2026-06-23 Australia Team

[joecastiglione]

Agenda

• ActivitySim Application & Analysis Guide

  • Review of Draft Land Use Guide

• Autocalibration

  • Off Model (FineTune) vs Integrated (RSG)
  • Interactive dashboard vs Static output tables and static PNGs
  • Oscillation Detection
  • Convergence and Tolerances
  • Complementary or Competing?

• EET Implementation & Results Investigation

  • Preliminary results
  • Discussed at 2026.06.18 Engineering meeting

• RNG Findings

  • No significant changes made

• Maintentance

  • Small Example Model Development
  • Testing methods and LOE

• Explicit Telecommute Implementation

  • Update next week

• Telecommute Frequency Model Update and Tests

  • Update in 2 weeks

Notes

Technical Updates

ActivitySim Application & Analysis Guide: Land Use Change Example

Joe F shared a first draft of the land use change application guide (one of three planned examples: land use change, telecommuting change, and network change). Amir (TfNSW) and Sumit (TransLink) were acknowledged for providing useful early feedback.

Next steps:

• Joe F will reformat the document to support inline commenting via a GitHub pull request, following a preference expressed on an earlier call.
• A revised, commentable version will be recirculated. Target for feedback: 1–2 weeks after that version is shared.
• SANDAG staff will conduct a practical validation review, working through the guide as a first-time modeler to identify gaps.

EET Implementation & Results Investigation

Jan reported that Sijia ran EET scenarios from Phase 10C, including an employment scenario and a transit improvement scenario for a defined SANDAG corridor. Key findings and issues discussed:

• A bug was identified in the workplace location simulation related to income/occupation segmentation in size terms — a YAML parameter was not specified, causing the model to default to income segmentation even when occupation-based segmentation was intended. A pull request has been opened.
• The current EET run uses a single iteration of workplace location without shadow pricing due to a separate segmentation issue discovered during the run.
• Joel noted a known limitation of the simulation method: when workers are re-simulated due to zone capacity constraints, utility loss is not considered, which can result in some workers being assigned to locations far worse than alternatives. A proposed solution exists that would also speed up the algorithm; Joel suggested scoping this for Phase 12 funding.
• Paris Brunton noted that Victoria DTP still uses shadow pricing and asked about the broader consortium view. No clear majority preference exists; both methods are in use across agencies.

3. Autocalibration & Governance

Pedro raised the importance of establishing clear governance for the auto-calibration tool before it becomes widely depended upon. His core concern: if the tool is created by a single entity but becomes central to the consortium's work, that entity's future decisions could constrain the direction of the tool for all users.

Discussion:

• Amir (TfNSW) agreed guardrails are needed but emphasized they should not discourage contributions.
• Joe C noted the ActivitySim governance document addresses some of these questions in principle but has not been strictly enforced.
• Joe C noted that ActivitySim has many dependencies over which it has no decisionmaking control.
• Action: Pedro will open a GitHub issue in the ActivitySim project to document the governance concern and keep it visible for future discussion.

Testing Framework

Joe flagged an ongoing conversation about the level of investment in automated testing. Jeff's earlier document on testing frameworks (from ~6 months ago) continues to surface in discussions. As Phase 12 planning begins, the question of whether to explicitly scope and fund a testing workstream will need to be addressed. Jeff suggested that 50% or more of the budget could be spent writing tests.

Explicit Telecommute Implementation

WSP is leading the explicit telecommute implementation, building on design work from Phase 9. An update is scheduled for next week's meeting (at an Australia-compatible time). Joe will request pre-read materials from Sijia ahead of that session.

Telecommute Frequency Model Update

RSG is leading a smaller task on the telecommute frequency model update and testing. This will be presented in two weeks (not Australia-compatible timing).

Action Items

#	Action	Owner	Timing
1	Recirculate reformatted Application Guide as GitHub PR for comment	Joe Flood	Soon
2	Provide feedback on Application Guide draft	All consortium members	1–2 weeks after new version shared
3	Open GitHub issue on auto-calibration governance concerns	Outer Loop	Near-term
4	Comment on workplace location simulation bug issue (Sijia's PR)	RSG	Near-term
5	Share pre-read materials for next week's telecommute update	WSP	Before next meeting
6	Complete member survey on RFQ priorities / ActivitySim pain points	All agency staff	Within days
7	Follow up with Amir on performance benchmarking table (threads/chunking/sample rate)	Joe C	Separate follow-up