The idea
Inspired by ADRI, built fresh for AI
The Australian Disaster Resilience Index scores every Australian LGA on coping capacity and adaptive capacity across eight themes and 82 indicators, social character, economic capital, infrastructure and planning, emergency services, community capital, information access, governance and leadership, and social and community engagement. It is peer-reviewed, government-published, and already at the geography you need.
The structure of the ADRI is hazard-agnostic. It does not measure how much bushfire is coming and instead measures whether a community can absorb a shock and adapt afterwards. That is what transfers to the QAIRI.
What people are actually worried about
The hazard taxonomy, from the authoritative source
We used the International AI Safety Report 2026, published 3 February 2026, led by Yoshua Bengio with over 100 experts. The expert panel nominated by more than 30 countries including Australia. The report sorts risks into three categories.
Malicious use
Someone intentionally uses AI to cause harm. Cyberattacks, deepfakes, and biological weapons uplift. The report notes AI agents identifying 77% of software vulnerabilities in competition settings.
Malfunctions
Systems fail or operate outside intended parameters. Hallucinations, evaluation gaming, and loss-of-control behaviours including sandbagging and reward hacking.
Systemic risks
Broad societal harms from widespread deployment. Labour market disruption and threats to human autonomy.
Provenance
The six component weights
The three-category taxonomy is sourced from the International AI Safety Report 2026. The specific outcomes tested against it are not. They are a derivation: the taxonomy mapped onto the things Australian local government actually touches. Nobody has published a local-government AI hazard register, so there was nothing else to cite.
The derivation follows a stated selection rule, applied to six candidate hazards. A hazard earns a place only if all four hold:
| Test | Why |
|---|
| 1 · Traces to the taxonomy | It is a recognised category of AI risk, not one invented to fit the data. |
| 2 · Lands through a local channel | Local government either controls the channel, or the community's exposure through it differs by place. |
| 3 · Varies measurably by LGA | There is public data that separates one council from another. A uniform national risk belongs in a national report, not this one. |
| 4 · Somebody local can act | A risk a council can do nothing about produces a scary map and no decision. |
Why water and electricity are scored differently. Water and sewerage pass test 2 directly: in regional Queensland the council is the utility. Electricity does not, because distribution and generation are state-owned and no council runs a grid. Who operates the asset is a separate question from how exposed the community is, and the index measures the second. Queensland has communities served by isolated electricity networks rather than the national grid, and an isolated network is a materially different risk object from a grid connection, regardless of who holds the licence. Essential services scores water and sewerage provision and isolated-network dependence as separate inputs.
Remoteness
Remoteness and the QAIRI
Across the six components, risk in remote areas according to the QAIRI varies. There are benefits to being a small, remote community such as opportunity to seal entry and exit which a large urban area could not; or having the social infrastructure in place to check and verify things by council staff visiting home by home.
Remoteness uses exposure, absorptive capacity and consequence combined to create this component.
| Component | Exposure | Absorptive capacity | Consequence | Net |
|---|
| Essential services | Neutral | Worse, one plant, one operator | Worse, days to restore | Remote worse |
| Population vulnerability | Cross-cutting | Cross-cutting | Worse, disadvantage compounds | Remote worse |
| Institutional capacity | Better, smaller attack surface, fewer systems | Better, manual fallback is real | Worse, no IT capability to respond | Plausibly remote better |
| Deployment evidence | Neutral | Neutral | Neutral | No clear sign, excluded |
| Synthetic warning | Neutral | Better, short verification chain, everyone knows the mayor | Worse, fewer channels, thinner coverage | Genuinely ambiguous, left out of the formula on purpose |
| Systemic labour | Worse, single-industry concentration | Worse, thin labour market | Worse, no alternative employer | Remote worse |
Institutional capacity inverts entirely: a highly digitised metropolitan council with two hundred integrated systems and no manual process is more exposed to institutional failure than a shire that still runs on paper. Synthetic warning's remoteness sign is genuinely ambiguous rather than resolvable either way, so the formula leaves remoteness out of that component rather than force a direction onto it.
The validation that matters
The actual result, not a prediction of one
Reusing ADRI's own inputs and assumptions would only reproduce ADRI with extra steps, rediscovering that remote communities are disadvantaged. The test is not an argument. It is a number, computed against ADRI held entirely out of the model and used only for comparison.
Spearman rank correlation between QAIRI and ADRI's resilience score: r = -0.408. Comfortably below the 0.9 threshold that would mean the index adds nothing, and below the 0.7 threshold that would mean it is merely a variant of the same thing. The two rankings are substantially different.
