4
Operational states
OBSERVE, PREPARE, PROTECT and ESCALATE, always reviewed by an authorized human.
OpenAI Build Week 2026 · Work & Productivity
From territorial noise to an explainable decision.
NERHIA helps municipal and emergency operations teams combine weather, territorial vulnerability, visual infrastructure and governed evidence into one reviewable action path. No login or test credentials are required.
Checking the public judging contract…
7
Governed layers
Core, Eye, Urban, Centinela, VA, Connect and Lab keep evidence and action boundaries explicit.
0
Credentials required
The production judge path and deterministic sample contracts work without a test account.
Guided functional evaluation
Four proofs in under three minutes
Open each live surface in a new tab, verify the expected behavior, then mark the proof complete.
1
Threat becomes territorial impact
Compare Pichasca and Buin. The same rain produces different decisions because exposure and vulnerability are local.
Expected proof: low confidence increases caution; it never produces a false green.2
Current conditions meet structural memory
Cross the threat with reported winter critical points and seasonal evidence.
Expected proof: missing records are labeled as incomplete coverage, never interpreted as safety.3
Evidence becomes a reviewable action path
Inspect incidents, resources, maps, recommendations, actions and their evidence in one operational contract.
Expected proof: NERHIA recommends; it does not automatically evacuate, close roads or dispatch resources.4
Every production claim is auditable
Review executable verification, SHA-256 artifacts, freshness policy and fail-closed evidence gates.
Expected proof: a successful network response is not silently promoted into operational truth.Deterministic sample scenario
One forecast, two territorial realities
Quebradas, river access, bridges, connectivity and distance shape the recommended state.
Drainage, canals, roads, schools and health services create a different exposure profile.
Sample contracts are versioned under public/data/; no judge needs to upload data or configure an API key.
How GPT-5.6 and Codex accelerated the work
Repository-scale engineering with explicit human decisions
1 · Human doctrineArchitecture, territories, authority and safety boundaries.
2 · Codex taskRepository exploration and bounded implementation plan.
3 · BuildContracts, adapters, UI surfaces and deterministic fixtures.
4 · VerifyValidators, CI diagnosis, exact repair and diff review.
5 · ShipReviewable PR, keyless deployment and smoke checks.
NERHIA existed before the event. PRs #1026–#1040 document the meaningful extension completed during the submission window. Later security maintenance is identified separately and is not presented as event-period work.
Try without rebuilding
Public production instance
Use the guided route above. No login, credential, browser extension or local setup is required.
Run from source
One deterministic judge command
npm install npm run validate:build-week
Node.js 20+ · Windows, macOS and Linux. See the repository README for local hosting.
Challenge requirement matrix
Everything a judge needs, in one place
| Requirement | Evidence | Status |
|---|---|---|
| Functional project | Live judge path and production routes | READY |
| Category | Work & Productivity | READY |
| Description | Repository submission document and this page | READY |
| Public demo video under 3 minutes | YouTube demo ↗ | VERIFY PUBLIC |
| Repository and testing | README, sample contracts and no-login demo | READY |
| Codex / GPT-5.6 evidence | Dated PRs, decisions and validation workflow | READY |
/feedback session ID | Entered privately in Devpost; never published here | DEVPOST ONLY |