Address in, bill of materials out.

Watch

Two and a half minutes for both. The first is the tool; the second is the reason it exists.

1Run it

Three ways in. All of them reach the same engine.

Command line — no key, no network

python3 bom/bom_estimate.py --ta ta-33 --speed 55 --zone-length 2000 \
    --jurisdiction maryland --summary

Drop --summary for JSON. --ta, --speed and --zone-length are the job; everything else has a default that prints in the assumptions block. Useful extras: --night, --duration-hours and --road-type freeway, because those are what the device gates read.

HTTP

curl -s -u name:password -X POST https://bom-dev.apps.keryk.ai/estimate \
  -H 'Content-Type: application/json' \
  -d '{"ta":"ta-33","jurisdiction":"maryland","speed_mph":55,"zone_length_ft":2000}'

Pass address instead of speed_mph and it geocodes, which is the only path that needs the key. speed_mph overrides the geocoder when you already know the posted speed.

The page

Same host, root URL. The bill of materials is the page; everything justifying it hides behind one disclosure. Every page and every API call needs a username and password.

Tests

./run-tests.sh

Under a second, stdlib only, nothing touches the network. It runs the engine-gap tests, the spot snapshot, the two Arizona calibrations and the coverage count.

2Read the answer

Four numbers come back for channelizing devices. They are not interchangeable and the difference between them is the honest part of this tool.

Derived from the standard line_devices

Pure algebra off the standard. This is the defensible number. Diff against the ta-bom track using this one, never the point estimate.

Likely on site estimate_low / estimate_high

Line devices plus the site allowance band — driveways, side streets, the funnels a real job grows. This is the range that prints on the BOM line, not the number above it.

Point estimate point_estimate

A rate fitted against two real jobs. Read it as "the fit", never as "the accuracy".

Load the truck order_quantity

The high end rounded up to ten. This is what to put on a truck.

Share that is algebra derivable_share

Read this first. What fraction of the estimate is algebra rather than allowance. 64% on the Gilbert job, 44% on Chandler.

Governing sheet — read it backwards

The word governing is not the good outcome. It appears only on a federal run, which means no state ruleset was applied at all. A state run reads declared fallback: the component list is the federal figure, and only the numbers — taper, buffer, sign spacing, device type — are that state's. Both are honest. Neither is a state's own drawing. None has been transcribed yet.

When it will not answer

Two refusals, and they are the product working, not the product failing.

no speed limit available

The map lookup came back without a speed. Type the posted speed and run it again. The tool never invents one, because the speed band moves every quantity on the page.

no jurisdiction available

The address did not resolve, so nobody's rules can be chosen. Fix the location or pick the state. It will not quietly fall back to federal — federal is a different rulebook, not a safe default.

A third case is not a refusal and is easy to miss: a state with no overlay answers normally, as plain federal, with an orange banner. Do not present that as the state's standard.

Where the speed came from

Every answer now names its speed source on the surface. Best first:

state DOT / local GIS road-attributes layer

A published record. Measured.

the road's own OpenStreetMap speed tag

A real value someone entered, and it can be years out of date. Measured.

the national HPMS road inventory

A published record, Federal-Aid roads only. Measured.

OSM highway-class default table

Estimated.

lookup by highway tag plus urban/rural

Estimated, and not tuned against production data.

Nothing in this tool reads a posted sign. Every speed is either a record or an assumption. Check it against the sign before you order.

3The jurisdiction layer

This is the part a federal-shaped engine does not have, and the reason the tool is worth keeping. The same job, run in every jurisdiction, on a TA-33 at 55 mph over 2,000 ft.

41 come in above federal, 9 match it exactly, and 7 come in under. The matches are a finding, not a failure — those states genuinely defer, and each one says so in its sources rather than inventing a deviation.

Where a gate fires

Tell the engine a job runs at night and for four days and these states change what you are pricing. The count barely moves; the unit cost does.

• alabama drum
• alaska drum
• arizona vertical_panel
• arkansas cone_28in
• california cone_28in
• chandler vertical_panel
• colorado cone_36in
• connecticut drum
• georgia drum
• gilbert vertical_panel
• illinois drum
• indiana drum
• iowa channelizer_42in
• kansas drum
• kentucky drum
• louisiana drum
• maricopa vertical_panel
• maryland drum
• mesa vertical_panel
• michigan drum
• minnesota cone_42
• missouri cone_42
• montana drum
• nebraska drum
• new-jersey drum
• new-mexico drum
• north-carolina drum
• ohio drum
• oregon cone_28in
• pennsylvania drum
• phoenix vertical_panel
• south-carolina cone_42
• south-dakota cone_42
• tempe vertical_panel
• tennessee drum
• utah drum
• washington drum
• west-virginia drum
• wisconsin drum
• wyoming cone_42

The number to lead with

Maryland more than doubles the federal device count on that job, 80 against 36. Illinois is higher still at 89. Those are the ones to put on a slide.

Quick reference