SafeFuture Lab
Work relied on
This page sets out what the lab's work is, where government or industry use of it is on the public record, and what we have not yet evidenced. Every item has a link, so you can check it yourself.
1. The work, and where it is publicly described
| Work | What it is | Public record |
|---|---|---|
| Statewide signal evidence (Victoria) | About 4,870 published signal-configuration sheets, many of them scanned with no text layer, read with AI vision and checked against road geometry. This gives per-approach evidence at 4,527 of 4,588 volumed signalised intersections (98.7%). The 61 remaining sites are each listed with a reason. | Case note: statewide signal evidence |
| Directional traffic volume for every Victorian road | 1,267,052 road sections in three evidence tiers (measured, intersection, modelled). About 71 billion vehicle-kilometres are closed to a documented envelope. Where official sources disagree, the disagreement table is published. | Case note: directional traffic volume |
| Before/after evaluation platform | Runs naive, comparison-group and Empirical Bayes methods side by side. Each method returns an estimate with its interval, or a stated reason it cannot be estimated. Worked example: crash modification factor 0.75 (95% CI 0.30–1.85). That is consistent with a reduction, but it is not evidence of one. | Evaluation · Worked example |
| Speed-zone evaluation and aggregation bias (NSW) | Six NSW speed-zone changes evaluated with weighted difference-in-differences against control roads. A connected-vehicle method separates path-level speeds from aggregate averages. The note presents the method; site-level effect sizes are not presented as findings. | Case note: speed zones |
| Skewed intersections (Victoria) | 34,220 Victorian intersections classified by approach angle and linked to crash history and heavy-vehicle networks. The code is open source. | Case note: skewed intersections |
| Camera-based vehicle counting | Computer-vision vehicle counts from 79 public traffic cameras in Victoria and NSW, about 735,000 rows a day. The night-time limitation is documented. | Network evidence |
| Spatial road-risk methods | Vertical grade, horizontal curvature, Movement & Place inference and network-wide infrastructure risk rating, plus a library of 99 QGIS processing models (2022–2026). | Spatial risk |
| Crash data systems and national fatality analysis | Design, upgrade and maintenance of crash analysis and reporting systems. Also national road-fatality and year-to-date road-toll reporting, and trend explanations tested against gates stated before the data arrive. | Crash systems |
| Verified AI and machine learning | Public AI assistants whose answers are machine-checked against sources and that refuse when the check fails. Also crash-likelihood modelling on open data, a blind benchmark of language models on official crash-type coding, and offline de-identification of crash narratives. | Verified AI |
2. Where government or industry has used it, on the public record
None listed yet. The lab's public pages describe the work above but do not name any government agency or company that has relied on it. We also searched public publication indexes when compiling this page and found no indexed publication record for the lab to link to.
We will list a use here only once there is a public document to link to, such as a published report, a journal article or a public statement by the user. Until then, each possible use sits in section 3.
3. Evidence we are still gathering
These are uses the work was built for, or that its descriptions point to, but that we cannot yet show on the public record. They are not claims. Each link goes to the public description of the work concerned.
| Work | What we still need to evidence publicly | Link |
|---|---|---|
| Crash analysis and reporting systems | The lab designs, upgrades and maintains crash data platforms. A public record of which organisations run them is not yet available to link. | Crash systems |
| National fatality and road-toll reporting | A public citation showing who uses or relies on this reporting. | Crash systems |
| Speed-zone evaluation (six NSW sites) | A public record that a road authority used these evaluations in a speed-limit decision or review. | Case note: speed zones |
| Before/after evaluation platform | A public record of a treatment or funding decision that relied on a platform result. | Evaluation |
| Statewide signal evidence and directional traffic volume | A public record of these layers being used in network planning, safety programming or exposure-based crash rates by an agency or consultant. | Signal evidence · Traffic volume |
| Road assessment data pipelines | A public record of the extracted assessment datasets being used outside the lab. | Network evidence |
| Skewed intersections (open source) | A public record of reuse of the open data or code, such as a citation, a fork or a report. | Case note: skewed intersections |
| Driver telematics study (200 drivers) | The study has UNSW ethics approval (with conditions). There are no results yet, so there is nothing that could have been relied on. | Evaluation |
| Peer-reviewed publications | Links to journal and conference records once they are listed in public publication indexes. | Team Lead's public profile |
How this page is kept honest
- Items move from section 3 to section 2 only when there is a public document to link to.
- We do not name organisations whose use is not on the public record.
- Figures are aggregates from the lab's own pipelines on open and public data. Site-level data and confidential records are not published here.