Ngaanyatjarra organisations, the Aboriginal Health Council of Western Australia and NACCHO: community organisation to state peak to national peak, spanning three states. It is the clearest governance path from a remote community to the national table in this data.
- Trustee boards look identical to governance and are not Four names recur together across the same Western Australian native title charitable trusts. The repetition is the signal: community governance produces overlapping but distinct boards, a professional trustee arrangement produces the same people on the same entities. Distinguishable structurally, but only a person should say which is which.
The preregistered prediction
HELD. Predicted before the run: most organisations governed by people holding no other board seat, and the sector connecting through regional organisational families rather than individuals spanning states. Result across 11,976 people and 13,481 roles at established Aboriginal community-controlled organisations: 91.9% hold exactly one seat, 969 hold more than one, 183 hold three or more, and only 139 span a state line. Both halves supported.
Registration
question, predictions and stop conditions locked 2026-07-31, before running anything
Nominee-block detection
codir-graph-v1 already flags 19,403 roles as nominee blocks, clusters of up to 740 roles under 142 identities. These are corporate trustee arrangements, not people, and any board-seat count that includes them is measuring law firms.
Identity guard
isEstablishedIndigenousOrg() restricts the organisation set to flags backed by a register or the organisation's own statement. Without it the set includes the Anglican Diocese of Brisbane.
Single-organisation resolution
202,541 roles where a name appears at exactly one organisation, so no ambiguity is possible. 197,724 identities, method single-org-v1, kept separate from codir-graph-v1 so provenance stays traceable. Graph coverage 8% to 68%; at established Aboriginal organisations 1.8% to 59.8%.
Refusing to merge what cannot be told apart
Names appearing at 20 or more organisations are not merged into one identity. Each role is held separately at low confidence, method high-degree-refused-v1. A refusal to merge is a better record than a wrong merge, and 42 roles landed here after nominee blocks were already excluded.
Authoritative resolution, not clustering
person_roles.properties carries person_uuid on 334,572 rows: the ACNC's own person identifier, 269,859 distinct people, and only 83 uuids showing more than one name spelling. The source had already resolved what an afternoon of clustering was rebuilding. Coverage at established Aboriginal organisations 99.6%, holding by basis: ORIC 99.9%, ACCO 100%, self-identified 97.9%.
Coverage is the headline, not a footnote
A first pass resolved 246 of 13,376 roles at established organisations, 1.8 per cent, and returned three people with multiple seats. That result is about coverage, not about the sector, and was discarded.
Coverage is still below any publishable floor
59.8% at established organisations is better than 1.8% and still not enough to say whether the sector connects through shared leadership. The floor has not been set, and until it is, no connection claim leaves the lab.
Entity resolution at scale
codir-graph-v1 has run on 27,851 of 339,698 roles. It produces sensible clusters where it runs. It needs to run across the rest.
A coverage floor
No connection claim is publishable below a stated resolution rate. The floor is a decision, not a default, and it goes on the page.
A human read of the top clusters
Professional trustees and community leaders wear the same shape. Ian Westley across seven WA native title trusts is a structural role; Preston Thomas across the Ngaanyatjarra organisations is community governance. Only a person can tell them apart.
This board is the honest state of the experiment. Nothing gets called finished until it survives the checks, and the gaps are listed because someone reading this might be able to close one.
Data sources
person_roles · 339,698 directorships and office-holders from ASIC, ACNC and ORIC registers periodic
person_identities (codir-graph-v1) · 27,851 resolved roles with cluster size, confidence and a nominee-block flag on demand
organizations + gs_entities · identity by name+ABN, restricted to register-backed or self-declared Aboriginal control periodic
Method, in plain words
Registered before running, per the lab rule. THE PREDICTION, stated first: after excluding nominee blocks, most Aboriginal community-controlled organisations in this dataset will be governed by people who hold no other board seat in it, and the sector will connect mainly through regional organisational families rather than through individuals spanning states. THE REFUSALS, decided in advance: no person is named on a public surface unless they hold a public role in a register; no cluster is presented as a finding while resolution coverage is below the stated floor; nominee blocks are excluded from every count and the exclusion is reported; and a low connection count is never published as evidence of local control until coverage can rule out the alternative explanation, which is that we simply have not resolved the graph.
Challenge this
Every figure has a receipt. If you can break one, we want to know, and the correction publishes here at the same volume. Challenge a figure.