Highly digitised metropolitan councils rank worse on QAIRI than they do on ADRI, because system dependence is a strength in a natural-hazard index and a liability in this one. Townsville, Brisbane, Sunshine Coast, Moreton Bay and Livingstone all shift more than fifty ranks worse on QAIRI than their ADRI position would predict.
Two real sources of divergence turned out to matter, and one turned out to cut the other way:
| Novel variable | Natural-hazard analogue | What actually happened |
|---|
| Institutional capacity / digitisation | Inverted, infrastructure is a strength in ADRI | Confirmed. This is the largest single driver of the metro councils ranking worse on QAIRI than on ADRI. |
| Systemic labour (AI-exposed occupations) | None at all | Cuts the other way. AI-exposed occupations concentrate in wealthy metro councils, the same councils ADRI already scores as resilient, which pulls correlation back toward zero rather than away from it. See Limitations. |
| Vendor concentration | None at all | Not measured. The one data source tried had no variance across all 78 councils. See Limitations. |
The pre-registration held, provisionally. Both stated predictions, metros ranking worse and correlation under 0.9, came true. One of the six components, systemic labour, was not anticipated: it pushes correlation back toward zero rather than away from it, and is reported here rather than excluded.
The register
Six components, as actually built
Formulas below match config/index.yaml exactly. Two of the original six candidate hazards did not survive the selection test in section above and are not here: a biological-event factor (travel time to intensive care) failed test 4, health access is not a council function in Queensland, and an automated-decision-failure factor never had a workable dataset. Both are documented in the Limitations page.
ESMalicious use · Malfunctions
weight 22%
Essential services
Water, sewerage or an isolated electricity network is compromised or interrupted in a community with no alternative supply and no rapid restoration path.
Scored factor, unregulated essential-service dependency
Under the Security of Critical Infrastructure Act 2018, a critical water asset is one serving at least 100,000 connections, and only then does the responsible entity owe a critical infrastructure risk management program. Almost every Queensland council that operates water and sewerage falls below that line: these councils run critical infrastructure that is not regulated as critical infrastructure.
Electricity is scored as exposure rather than operation. No Queensland council runs a grid, yet a community on an isolated network rather than the interconnected grid is a different risk object, a property of the place, not of who holds the licence, so it is not paired with a "provider" input the way water is.
essential_services =
0.30 × above_soci_water_threshold (false_worse)
+ 0.25 × is_water_service_provider (true_worse)
+ 0.25 × isolated_power_network (true_worse)
+ 0.20 × mean_remoteness_score (higher_worse)
Data
- Water and sewer connections per council, QLD comparative information report, 70/78
- Isolated electricity networks, Ergon's own published list, 78/78 verified
- Remoteness class, ABS ASGS, 78/78
Why it varies by LGA
South-east councils have separated water into distributor entities. Regional and remote councils are the utility, often with one plant, one operator, and days of road access.
PVCross-cutting
weight 18%
Population vulnerability
Who absorbs any shock badly, regardless of what caused it.
Scored factor, validated disadvantage measures
Every input here is a validated vulnerability measure already used for natural hazards. It does not depend on any particular AI scenario, which is why it belongs under a hazard-agnostic architecture: it would score similarly for a flood or a pandemic.
population_vulnerability =
0.40 × seifa_irsd_score (invert, lower = more disadvantaged)
+ 0.35 × avg_persons_per_bedroom (higher_worse)
+ 0.25 × indigenous_share_pct (higher_worse)
Data
- SEIFA 2021, ABS, 78/78
- Census 2021 crowding, 78/78
- Indigenous share, derived from Census, 78/78
Why it varies by LGA
Cape York and Torres Strait communities score at the extreme on every input simultaneously. That is not a modelling artefact, it is the actual situation.
ICMalfunctions
weight 25%
Institutional capacity
The council itself is incapacitated, by intrusion or by dependence on a single failed system, and has no internal capability to run the service manually.
Scored factor, institutional thinness
Two things determine whether a council can survive its systems going down: how many people it has indoors, and how much of its money it controls. Grant-funded money is committed money; own-source revenue is the only discretionary capacity a council really has, and Indigenous councils have almost none because DOGIT land is largely not valued and therefore not rateable.
