Methodology · Last reviewed 2026-04-30
Show the working.
propautopilot is an editorial research desk for Australian property. Seven specialists work from primary government data sources, every claim is cited to its primary source, and every dated property recommendation is graded held / missed / pending after twelve months in a public ledger. This page documents how. It is updated quarterly.
The seven specialists.
Each specialist consults the same primary data foundation but applies a different decision lens. Suburb-level conclusions are cross-checked across multiple specialists before a verdict is published.
- CoachDiscover
- Buyer profiling, brief shaping, capacity capture
- Suburb ScoutResearch
- Area-level reads, verdict calls (Buy / Watch / Avoid)
- HunterResearch
- Listing-level pull and shortlist
- ValuerEvaluate
- Fair-value triangulation across comparable ladder + statutory inputs
- NegotiatorOffer
- Offer position, draft email, walk-away discipline
- StewardHold
- Long-hold portfolio, tax position, refinance windows
- AdvocateSettle
- Buyer's-side process, settlement, escalation
Primary sources we read.
Every claim on a public suburb report or in a Compass answer is cited to one of these. The full source register lives in the methodology paper PDF (available to Explorer-tier accounts).
| Source | What we use it for | Cadence |
|---|---|---|
| ATO postcode tax statistics | Income distribution, rental income, CGT realised at exit | Annual (financial year) |
| ABS Census 2021 | Population mix, household types, dwelling stock, demographic flow | 5-yearly |
| NSW Fair Trading tenancy bond lodgements | Actual paid rents (not asking rents) | Quarterly publication |
| State land-record transactions | Settled property prices, the foundation of yield + growth math | Weekly to monthly per state |
| Public planning records | Supply pipeline + planned dwellings, an early signal of dilution | Continuous |
| ASIC company registers | Corporate holdings, trustees, beneficial-ownership traces | Continuous |
| News RSS feeds (council + market commentary) | Suburb-specific commentary tagged + cited | Daily |
| 200+ further sources | Hazards, transit, schools, demographics, infrastructure pipelines, crime, environmental risk | Per-source |
Citation chips.
Every cited claim renders an inline rust-accent chip. Tap the chip to open the citation drawer. The drawer shows the primary source name (statute-OK references only, no commercial vendor names), the field path, the last-fetched date, and any cross-source agreement metadata. The chip prefix is §, the trailing ↗ indicates an outbound primary-source link is available where the source publishes one.
Five confidence states.
Every metric on a suburb report carries a confidence state. The state determines whether a number is rendered, suppressed, or shown with explicit hedging. Never softened or fabricated.
- Primary
- Direct read from a primary source. Highest confidence.
- UI: Solid chip, full cite drawer
- Derived
- Computed from primary inputs (e.g. yield = annual rent ÷ purchase price). Confidence inherits the weakest input.
- UI: Solid chip, derivation expanded in drawer
- Provisional
- Pending. Input data exists but hasn't been recomputed yet (data lag, refresh window, model recompute). Renders 'Pending' rather than a stale figure.
- UI: Dashed chip, eta in drawer
- Conflict
- Two or more primary sources disagree by more than 5%. Renders the conservative read with the divergence flagged.
- UI: Amber chip, both readings shown in drawer
- Stale
- Primary source hasn't published in its expected window (e.g. quarterly source 5 months silent). Last-known value rendered with last-reviewed date prominent.
- UI: Greyed chip, age + 'last reviewed' in drawer
The public prediction ledger.
Every dated property recommendation we publish is graded twelve months later as held, missed, or pending. The ledger is public. See /ledger.
Held: outcome lands within ±5% of the recommendation's primary metric (e.g. predicted growth of 6% landing between 5.7% and 6.3%).
Missed: outcome lands more than 5% in the wrong direction. Misses are shown alongside the original reasoning so the failure mode is auditable.
Pending: the twelve-month outcome window has not elapsed yet, or the primary source for the outcome metric has not published the relevant period. Pending entries are visible from day one, so buyers can audit our reasoning before any verdict has resolved.
We don't backfill. We don't retroactively adjust verdicts. We don't censor misses. The ledger compounds. By year three, ~70K graded entries make the system's hit rate impossible to fake.
Quarterly refresh cadence.
Tax brackets, HGS caps, LMI schedules, state Duties Act rates, and source register reviews land every quarter. When a rate or rule changes, the update lands within two weeks and a last-reviewed date is surfaced on the affected calculator + the affected suburb fields.
This document is reviewed at the same cadence. Last reviewed: 2026-04-30.
What we never do.
- Round to look agreeable. If the median is $1,182,500, we render $1,182,500, not "around $1.18M". Buyers see the exact figure, not a softened one.
- Pay-to-play. No suburb appears higher because someone paid. No vendor name appears in any verdict prose.
- Fear-first framing. Calculators state current rules and model what changes if they do. They don't amplify panic about negative gearing reform or rate moves.
- Backfill verdicts. Once a recommendation is published, the date stamp is permanent. We grade it twelve months later (held, missed, or pending) and we publish either way.
- Hide misses. Missed verdicts stay in the public ledger with the original reasoning. The system gets better by being audited, not by being polished.
Read further.
- The Buyer's Playbook This methodology applied to the end-to-end buying journey, narrated by the seven specialists.
- Public prediction ledger Every dated verdict, graded after twelve months.
- About propautopilot What it is and what it isn't.
- How to find investment properties: analyst's framework 10-step worked example using this methodology.
- hello@propautopilot.ai Questions about the methodology, or report a citation that doesn't resolve.