Mission dossier
Folsom Street signal timing (San Francisco)
Folsom St: 2nd St → Hawthorne St, San Francisco · Data as of 2026-10-19
Exploratory research mission. It has not been validated, reviewed or adopted, and nothing here is a recommendation.
This mission at a glance
- Where you are
- A public mission dossier. You can read it without an account or the 3D Earth.
- What you can do
- Read the problem, its sources and what is unknown
- Open the workspace and test the scenarios yourself
- Add what you have seen, kept in your browser
- Who takes part
- You, in your own browser. Shared participation with other people is not open yet.
- What the agents really do
- 5 software identities, each with one permitted tool per task. They run deterministic tools in your browser: no language model, no real-world action.
- What is verified
- 1 public source(s), each with its licence and retrieval date, plus 1 CORE record(s), not independent evidence
- Evidence checks run by code
- What remains
- 3 declared gaps in the evidence
- Scientific model validation: Not done yet
- Independent expert review: Not available yet
- Institutional decision: Outside CORE
- Measured real-world outcome: None so far
- What a human has validated
- Nothing. No reviewer or institution has validated this mission.
The problem
Cars and buses queue along Folsom Street in South of Market. Could giving one mapped approach more green time reduce delay, and what evidence would anyone need before acting on that?
The exact research question
Would a ten-point green-time increase on the mapped Folsom Street approach reduce modelled delay, how robust is that to unknown demand, and what evidence is missing before any real-world claim?
How it was found
The CORE team defined this question from dated public street data, and a recorded AI investigation examined it. No resident report started it and no traffic was measured.
Why it matters
Congestion costs people time, slows buses that many low-income residents rely on, and adds emissions, but a wrong intervention can make things worse.
Where
Folsom St: 2nd St → Hawthorne St, San Francisco
Evidence and what is unknown
Sources used by this mission
- Streets – Active and Retired (3psu-pn9h), in the urban model catalog
City and County of San Francisco (DPW Basemap) · ODC Public Domain Dedication and License (PDDL) · retrieved 2026-10-13 · source
- Recorded local-model investigation of this missionCORE’s own record, not independent evidence
CORE · CORE published record (this repository) · retrieved 2026-10-19
What the sources show
- Observed in a public source
69 of the model’s 73 signals match a signal recorded by DataSF (within 35 m); 74 signals recorded in the area.
Traffic Signals (DataSF) · retrieved 2026-10-16 · source
Presence of signals only; their timing is not published.
What is still unknown
- Traffic counts: no public dataset found: demand is synthetic.
- Real signal timing: not published: signal plans are generated by SUMO.
- Muni schedules (GTFS): access requires accepting a licence (indemnity, California law), a founder decision; the bus line is an assumption.
- Uncalibrated fluid queue at one approach: no turns, spillback, transit or pedestrians.
- Demand and signal timing are assumptions: no authorized counts or signal plans are bundled.
- Results are exploratory computations, not predictions or measured effects.
Agents and their limits
Agents are software identities with one permitted tool per task. On this public site they run deterministic tools in your browser: no language model, no network, no real-world action. They only propose; people decide.
- CORE-DAT-003 Geodata analyst
- Confirm the mapped approach and that its source is authorized for modeling.
geo.segmentread-bundled
- Confirm the mapped approach and that its source is authorized for modeling.
- CORE-RES-0043 Evidence verifier
- Read what the recorded investigation established, claimed and left open.
records.inspectread-records - List the scientific requirements still unmet for this approach.
evidence.readinessread-bundled
- Read what the recorded investigation established, claimed and left open.
- CORE-TRF-0010 Direction: Transport and mobility
- Run the exploratory simulator on the declared assumptions.
sim.comparecompute-local - Test whether the modeled effect survives the unknown demand.
sim.sweepcompute-local
- Run the exploratory simulator on the declared assumptions.
- CORE-RES-0029 Direction: Research and evidence verification
- Compare computed outputs with recorded claims, assumptions and readiness.
evidence.comparecompare
- Compare computed outputs with recorded claims, assumptions and readiness.
- CORE-CNL-0001 Deliberation: question, evidence, hypotheses, dissent, verification requests, review recommendation
- Draft testable hypotheses (local model only; otherwise unavailable).
reason.hypotheseslocal-model - Compose the report and hand it to human review.
report.composereport-local
- Draft testable hypotheses (local model only; otherwise unavailable).