This is the component that inverts. A shire that still runs half its processes on paper has a smaller attack surface and a real manual fallback; a metropolitan council with two hundred integrated systems and no non-digital path has neither. Digitisation is a strength in the natural-hazard index and a liability here. This inversion is the largest single driver of QAIRI diverging from ADRI, confirmed in the validation section above.
institutional_capacity =
0.50 × own_source_revenue_share (invert)
+ 0.30 × indoor_staff_share (higher_worse, digitisation proxy)
+ 0.20 × staff_per_1000_residents (invert)
Data
- Personnel, indoor and outdoor FTE, QLD comparative information report, 73 to 76/78
- Financial inputs and rate revenue, same source
Not measured
Vendor concentration, a single vendor outage hitting many councils at once, has no natural-hazard analogue and would be the cleanest source of divergence available. Not scored here. See Limitations.
DEAI race (Hendrycks)
weight 15%
Deployment evidence
Confirmed AI already running in council operations, the adoption-pressure signal.
Scored factor, confirmed AI in the field
Press-derived, and only 8 of 78 councils are confirmed. "No evidence found" is scored as missing, not as safe. A council with no press coverage is not confirmed free of AI, it is simply not covered. Missing values are excluded from this component and the remaining components are renormalised, rather than counted as zero risk.
deployment_evidence =
1.0 × ai_deployment_confirmed (true_worse, missing if unconfirmed)
Data
- News and council disclosures, 8 confirmed, 8 AI-adjacent, 62 no evidence found
Why it's weighted low
This is the closest fit in the whole index for Hendrycks' AI race category, competitive pressure to deploy fast. It is weighted low deliberately: the evidence base is thin by construction.
SWMalicious use
weight 10%
Synthetic warning
A fabricated, AI-generated evacuation message during a live emergency, with no second channel to check it against, moving people toward a hazard instead of away from it.
Scored factor, warning-channel exposure
The only component in the whole index whose scored factor genuinely depends on AI as the cause, not a general-vulnerability proxy wearing an AI label. Weighted by how often the LGA is actually in an emergency, how many channels exist to check a warning against, and whether the official channel reaches non-English-speaking households.
Deliberately does not use ADRI's own information-access theme as a channel proxy, even though it is the closest fit, because that column is held out as the control and using it here would leak the control into the model. Remoteness is deliberately left out too: the correction above found its sign genuinely ambiguous for this component, not a defensible single direction.
synthetic_warning =
0.40 × disaster_events_alltime (higher_worse)
+ 0.35 × mbsp_stations_per_1000_residents (invert, more coverage = more channels)
+ 0.25 × pct_language_not_english_home (higher_worse)
Data
- Disaster Recovery Funding Arrangements activations by council, Queensland Reconstruction Authority, 2010 to 2011 onward, 78/78
- Funded mobile base stations, DITRDCA Mobile Black Spot Program, 78/78
- Non-English-speaking households, Census 2021, 78/78
Why it varies by LGA
Brisbane has radio, television, three newspapers, an SES presence and a council comms team. A Cape York community in cyclone season may have one Facebook page and a satellite link.
SLSystemic risk
weight 10%
Systemic labour
Employment concentrated in occupations most exposed to generative AI, in an LGA with a rates base that depends on that employment.
Scored factor, AI-exposed occupation share
Bengio's third category, systemic risk, was almost entirely absent from the index until this component. Proxy: share of employed persons in Professionals, Clerical and Administrative Workers, and Sales Workers (ANZSCO major groups 2, 5, 6), the occupations most commonly cited in generative-AI exposure research.
systemic_labour =
1.0 × pct_ai_exposed_occupation (higher_worse)
Data
- Census 2021, occupation by LGA, ABS SDMX API (table G61), 78/78, real range 19% to 53%
Proxy limitations
One judgement call about which occupations count as exposed, not a validated index, and not cross-checked against a second published exposure measure. It also correlates with council wealth and capacity, which pulls the overall index toward ADRI rather than away from it. See Limitations for the full discussion.
