[dgoodenough]

Transfers analysis and infill stations

• Followed every modelled trip through the network: where people change, which trips need two or more transfers, and 148 places where two lines cross or pass close with no station (reports/TRANSFERS.md).
• Tested ~40 candidate infill stops, walk links and short extensions one at a time in the ridership model, then together with each one left out in turn. Kept the 13 that still added riders next to the others:
  • Infill stations (plan and 2M): Red/Orange at Main / Sherman Street (downtown Richardson) and NW Highway / Shady Brook; Red/Blue/Orange at Central / Hall; Blue at Shiloh / Forest; Orange at John W Carpenter / Freeport; TRE at Rock Island / Briery and Airport / Beach (with FW Green); TEXRail extension and FW Berry at Pennsylvania / Henderson; Silver Line at Marsh / Arapaho; Midlothian at I-20 / Oriole.
  • One-seat links (2M): SH-114 trains run on over the Green Line to Victory (new run_over edit); the Stemmons-Brookhollow BRT starts at Victory.
  • Walk link (2M): Commerce / Harwood – Marilla / Pearl.
• Result: more riders, but not fewer transfers. Holding the commutes fixed, transfers per commute stay about level (plan 0.33 → 0.34, 2M 0.47). With the model's 3-minute transfer penalty, a new interchange often wins riders by making a trip faster with one change. A longer transfer penalty is noted for a later round.

SH-114 Line (owner's hand-drawn realignment)

• Runs in the LBJ/SH-114 freeway corridor without the loops at Las Colinas, Freeport North and Southlake.
• Runs beside TEXRail through Grapevine.
• New Trophy Club and Westlake stations on SH-114, replacing the Kirkwood / Dove Road / Westlake Solana detour.
• Belt Line / Royal moves onto the SH-114 crossing (renamed Belt Line / SH-114).
• gen_expansion: new road_weights option, and per-leg "follow": "rail" and "follow": "straight".

Diagram

• Rail / BRT / All versions, each laid out for its own lines; Rail is the default.
• New straightness rules: long runs laid on their axis, a cost per bend and more for zig-zags, straight-through lines kept opposite at junctions, and no run laid across another line.
• The Rail diagram has about half the bends of the old all-modes diagram, with crossings equal to real geography (79). The All version is still bendier on the new network (known limitation).

Results (full build)

	Plan	2M
Daily trips 2050	388k (was 370k)	1.06M (was 1.03M)
Daily trips 2100	788k (was 771k)	2.25M (was 2.23M); 2M reached 2077
Capital through 2100	$103.2B	$136.6B (rescheduled: DNT East and I-20 earlier, I-30 and Las Colinas–Mesquite later)

Both plans stay within budget and no stage opens after 2100.

Testing

• 70 unit tests pass, including new tests for run_over and the diagram layout rules.
• python -m dfwt validate: no errors. The two errors in the archived v1/v2 scenarios are unchanged from main.
• Full build, then a quick build for the diagram. Headless Chromium check of all three diagram versions (no page errors) and of two-station trip clicks.

---

Generated by Claude Code