Recorded investigation
A local language model investigated this question on 2026-10-19: 7 tasks, 8 findings. It is a recording, not live work, and human review is still pending. It contains errors that remain visible, including placing this street in Oakland.
Scenarios and their limits
- Recorded simulation · synthetic demand, uncalibrated
Modeled baseline (real OSM network, synthetic demand): mean car delay 182 s (± 4 s), bus travel time 358 s on the synthetic Folsom line.
SUMO 1.27.1 simulations (5 seeds, 30 simulated minutes) · retrieved 2026-10-16 · source
UNCALIBRATED, no result describes real traffic in San Francisco.
- Recorded simulation · synthetic demand, uncalibrated
Scenario A · Bus-only lane on Folsom: car delay +76.8%, bus time +175.8%, estimated CO2 -0.4% versus the baseline.
SUMO 1.27.1 simulations (5 seeds, 30 simulated minutes) · retrieved 2026-10-16 · source
Folsom has 6 segments with fewer than 3 lanes where the bus lane breaks off: cars lose a lane there, queues spill back up the corridor, and buses get stuck in them. UNCALIBRATED, no result describes real traffic in San Francisco.
- Recorded simulation · synthetic demand, uncalibrated
Scenario B · Adaptive (actuated) signals: car delay -46.1%, bus time -14%, estimated CO2 -17.9% versus the baseline.
SUMO 1.27.1 simulations (5 seeds, 30 simulated minutes) · retrieved 2026-10-16 · source
Actuated signals cut dead time at lightly loaded intersections; the gain depends on real signal plans we do not know. UNCALIBRATED, no result describes real traffic in San Francisco.
- Recorded simulation · synthetic demand, uncalibrated
Scenario C · Protected bike lane (road diet): car delay +83.1%, bus time +200.4%, estimated CO2 +1.6% versus the baseline.
SUMO 1.27.1 simulations (5 seeds, 30 simulated minutes) · retrieved 2026-10-16 · source
One car lane fewer with unchanged demand: capacity drops and queues grow (the model does not simulate the additional cyclists). UNCALIBRATED, no result describes real traffic in San Francisco.
- Recorded simulation · synthetic demand, uncalibrated
Scenario D · Demand management (−15% car trips): car delay -11.9%, bus time -0.8%, estimated CO2 -18.4% versus the baseline.
SUMO 1.27.1 simulations (5 seeds, 30 simulated minutes) · retrieved 2026-10-16 · source
Fewer car trips: a mechanical effect of a demand assumption, not of a measure whose feasibility is known. UNCALIBRATED, no result describes real traffic in San Francisco.
- Recorded simulation · synthetic demand, uncalibrated
Scenario search: 15 configurations generated, 1 rejected by a constraint, 14 simulated (3 seeds); Pareto front: act at 90% demand, act at 85% demand.
SUMO 1.27.1 simulations (5 seeds, 30 simulated minutes) · retrieved 2026-10-16 · source
Grid of 15 configurations; constraint filtering; 3 seeds each; no surrogate model used.
Hypotheses, not findings
- Actuated signals
In this model, actuated signals reduce delay for cars and buses; this must be checked against real signal plans.
- Continuous bus lane
A bus lane only helps buses if it is continuous: in this model, its gaps cause queues to spill back.
- Demand reduction
Fewer car trips reduce delay; this remains a demand assumption until a concrete measure is modelled.
Recorded objections
- An uncalibrated model cannot justify recommending an intervention: rankings may flip under real demand.
- The road diet worsens car time in this model, but the model simulates neither cyclists nor safety: do not judge it on car delay alone.
Collaborate
Invited members work on this mission together in an invite-only shared space: observations, reviews with written reasons, a verified history. Nothing from it is published here.
Open the shared mission (invite-only)
Without an invitation, what you add below stays in your browser, and nobody receives it.
What would validate it
- Evidence checksWorking
- Exploratory simulationsWorking, uncalibrated
- Scientific model validationNot done yet
- Independent expert reviewNot available yet
- Institutional decisionOutside CORE
- Measured real-world outcomeNone so far
Decisions and results
- Human review
- Not done
- Institutional decision
- None
- Measured real-world result
- None
Traffic signals in San Francisco are the responsibility of the San Francisco Municipal Transportation Agency (SFMTA). CORE has no agreement with it and sends it nothing.