The other half
Mapping ADRI's capacity themes to AI
Exposure is only one side. These are the eight ADRI themes, and what each becomes when the hazard changes.
| ADRI theme | Capacity | AI equivalent | Available indicator |
|---|
| Social character | Coping | Who can detect and challenge a wrong automated answer | Age structure, education, English proficiency, disability |
| Economic capital | Coping | Can the council fund remediation without a grant round | Own-source revenue share, rate revenue per capita |
| Infrastructure & planning | Coping | Connectivity redundancy and manual fallback | Mobile/NBN coverage, single-road access, backup channels |
| Emergency services | Coping | Local disaster management capability and plan currency | LDMP currency, IGEM assessment status, SES presence |
| Community capital | Coping | Trusted intermediaries who can correct a false message | Local organisations, community-controlled services, media |
| Information access | Coping | Whether a resident can independently verify anything | ADII access / affordability / ability sub-scores |
| Governance & leadership | Adaptive | Is there a policy, a delegation, a human rights process | Published AI policy, delegations register, s 58 process |
| Social & community engagement | Adaptive | Can the community influence what gets deployed on it | Consultation mechanisms, community-controlled governance |
ADRI itself has 67 of its indicators in coping capacity and far fewer in adaptive capacity, because national-scale adaptive indicators are hard to find. The same scarcity applies to AI-specific adaptive indicators at the LGA level.
Status
What data QAIRI is actually built on
All six components live, all 78 Queensland LGAs, in data/qld_lga_master.csv.
| Component | Coverage | Note |
|---|
| Essential services | 70 to 78/78 | Water threshold and provider flags are 70/78, SEQ councils that separated water into a distributor entity; isolated power and remoteness are complete |
| Population vulnerability | 78/78 | Complete |
| Institutional capacity | 73 to 76/78 | Staff and finance figures are 2015 to 2016, current years blocked at source, see Limitations |
| Deployment evidence | 16/78 confirmed either way | The rest excluded as missing, not scored as safe |
| Synthetic warning | 78/78 | Disaster frequency from QRA's own activation records, authoritative, not approximated |
| Systemic labour | 78/78 | One occupational proxy, not cross-validated, see Limitations |
The honest headline. Every component is scoreable and scored, but "scoreable" is not the same as "settled." Deployment evidence, synthetic warning and systemic labour all rest on thinner, more contested evidence than essential services, population vulnerability and institutional capacity, and are weighted lower for exactly that reason. Full account: the Limitations page.
ADRI is the control, not an input
ADRI's composite is built from 77 indicators, overwhelmingly ABS Census, named variables include AGEP, SEXP, HCFMD, HEAP, LFHRP, OCCP, VEHD, VOLWP. The raw indicator values are not published, the API serves only the eight themes, the two capacities and the composite. Verified by reading the application's own bundle.
Every adri_* column is held out of the model and used only as the comparison. The overlap is heavier than it looks: median income, age structure, household composition and SEIFA all draw on the same Census variables ADRI does, and council staff per dwelling is itself an ADRI indicator. Genuinely independent of ADRI: own-source revenue share, water connections, the SOCI flag, and Indigenous council status. That is a thinner independent signal than it first appears.
Sign inversions, the actual contribution
This does not require building a competing index. ADRI's own indicators are reused, with the sign flipped where the causal mechanism inverts. Every flip below traces to a stated argument, so the divergence is derived rather than manufactured.
| Indicator | Natural hazard | AI | Why it flips |
|---|
| Connectivity (information_access) | Good, the warning gets through | Bad | It is the entire exposure surface for synthetic content, scams and AI-mediated services. Partly offset: connectivity also enables verification |
| Education, % managers and professionals | Good, literacy, resources, planning | Bad | Professional and managerial work is the most automatable. The cleanest flip in the set, and it moves metro councils |
| Dwellings per FTE council staff | More staff = capacity | Inverts | ND reads staffing; AI reads system dependence. Many staff and few systems means real manual fallback |
| Remoteness (distance to facility / airport) | Bad, slow everything | Partly protective | Sealable, low digital penetration, fewer systems to compromise. Consequence still worse |
| Volunteering, residence >5 years | Good | Good, under-weighted | Not a flip, a re-weight. These are the mechanism of social verification: a fake can be checked by asking someone |
| % caravan & improvised dwellings | Bad, structural vulnerability | Bad, different reason | Exactly the false-positive class for aerial change-detection compliance enforcement |
One line of attack that turned out not to work. ADRI 1 measured connectivity as % area with excellent or good ADSL cover, from 2016 broadband data, which looked like an exploitable staleness. It is not. Comparing ADRI 1 and ADRI 2 across matched areas, information_access correlates at only 0.605 and its mean rose from 0.486 to 0.571, exactly what the NBN rollout would produce. The indicator was refreshed; that line of attack does not hold. (Caveat on our own method: the two versions use different ASGS boundary editions, so the correlation is contaminated; the means are not.) Incidentally govt_leadership moved most of all, at r = 0.426, nobody appears to have looked at why.
