U
U-Flowby Utopic Traffic

Your ATSPM controllers log every detector event. Your retiming plan is still 3 months stale.

U-Flow reads the data your cabinet already produces — no new sensors, no integration project — and turns it into a queue of findings your engineers can act on this week.

No new hardware·Engineer-in-the-loop approval·Audit trail by default
Signal operations queue
SR99 @ 320th · Analysis snapshot · 28 days4 work items
Northbound green-band performance jumped 3.8× mid-monthBlockedHIGH
71% of ped button presses come from one crosswalk; recall uniformReadyHIGH
Controller has 26 timing patterns; only 1 has run in 28 daysNewMED
Arrival pattern on P6 may have shifted — distribution checkShadowingLOW
Engineer-in-the-loop · Hypothesis ledger per finding · Audit trail saved
The retiming problem

Your engineers spend three months on a retiming plan. The city deploys it six weeks later. By then, the intersections have already shifted.

A typical mid-size city spends six figures a year on manual tube counts and consultant retiming — for a plan that’s already out of date when it ships. Meanwhile, your ATSPM controllers are generating millions of detector events that nobody is using to decide tomorrow’s timing.

Per consultant retiming contract · industry typical
$100–200K/ engagement

Recurring line item in DOT operations budgets. ATSPM cuts the data-collection portion; the manual-engineering portion still recurs.

Between retimings · funding-driven
>1 yeartypical gap

Frequency depends on agency budget cycles, not a fixed schedule. Mid-cycle congestion waits until the next contract.

Tube counter accuracy · industry-typical
~10%miss rate

Plans sized to data that misses roughly 1 in 10 vehicles.

Start here · no integration, no contract

Send a 28-day ATSPM export. Get back a live findings queue in 48 hours.

The pilot fits inside the data agreements DOT signal-ops teams already use. No new sensors, no new contracts, no controller writes. We return a packet your engineers can review, approve, modify, or reject — and decide if it’s worth a conversation.

What you send
  • 28-day ATSPM export
    Or read-only access to your ATSPM endpoint. High-resolution controller events.
  • Controller config snapshot
    Pattern library, splits, ring/barrier structure, recall configuration. For audit + diff.
  • Top intersections list
    Up to 10 sites where corridor performance, pedestrian wait, or detector health matter most.
  • Maintenance log · optional
    Lets us distinguish detector recalibration from unlogged timing changes when anomalies appear.
What you get
  • Findings queue
    Anomalies, opportunities, and open hypotheses — each with a workflow state (new / blocked / shadowing / ready).
  • Evidence packet
    Per-finding hypothesis ledger, ruled-out alternatives, remaining unknowns, and external records still needed.
  • Timing-change candidates
    Controller-ready timing packet for review — engineer deploys through your existing vendor workflow.
  • Shadow-eval plan + methodology PDF
    How each proposed change is evaluated against the live plan before approval. Open methodology, no black box.
How U-Flow works · two rails, one queue

Two product lines read the same ATSPM stream and feed one engineer queue.

The Optimization Engine proposes specific timing changes, each anchored to what current operations already do. The Agentic Auditor surfaces anomalies, blockers, and the questions data alone cannot answer. Both land in the same engineer queue with the audit trail already started — the engineer’s job is to accept, modify, block, or reject, not to assemble context from scratch.

Rail 1 · Optimization Engine

Ingest → Recommend → Shadow → Export

  1. 01 · Ingest
    Read your existing ATSPM stream — advance detectors, stop bars, ped calls. Backfill 28 days, then poll continuously.
  2. 02 · Recommend
    Derive time-of-day plans, per-(TOD, phase) pedestrian recall rules, and HCM capacity flags from the stream.
  3. 03 · Shadow
    Each proposed change runs in shadow against the live plan before any engineer sign-off.
  4. 04 · Export
    Output is a controller-ready timing packet — engineer deploys through your existing vendor workflow. No autonomous deploys.
Rail 2 · Agentic Auditor

