Limitations
What SENTINEL doesn't know.
Every product has limitations. Most don't list them. We do. This page is updated as we find new gaps; if you spot one, tell us.
Coverage gaps
- Tasmania, NT, ACT, SA roads — no public live road incident API. We can't show road closures or hazards in these jurisdictions until those agencies publish an open feed (or we arrange data access).
- VIC and WA roads — APIs exist but require approved access tokens. In progress.
- NT and remote regional areas — OpenStreetMap amenity tagging is patchy. Schools, hospitals and emergency facilities are well-tagged in metro and major regional areas, less so in remote communities.
- Indigenous community hazard data — limited public datasets. We surface what's available; gaps reflect publication, not relevance.
Geocoder limits
- Coordinates from Nominatim are accurate to building level in most metro areas, street level in most regional areas, and sometimes only suburb level in remote areas. Every result shows its confidence rating.
- New subdivisions and very recent address changes may not be in the dataset.
- For rural properties without a numbered street address, results may centre on the nearest road or property entrance.
Historical event records
- Coverage starts at 2010. Earlier major events (1983 Ash Wednesday, 1974 Brisbane floods, etc.) will be added in future revisions as we verify primary sources.
- Extent polygons are sourced from the responsible agency. Some agencies publish polygons with more precision than others — we surface the polygon we have.
- Statistics (deaths, properties destroyed) come from official inquiries, royal commissions, or post-event reviews. Numbers in the immediate aftermath of an event are often revised later; we use the most recent authoritative figure and cite the source.
- We catalogue events of state or national significance. Local events with significant local impact may not yet be represented.
Modelled risk overlays
- Flood extent maps reflect modelling at a specific point in time with specific assumptions. A 1% AEP flood extent from a 2018 study is not equivalent to a 1% AEP flood from a 2024 study — terrain, hydrology and infrastructure all change.
- Bushfire-prone area mapping is a statutory designation, not a prediction. A property outside the mapped area is not "safe"; a property inside is not necessarily at imminent risk.
- Modelled overlays do not account for individual property characteristics (building materials, asset protection zones, defendable space).
- Climate change projections are not currently reflected in the historical-baseline modelling we surface. Future scenarios will be added as a separate layer with clear labelling.
Comparison engine limits
- Spatial overlap (Jaccard index) measures geometric similarity, not impact similarity. Two fires with the same Jaccard might have wildly different consequence depending on what's in the polygon.
- Quantitative comparison only includes statistics we have for both events. When data is missing for one, that metric is omitted.
- Comparison shows similarity to past events. It does not predict what will happen — agency models do that, and we link to them rather than predict ourselves.
Freshness
- Live feeds are polled on a schedule (60 sec to 15 min depending on source). Between polls, displayed data is the last snapshot we have. Each panel shows when it was last fetched.
- During severe events, the source agency feed may itself be delayed or experience outages. We display the most recent snapshot and timestamp it; if you see a stale timestamp, check the agency directly.
Things we don't do (deliberately)
- We don't generate our own warnings or alerts
- We don't recommend specific actions in specific situations
- We don't quantify property risk in dollar terms
- We don't forecast where current incidents will spread
- We don't replace agency advice
- We don't use AI to summarise live agency text without human review
Telling us
If you find a limitation we should add to this list, or an error in our data: hello@sentinel.example. We log every report and respond.