Constraints on interpretation
What this index cannot do
This index cannot be validated against historical outcomes. ADRI's indicators are grounded in hazards that recur; there are decades of floods and fires to check them against. The catastrophic AI outcomes in this register have not occurred. Every weight in every formula above is a judgement, not a fitted parameter, and no amount of arithmetic changes that. An index that looks precise about an unprecedented event is the most dangerous artefact to hand a council.
Four practices follow from that.
- The composite score is not published alone. The interactive map shows the six component scores alongside it, so a council can see which factor drives its own exposure.
- The measured is kept separate from the assumed. Water and sewerage connections, remoteness, connectivity and own-source revenue are measured. The hazard scenarios are assumed.
- The weights are a control, not a constant. The interactive map lets a reader move the sliders and watch the ranking change, which demonstrates the sensitivity of the result rather than asserting one.
- The falsifier is stated. If exposure turns out uncorrelated with remoteness across every factor, the theory of the case has failed.
The core claim. Australia has a government-published, peer-reviewed index of which communities absorb shocks badly. It was built for natural hazards. Every catastrophic AI scenario in the international literature lands through the same channels: essential services, health access, warning systems, institutional capacity. The communities already known to be least resilient are the ones with the most to lose from an entirely new hazard class, and no one has checked.
Appendix
List of data sets to build QAIRI
The original 51 columns below, 78 Queensland LGAs, kept as the reference dictionary for this write-up. The live master table, data/qld_lga_master.csv, has grown since (mobile coverage, occupation, disaster events, area and density), documented in data/DATA_DICTIONARY.md rather than duplicated here.
The column that matters most is “In ADRI?”. Anything marked clean is genuine independent signal. Anything partial or indirect shares Census inputs with the natural-hazard index and will drag your correlation up on its own. Anything red must be held out of the model entirely.
| Column | What it shows | Source | In ADRI? | Coverage |
|---|
| short_name | LGA name as ADRI publishes it; the join key across every file. | ADRI | n/a | 78/78 |
| council_name | Full legal name of the council. Corrected against Wikipedia's LGA table, 33 of 78 were wrong. | Derived | No | 78/78 |
| abs_lga_name | The same LGA as the ABS names it. | ABS | No | 78/78 |
| abs_lga_code | ABS LGA code; the key to any other ABS dataflow. | ABS | No | 78/78 |
| stratum | Sampling band from remoteness: metro through very remote, plus Indigenous. | Derived | No | 78/78 |
| is_local_government | Whether it is a council at all, False only for Weipa Town Authority. | Derived | No | 78/78 |
| is_indigenous_council | One of Queensland's 17 Indigenous local governments. | Derived | No, clean | 78/78 |
| website | The council's verified website; 54 of 78 refuse automated clients. | Verified fetch | No | 78/78 |
| population_latest | How many people live there, 2025 estimate (265 to 1.375m). | ABS ERP 2025 | No | 78/78 |
| population_latest_year | Which year that estimate is from. | ABS ERP 2025 | No | 78/78 |
| population_10yr_prior | Population a decade earlier, giving growth or decline. | ABS ERP | No | 78/78 |
| census2021_total_persons | Raw Census head count, lower than ERP by design. | Census 2021 | No | 78/78 |
| census2021_indigenous_persons | Aboriginal and/or Torres Strait Islander persons counted. | Census 2021 | No, clean | 78/78 |
| seifa_irsd_score | How disadvantaged the area is; national mean 1000, lower is worse off. | SEIFA 2021 | Indirect | 78/78 |
| seifa_irsd_decile_aus | Which tenth of Australia it falls in, 1 is the most disadvantaged. | SEIFA 2021 | Indirect | 78/78 |
| seifa_irsd_percentile_aus | The same rank at finer grain, 1 to 100. | SEIFA 2021 | Indirect | 78/78 |
| seifa_irsd_decile_state | Disadvantage decile against other Queensland areas only. | SEIFA 2021 | Indirect | 78/78 |
| seifa_irsad_score | Advantage and disadvantage combined, a two-ended measure. | SEIFA 2021 | Indirect | 78/78 |
| seifa_irsad_decile_aus | National decile for that combined measure. | SEIFA 2021 | Indirect | 78/78 |
| seifa_ier_score | Economic resources alone: income, housing, wealth. | SEIFA 2021 | Indirect | 78/78 |
