2025-10-30 Engineering Team

[joecastiglione]

@ActivitySim/engineering

Agenda

• Updates on 11B tasks

  • Consolidated Error Definitions (CS) -- ActivitySim#1000. Awaiting @jpn-- review
  • Trace files consistent hash (CS)
  • UV Support (WSP + All) -- ActivitySim#1007 Engineering team should review.
  • Clean up loggging (CS) -- ActivitySim#1004 ✅ Merged
  • Consistent Alternatives File Formatting (RSG)
  • Trip Scheduling Choice Reproducible Random (RSG) -- ActivitySim#1005 Awaiting @jpn-- review
  • Fix Shadow Price Zones Reopening (RSG)
  • Global Option to Skip Households on Fail (WSP)
  • Example SANDAG Model Update (WSP) -- sandag-abm3-example#37 approved by SANDAG, needs review from @jpn--
  • SANDAG Sharrow Performance Optimization (DL) -- Found bug: SANDAG parking location model uses DISTWALK skims, which exist only as MAZ-to-MAZ (no TAZ level skims are included). Legacy ActivitySim assumes missing TAZ values are all zero, Sharrow just fails (it shouldn't).

• Park and Ride, continue technical discussion

[dhensle]

Suggested agenda item: GitHub permissions -- I can no longer close GitHub issues on the ActivitySim/ActivitySim repo.

[joecastiglione]

@dhensle - You should have permission now.

[jpn--]

ActivitySim Engineering Team Meeting Notes

Meeting Overview

Agenda

• Phase 11B task updates and status reviews
• Discussion of Phase 11C scoping process and challenges
• Technical presentation and discussion on park and ride implementation
• Planning for upcoming meetings

Participants

• Joe Castiglione (@joecastiglione), Zephyr Foundation
• Jeff Newman (@jpn--), Driftless
• David Hensle (@dhensle), RSG
• Sijia Wang (@i-am-sijia), WSP
• Stefan Coe (@stefancoe), PSRC
• Bo Wen (@bwentl), TransLink

Duration

65 minutes

Executive Summary

The engineering team reviewed progress on Phase 11B tasks, with significant attention to resolving SANDAG model compatibility with Sharrow and addressing missing walk distance skims. A substantial discussion focused on challenges with the Phase 11C scoping process, particularly consortium members reporting insufficient information to prioritize tasks effectively. The team also continued technical discussions on park and ride implementation being developed for TransLink and Oregon DOT, which uses a simulation-based capacity constraint mechanism rather than traditional shadow pricing.

Meeting Notes

Phase 11B Task Updates

Jeff Newman reported progress on integrating the SANDAG model with Sharrow. The primary issue identified was that SANDAG's walk distance data only exists in transit skims, not in TAZ-to-TAZ skims. Legacy ActivitySim handled this by defaulting to zeros when data was missing, but Sharrow's compiled code was falling back to null values instead. This has been resolved, though there's a question about whether SANDAG should add actual TAZ-to-TAZ walk distance data to avoid potential bugs where parking locations might be assigned at infeasible distances.

Other task updates included consolidated error definitions completed by Joe Flood and ready for review, trace files requiring consistent hashing assigned to CS, and UV documentation in pull request ready for broad engineering team review. The example SANDAG model work is pending completion of the Sharrow integration. Sijia Wang added an auto-diffing script to help keep the example model synchronized with SANDAG's production model, though decisions about automation frequency will be made later.

Phase 11C Scoping Process Discussion

Joe Castiglione raised concerns about the Phase 11C prioritization process, noting that consortium members consistently report insufficient information to make informed prioritization decisions. This feedback has persisted despite previous efforts to add more detail to task descriptions. The process has been ongoing for over two months without converging on final task orders.

The discussion explored different historical approaches to scoping, from detailed consultant proposals in response to specific requests to more collaborative ad hoc discussions. David Hensle suggested a middle-ground approach of providing descriptions with level-of-effort estimates for multiple options, which he demonstrated with recent ET work. However, questions arose about whether consultants should be compensated for this scoping work and whether it's appropriate to have one firm develop scopes that all firms would potentially bid on.

Sijia Wang emphasized the value of collaborative feature development from the start to avoid locally optimized solutions that require significant rework later. She suggested having all consultants provide effort estimates for open tasks to help agencies understand resource requirements.

Jeff Newman noted that the current process differs from historical IRFPs where the consortium would narrow priorities to a small number of items before requesting detailed proposals. He suggested focusing on high-level strategic priorities (performance, usability, research) rather than having all agencies evaluate every possible task. The discussion concluded without immediate process changes but with acknowledgment that the current approach needs refinement.

Park and Ride Implementation

David Hensle presented ongoing work on park and ride modeling being funded by TransLink and Oregon DOT. Bo Wen explained TransLink's motivation: their current park and ride model exists outside ActivitySim, is over five years old, difficult to maintain, and creates challenges with data management between assignment and demand models. The new approach would handle skim manipulation within ActivitySim and better integrate with mode choice and other decisions.

The implementation uses simulation-based capacity constraints rather than traditional shadow pricing, similar to the workplace location approach. Tours are sorted by departure time, and lot assignments are made sequentially with capacity tracking. When lots reach capacity, subsequent tours are re-simulated to choose alternative lots or modes. The utility calculation incorporates driving costs from origin to lot and transit costs from lot to destination in both directions.

Discussion covered whether capacity constraints should include utility penalties for near-full lots (similar to highway volume-delay functions) and how time-dependent availability affects user behavior. The team agreed to continue the technical discussion in future meetings, with the goal of eventually integrating this as a configurable option in ActivitySim rather than requiring agencies to maintain forks.

Action Items

• Jeff Newman: Review consolidated error definitions pull request from Joe Flood
• Jeff Newman: Review trip scheduling choice pull request marked ready for review
• Jeff Newman: Complete and test Sharrow changes to handle missing skim data
• Engineering Team (All): Review UV documentation pull request and provide feedback
• Joe Castiglione: Share David Hensle's ET scoping document with product and community team for feedback
• Joe Castiglione: Gather additional feedback from consortium members on scoping process
• David Hensle: Continue park and ride presentation in next engineering team meeting
• Engineering Team: Schedule meeting discussion for November (week before Thanksgiving) without Jeff Newman to continue park and ride discussions