Methodology
Validation & Transparency
What we test, what we can't test yet, and what we're honest about.
Postcodes
3,193
scored
Structural
11
indicators per postcode
Exposure
6
domains
Signals
20+
automated sources
Coverage
WA + NSW
station-level, national for other signals
What's Validated
The scoring engine underneath the exposure profiles has been validated against five quantitative tests. We chose to present profiles instead of scores because a profile is more actionable but the validated methodology is what gives those profiles their grounding.
Normalisation Robustness
ValidatedMin-max vs percentile-rank normalisation compared across all postcodes. Method choice does not change rankings.
BRIC rank correlation (Spearman)
0.976
threshold: 0.85 met
INFORM rank correlation (Spearman)
0.952
threshold: 0.85 met
Rankings are robust to the choice of normalisation method. Both scoring frameworks remain highly correlated regardless of whether min-max or percentile-rank normalisation is used.
Weight Sensitivity
ValidatedIndicator weights perturbed by +/-20%. Stable under parameter uncertainty.
Min BRIC rank correlation
0.999
threshold: 0.90 met
Min INFORM rank correlation
0.996
threshold: 0.90 met
Max quadrant change rate
9.1%
threshold: 10% met
Perturbation rounds
4
Even at maximum perturbation (+/-20% on all indicators simultaneously), fewer than 1 in 10 postcodes change classification. The structural data is stable under parameter uncertainty.
External Validation (SEIFA)
ValidatedStructural data correlated with the ABS Socio-Economic Index for Areas (SEIFA IRSD). Validates that structural indicators capture genuine socioeconomic patterns.
Pearson r
0.909
threshold: 0.40 met
Spearman rho
0.912
threshold: 0.40 met
Postcodes compared
2,624
Strong correlation with SEIFA IRSD validates that our structural data captures genuine socioeconomic patterns. The gap between r=0.91 and 1.0 is the point: perfect correlation would mean the two measures are identical and one is redundant. The CRI captures dimensions SEIFA does not, particularly diversity and crisis exposure.
Diversity Weighting
ValidatedCompares standard and diversity-weighted scoring across mining-dependent postcodes. Concentrated economies are correctly flagged.
Standard vs diversity Spearman
0.997
Mining postcodes scoring lower
22 / 28
threshold: majority met
Diversity weighting reveals structural brittleness that standard scoring misses. Communities with concentrated economies appear resilient on volume-based indicators but are exposed when diversity is weighted. 22 of 28 mining-dependent postcodes score lower under diversity weighting, consistent with the hypothesis that concentration mimics resilience but creates systemic fragility.
Exposure Profile Validation
The exposure profile maps structural characteristics to exposure domains using deterministic rules. Here is what we can verify and what we cannot.
What Can Be Verified
AuditableDeterministic rules, reproducible outputs, verifiable positions.
Exposure rules are deterministic and auditable
Every exposure mapping is a deterministic function of structural inputs. No randomness, no LLM inference. You can inspect the code to trace any exposure rating back to its inputs.
Signal contextualisation uses parameterised templates
When signals are contextualised for a postcode, the text is generated from parameterised templates with known inputs. Reproducible, not generative.
Structural characteristics match against national percentiles
Every structural characteristic is positioned against the national distribution. You can verify your community's position against the ABS source data.
What Cannot Be Validated Yet
Honest gapLimitations we name openly rather than hide.
Cascade timeline estimates are structural, not empirical
Timeline estimates (e.g. 'fuel price increases reach grocery shelves in 2-4 weeks') are based on industry knowledge and published cost coefficients, not empirical measurement of actual propagation delays. We present them as estimates, not forecasts.
Exposure mapping rules are based on domain expertise, not statistical derivation
The rules that map structural characteristics to exposure domains are authored from domain expertise, not derived from statistical models. This is a deliberate design choice: transparent rules that anyone can inspect and challenge are more trustworthy than an opaque model with higher accuracy claims.
Station availability gap detection is a proxy
When fuel station reporting gaps are used as a demand pressure signal, it is a proxy measure. Stations may stop reporting for operational reasons unrelated to supply pressure. We flag this uncertainty in the signal context.
Data Transparency
We show what data is available and name what is missing. Every data point is sourced and dated.
What We Disclose
TransparentData provenance, age, and coverage gaps are surfaced, not hidden.
Most structural data is from 2021 Census (5 years old)
We say so on every profile. The 2026 Census will allow us to refresh structural indicators. Until then, the data reflects 2021 conditions and we are explicit about that.
Signal availability varies by state
WA has the most transparent fuel data (FuelWatch, station-level, daily). NSW provides data via CKAN and FuelCheck. Other states have limited or no automated public data. We surface what is available and name what is missing.
Every data point is sourced and dated
Structural characteristics show their data source and vintage. Live signals show their fetch timestamp and source authority. Derived signals like cascade pressure are labelled as estimates.
Missing data is named, not hidden
When an indicator has no data for a postcode, we show that gap rather than imputing a value or hiding the field. When a signal source is unavailable for a region, we say so.
OECD 10-Step Compliance
The scoring engine follows the OECD/JRC Handbook on Constructing Composite Indicators (2008) and the England BRIC adaptation by Camacho et al. (2024), which achieved 100% OECD compliance and is the closest methodological precedent. We preserve this rigour even though the primary output is now an exposure profile rather than a composite score.
| Step | Requirement | Status |
|---|---|---|
| 1 | Theoretical framework | CompleteDROP model + INFORM + coherence/entrainment lens |
| 2 | Data selection | Complete11 structural indicators with documented rationale |
| 3 | Imputation of missing data | CompleteMissing data handling with explicit gap surfacing |
| 4 | Multivariate analysis | CompletePCA + Cronbach alpha per capital |
| 5 | Normalisation | CompleteMin-max + percentile-rank, robustness validated |
| 6 | Weighting and aggregation | CompleteEqual weighting within capitals, sensitivity tested |
| 7 | Robustness and sensitivity | CompleteNormalisation robustness + weight perturbation analysis |
| 8 | Back to the data | CompleteFull decomposition in exposure profiles |
| 9 | Links to other indicators | CompleteSEIFA IRSD external validation (r=0.909) |
| 10 | Visualisation | CompleteExposure profiles, signal cascade, action templates |
Validation run: 20 March 2026. Source data: ABS Census 2021, SEIFA 2021, MMM 2023, CER postcode data. Signal sources: WA FuelWatch, AIP terminal gate prices, AEMO, RBA, ABS SDMX, DCCEEW, Yahoo Finance.