| seifa_ieo_score | Education and occupation alone: skills and qualifications. | SEIFA 2021 | Indirect | 78/78 |
| median_age | Median age of residents, 23 to 51 years. | Census 2021 | Partial | 78/78 |
| median_personal_income_weekly | What a typical individual earns per week, $285–1,558. | Census 2021 | Partial | 78/78 |
| median_household_income_weekly | What a typical household earns per week, $684–2,978. | Census 2021 | Partial | 78/78 |
| avg_household_size | People per household, 2.1 to 4.5. | Census 2021 | Partial | 78/78 |
| avg_persons_per_bedroom | Crowding, people per bedroom, 0.7 to 1.5. ADRI has no equivalent. | Census 2021 | No, clean | 78/78 |
| median_family_income_weekly | What a typical family earns per week, this is literally an ADRI indicator. | Census 2021 | Yes, direct | 78/78 |
| median_rent_weekly | Typical rent paid, $70–500 per week. | Census 2021 | No | 78/78 |
| median_mortgage_monthly | Typical monthly mortgage repayment. | Census 2021 | No | 78/78 |
| staff_data_year | Which financial year the staff figures are from. | Qld CDC | No | 77/78 |
| staff_fte_indoor | Office-based staff: admin, planning, corporate, IT (8–4,662 FTE). | Qld CDC 2015-16 | Partial | 73/78 |
| staff_fte_outdoor | Field staff: roads, waste, parks, water operations (22–2,896 FTE). | Qld CDC 2015-16 | Partial | 73/78 |
| staff_fte_total | Total council workforce in full-time equivalents. | Qld CDC 2015-16 | Partial | 75/78 |
| finance_data_year | Which financial year the money figures are from. | Qld CDC | No | 77/78 |
| total_operating_income_k | Everything the council brings in, $'000. | Qld CDC 2015-16 | No, clean | 76/78 |
| net_rates_and_utility_charges_k | What it raises itself from rates and utilities, $'000. | Qld CDC 2015-16 | No, clean | 76/78 |
| employee_expenses_k | What it spends on staff, $'000. | Qld CDC 2015-16 | No, clean | 76/78 |
| water_sewer_data_year | Which financial year the connection counts are from. | Qld CDC | No | 77/78 |
| water_connections_total | How many water connections the council services (0–247,200). | Qld CDC 2015-16 | No, clean | 75/78 |
| water_connections_residential | The residential subset, only 58 of 78 reported, too thin to build on. | Qld CDC 2015-16 | No, clean | 58/78 |
| sewerage_connections_total | How many sewerage connections it services. | Qld CDC 2015-16 | No, clean | 75/78 |
| adri_andri | The natural-hazard resilience score itself, 0–1, higher is better. | ADRI 2024 | It IS the index | 78/78 |
| adri_coping_capacity | Ability to prepare for, absorb and recover from a shock. | ADRI 2024 | It IS the index | 78/78 |
| adri_adaptive_capacity | Ability to learn and transform after one. | ADRI 2024 | It IS the index | 78/78 |
| adri_information_access | Capacity to engage with hazard information, telecoms and internet. | ADRI 2024 | It IS the index | 78/78 |
| mean_remoteness_score | How remote the area is, 1 metropolitan to 5 very remote. | ADRI 2024 | Partial | 78/78 |
| indigenous_share_pct | Indigenous share of the population, 1.7% to 96.4%. | Derived | No, clean | 78/78 |
| is_water_service_provider | Whether the council runs water at all. | Derived | No, clean | 70/78 |
| above_soci_water_threshold | Whether it crosses the SOCI Act's 100,000-connection line into regulated critical infrastructure. | Derived | No, clean | 70/78 |
| own_source_revenue_share | How much of its income it controls, 0 to 0.862, and exactly 0 for Indigenous councils. | Derived | No, clean | 76/78 |
| staff_per_1000_residents | Workforce relative to population, but mixes 2015-16 staff with 2025 people. | Derived | Partial | 75/78 |
The independent signal, in full. Fourteen columns have no ADRI analogue at all: Indigenous population and share, persons per bedroom, the four council finance measures, own-source revenue share, the four water and sewerage measures, and the two derived service flags.
Two coverage warnings. water_connections_residential is only 58/78, do not build a factor on it. Every Queensland council column is 2015–16 regardless of what the portal says; the *_data_year fields carry the vintage so it travels with the number. All 30 ABS and ADRI columns are complete at 78/78 and current.
Where the data comes from
Sources
Mobile Black Spot Program
Funded base stations per LGA, by carrier and round. The channel-redundancy input to synthetic warning.
infrastructure.gov.au, DITRDCALimitations
Everything named above as not measured, approximate, or a single judgement call, in one place.
docs/limitations.html