Prototype: compile and simulate
The P1.6 prototype is the smallest end-to-end path that exercises the contracts. It compiles the
synthetic fixture into a validated dataset, replays a plan or schedule, and writes a run directory with
the account values, state values, schedule, report, and run.json.
It is not the calibrated model. The growth rule is linear and the treatment transition is a simple fractional reduction. The calibrated model is P4.
Compile
brushline build --fixture examples/synthetic-corridor --out build/compiled
The compiler reads dataset/segments.csv, corridor.json, and state.json, joins the service
standards and corridor attributes onto the segments, and writes:
segments.parquet: segment id, corridor, length, geometry, span, sequence, treatment unit, service standard, access point, maximum height, and minimum clearance.states.parquet: segment, year, functional group, state class, height, cover, density, age, confidence, source, and source version.manifest.json: thebrushline.dataset.v1manifest with both tables, checksums, and the content digest.
Simulate
brushline simulate \
--dataset build/compiled \
--out runs/prototype \
--plan examples/synthetic-corridor/plan.json \
--years 5 \
--start-year 2026
Or replay a schedule CSV:
brushline simulate \
--dataset build/compiled \
--out runs/prototype \
--schedule path/to/schedule.csv \
--years 5
The schedule CSV columns are segment_id, year, action_id, method, intensity,
efficacy, and cost. Exactly one of --plan and --schedule is required.
The prototype writes:
schedule.csv: the replayed actions withappliedoroutside_horizonstatus.account_values.csv: long account values by year.state_values.csv: per-segment, per-year state values.reports/summary.md: a short run report.run.json: thebrushline.run.v1manifest with the artifact index and summary.
One-command path
brushline prototype \
--fixture examples/synthetic-corridor \
--out runs/prototype \
--years 5
The command compiles under runs/prototype/dataset and simulates under runs/prototype/run.
Prototype rules
Growth: height increases by 0.25 m per year and cover by 2 percentage points per year, capped at 10 m and 100 percent.
Density: grows by 5 percent per year when it is known.
Treatment: height and cover are multiplied by
1 - efficacy; density is multiplied by1 - 0.5 * efficacy; the action cost is charged in its year.An unknown state does not grow and remains unknown.
A clearing violation is a known state whose height exceeds the segment’s service standard maximum.
No action is invented: the simulator replays only the plan or schedule.
Accounts
The prototype writes seven accounts:
Account |
Unit |
Meaning |
|---|---|---|
|
m |
Mean height across known states. |
|
% |
Mean cover across known states. |
|
m |
Maximum height across known states. |
|
count |
Segments whose state is unknown. |
|
count |
Known states above the service standard maximum. |
|
currency |
Cost charged in the year. |
|
count |
Actions applied in the year. |
Reproducibility
The simulator is deterministic given the dataset and the schedule. The same schedule reproduces the same
account_values.csv and state_values.csv. run.json records the dataset content digest, the
prototype model versions, the configuration, and the artifact checksums.
Validation
The test suite compiles the fixture, simulates the plan, checks the five-year accounts, replays the written schedule, compares the outputs, and exercises the three CLI paths. It also checks that a schedule with an unknown segment fails.