Hypothesize → Query → Rule-out → Surface

  1. 01 · Hypothesize
    Auditor proposes candidate causes for any pattern that looks off — split edit, cycle change, detector drift, upstream retiming.
  2. 02 · Query
    Pull the relevant ATSPM slice for each hypothesis — pattern log, EventCode 1↔7 pairing, weekly volume integrals.
  3. 03 · Rule out
    Mark each hypothesis ruled out (with evidence) or remaining open (with the data slice that would settle it).
  4. 04 · Surface
    Leave a finding in the queue with its full reasoning chain — defendable in an incident replay, not just at approval.
Both rails feed →Engineer operations queueexamples of what shows up
BlockedBlocked findingNeeds maintenance log
ReadyReady timing changeShadow passed · awaiting sign-off
BlockedExternal evidence neededRequest upstream timing sheets
ShadowingShadow check queuedP6 progression band shift
The second rail · Agentic Auditor

The Auditor doesn’t pretend to know everything. It uses up the data to its limit, then routes the sharpened question to the engineer who can answer it.

DOT operations is an audit culture. Every retiming gets logged, defended, and replayed when something goes wrong — so a black-box recommendation cannot survive that workflow. The Agentic Auditor runs alongside the Optimization Engine to make every finding defendable from the moment it surfaces.

Reading the same ATSPM stream as the Optimization Engine, the Auditor surfaces hypotheses, rules out everything the event stream can rule out, and writes the question that remains — including who to ask, and what evidence to attach when you ask. The audit trail is already started by the time it reaches the engineer’s queue.

Northbound green-band performance jumped 3.8× mid-month
Anomaly · confidence high · Blocked on agency input
Hypothesis ledger
Split edit pushed silentlyRuled outPattern-change log silent; per-week splits stable
Cycle length changedRuled outCycle held at 132–135s across all 4 weeks
Detector drift at this siteRuled outWeekly counts match within 2%
Upstream signal retimed? OpenSingle-intersection data can't see it
Detector recalibrated off-log? OpenNeeds maintenance log from agency
Methodology & data provenance

Audit findings are computed directly from raw ATSPM events.

Recommended timing changes go through shadow evaluation before any deployment. No simulation, no interpolation, no autonomous writes to the controller. Full method below — open to audit.

Read the full methodology
Source data

Raw ATSPM event stream from one signalized intersection, agency-partner site. 28 consecutive days. ~1.7M events / week. EventCodes 0–82 retained; configuration parsed from controller export.

Per-phase AOG

Advance-detector latency = distance / speed × 1.467. Pair EventCode 1→7 within a cycle window. Numerator = vehicles arriving during green; denominator = vehicles arriving total. No simulation, no interpolation.

Pedestrian wait

Time from EventCode 89 (call) to EventCode 21 (walk). 26,702 samples over the window. Distribution reported as p50 / p90 to avoid mean-skew from long tails.

Configuration audit

Pattern splits cross-checked against EventCode 31 history. NEMA dual-ring barrier-equal constraint verified per cycle. Anomaly threshold for “regime change”: ≥2× shift in per-phase AOG with no corresponding EventCode 31.

Why U-Flow

vs traditional adaptive systems

FactorSCATS / SCOOTProprietary AIU-Flow
Hardware costHigh — central system + commsVery high — new sensors per intersectionZero — uses existing ATSPM
TransparencyLimited — vendor-specific algorithmBlack boxFull why-explanation per recommendation
Reasoning & audit trailNot exposed to engineerNoneHypothesis ledger + ruled-out alternatives per finding
Engineer controlLimited overrideVery limitedFull approval workflow + 24h shadow
Deployment timeMonthsMonthsWeeks
Procurement pathSole-source vendorsSole-source vendorsOpen data + open methodology
Real controller data

28-day analysis on agency partner corridor. Raw ATSPM event stream — not synthetic, not simulated.

Sharpens, then routes

When data cannot fully explain a finding, the Auditor names the specific question that remains, identifies who can answer it, and starts the audit trail before the engineer opens the ticket.

Engineer-in-the-loop

Recommendations and findings go into the same queue with Accept / Modify / Block / Reject buttons. No autonomous writes to controllers. Every action is logged.

Compliance-ready

Section 508 / WCAG 2.1 AA. FedRAMP-ready hosting path. Audit log built in.

Send a 28-day ATSPM export

We’ll return a pilot audit packet for your top intersections.

Send a 28-day ATSPM export (or read-only access), the controller config snapshot, and a list of intersections that matter to you. We return a findings queue, evidence packet, and timing-change candidates — reviewed through shadow evaluation before any deployment.

Book a 20-minute call →Open the dashboard →

Calls are confidential. We sign DOT-standard data-use agreements before touching any logs.