Public evidenceVersioned methodsAuditable decisions
Urban decisions,
with an evidence trail.
Build reproducible urban intervention analyses that expose the evidence, assumptions, uncertainty—and when no recommendation is justified.
Read-only public walkthroughVerified release snapshot · v0.8.0
G discovery S screening D deep evidence
From evidence to release
One traceable path.
Five evidence types.
CivicDecision OS is built for urban analysts and civic data teams who need a reviewable record of how an intervention screen reached—or refused—a result.
Type the evidence
Keep observed, estimated, simulated, optimized, and proposed claims separate.
Run the decision
Replay, simulate, and optimize within explicit data and method gates.
Test reversals
Show which assumptions can change the selected bounded option.
Release or withhold
Emit a DecisionPack—or preserve an auditable negative result.
Guided reference case
Suffolk heat access.
Including the right to say no.
Two deterministic runs use the same bounded public-data sample: one satisfies the declared constraints; one deliberately makes the candidate set infeasible.
Reference DecisionPack
A bounded option satisfies every declared hard constraint.
Run
run-aff7c38b12c1Resultplan-25025000202-25025000502
Combinations evaluated55
Feasible combinations16
Estimated proxy covered3,121.507
Overall coverage rate96.35%
Five service-radius tests produced three different selected bounded options, making the result visibly assumption-sensitive.
10 parsed public-data rows in the bounded reference sample.
3,239.695 people represented by an area-level need proxy.
Straight-line coverage around tract-centroid candidates.
55 bounded combinations evaluated against declared constraints.
Candidate points are demonstrations—not verified facilities.
Tract centroids are not verified facilities; straight-line radius is not travel time; the population proxy is not individual demand. The selected bounded option is a methods result, not an implementation recommendation.
Fail-closed by design
What the system
refuses to overclaim.
A valid output can be completed, infeasible, or insufficient-evidence. Negative evidence is retained instead of being hidden.
Simulation ≠ observed impact
Modeled draws remain conditional on the declared inputs and assumptions.
Optimization ≠ adoption
A selected bounded option does not prove institutional approval or implementation.
Coverage ≠ readiness
A city point or public-data screen does not establish a local intervention evidence base.
Tests ≠ policy validity
Software verification establishes implementation behavior, not real-world effectiveness.
Five-minute quickstart
Inspect the method.
Reproduce the artifact.
Install the repository, open the full local Evidence Explorer, or reproduce the bounded heat-access DecisionPack from committed inputs.
local terminal
git clone https://github.com/limingrui679-design/civicdecision-os.git
cd civicdecision-os
python -m venv .venv
. .venv/bin/activate
python -m pip install -e '.[api]'
civicdecision serve --root .Then open http://127.0.0.1:8000