Urban Decision OS · governance & human review
Governance & Human Review
The platform produces evidence-backed review context for named municipal reviewers. It does not produce automated municipal decisions, does not write to municipal records, and does not act on its own. Every consequential transition is initiated by a human reviewer and recorded in a digest-verified audit log.
Governance posture
- Advisory only
- Human review required
- Evidence-linked
- No automated action
Section 1
AI governance
The platform is aligned to two published frameworks. Alignment is a structural posture, not a certification, and no third-party authority has endorsed this alignment.
Framework · NIST AI Risk Management Framework
Function
Govern
Constitutional commitments, owner of record, and a published policy set behind every surface.
Function
Map
Each case carries a documented context, an input class, and a named reviewer cohort.
Function
Measure
Rule-based qualitative bands and a digest-verified audit log — no opaque scoring.
Function
Manage
Caveats, escalation, and reviewer-initiated state changes — never auto-routed.
Framework · ISO/IEC 42001
- AI management-system alignment — not certification.
- Owner of record, documented policies, versioned changes, and a defensibility export path.
Section 2
Human review
A case moves through a published review state machine. Reviewers are assigned by the institution; the platform reads role assignments from a configuration file and never assigns reviewers itself.
Review flow
Step 1
Queued
Case is registered; intake reviewer has not yet picked it up.
Step 2
In review
A named case reviewer is working the evidence pack.
Step 3
Awaiting evidence
Reviewer has paused the case pending additional cited inputs.
Step 4
Ready for executive review
Senior reviewer has concurred; the pack is prepared for executive sign-off.
Step 5
Closed
Executive reviewer has recorded a decision (accept / defer / decline).
Operating guarantees
A reviewer of record exists for every consequential transition.
Review is required; the platform never advances a case on its own.
No auto-approval and no auto-routing between reviewers.
No enforcement, no penalty, no automated municipal action.
Section 3
Algorithmic transparency
The platform publishes — in plain language — what it uses, what it produces, what it does not do, and what recourse a reviewer has if they disagree with a readiness band.
Block
What the system uses
- Public-source imagery, terrain, and overlays (cited).
- Public-source statistical and policy documents (cited).
- Tenant-provided inputs under a signed data-sharing agreement.
- Platform-generated audit records (append-only, digest-verified).
Block
What the system outputs
- Evidence packs assembled from cited inputs.
- A qualitative decision-readiness band per case.
- A reviewer-initiated hand-off into the executive review step.
- A digest-verified record of every consequential state change.
Block
What the system does not do
- Make municipal decisions.
- Take automated action.
- Write to municipal records.
- Produce predictive enforcement output.
- Monitor citizens, properties, or organisations.
- Compute or display a numeric decision score.
Block
Recourse path
- A reviewer who disagrees records a caveat on the case.
- The case can be moved to awaiting-evidence to demand more inputs.
- Escalation to a senior reviewer is always available.
- The platform never blocks a reviewer's escalation.
Section 4
Data classification
The platform accepts five operating data classes and explicitly excludes a sixth (personal identifiers).
Accepted classes
Class
Public-source data
Imagery, terrain, vector overlays, statistics, policy documents — all cited.
Class
Tenant-provided data
Accepted only under a signed data-sharing agreement with a residency statement.
Class
Restricted municipal data
Read-only, scoped to the case under review, never exported outside the tenant boundary.
Class
Geospatial data
Illustrative spatial framing — never coordinate-accurate, never cadastral-authoritative.
Class
Audit data
Append-only, digest-verified, stored within the tenant boundary.
Excluded by policy
- Personal identifiers (national ID, deed number, device ID).
- Citizen profiling of any shape.
- Per-person monitoring, tracking, or behavioural traces.
Section 5
Decision readiness
Readiness is reported as one of four qualitative bands derived from a rule-based composition. No numeric decision score is computed or displayed.
Band
Not ready
A structural problem blocks executive review — typically a governance-clarity gap.
Band
Evidence partial
The pack is being assembled; cited gaps remain open.
Band
Review required
The pack is complete; case-reviewer or senior reviewer is working it.
Band
Ready for executive review
Pack complete and senior concurrence recorded — awaiting executive hand-off.
Important
Decision readiness reflects whether the case is ready for executive review. It is not a decision and does not recommend an outcome.
Section 6
Claims language
Every user-visible string is bound by a published language policy. The list below is illustrative — the full vocabulary is enforced by the linter against every touched file.
Safe language
- Advisory context
- Review readiness
- Evidence-backed review
- Source-linked evidence
Forbidden positioning
- Surveillance
- Enforcement
- Automated decision
- Official integration
- ROI guarantee