ContinuityIQ

What problem does ContinuityIQ solve?

Leadership usually knows what an employee is supposed to own. The harder question is what actually depends on that person in practice.

Critical work is often scattered across job descriptions, meeting notes, procedures, project lists, help desk tickets, vendor records, system access, and undocumented personal knowledge. When that person becomes unavailable, management can discover too late that hidden responsibilities have no effective backup, the named backup lacks the access or authority to execute, or a recurring deadline depends on knowledge that exists only in one person's head.

ContinuityIQ asks:

What breaks if this person leaves tomorrow—and what should management transfer during the next 10 business days?

What it does

A manager loads one synthetic employee case package containing exactly seven sources. ContinuityIQ compares leadership's assumed view with evidence of day-to-day work.

The demo begins with three responsibilities leadership believes the role owns. The evidence reveals seven actual responsibilities, including:

• critical month-end work absent from the job description and formal procedures;
• a severity-one vendor escalation path with no effective backup; and
• a responsibility that appears covered on paper, but whose named backup lacks the access, instructions, context, and decision authority to execute it.

The resulting Continuity Risk Brief includes:

• a seven-row responsibility map comparing official and practical ownership;
• an explainable continuity-risk score of 93/100, with six visible contributing factors;
• three evidence-backed hidden-risk reveals;
• a compact Systems & Access Inventory covering systems, critical data assets, access or authority requirements, backups, and readiness gaps;
• clickable citations that open the exact synthetic source record and highlighted excerpt; and
• a prioritized five-phase, 10-business-day knowledge-transfer and handoff plan.

All data is fictional and synthetic.

How it works

ContinuityIQ uses one server-side Continuity Orchestrator to control exactly five bounded GPT-5.6 specialist stages:

1. Document Extraction
2. Responsibility Classification
3. Dependency & Coverage Mapping
4. Contradiction & Missing-Information
5. Executive Reporting & Handoff

Each stage is a separate GPT-5.6 invocation with one turn, strict structured output, no tools, no MCP servers, no sessions, no peer-to-peer communication, and no authority to publish a result.

Deterministic application code—not an agent—runs five validation gates:

• E1 validates source records, exact excerpts, evidence origins, duplicates, and canonical IDs.
• R1 reconciles the complete seven-responsibility signal map.
• D1 validates dependency types, data assets, backup coverage, and access readiness.
• F1 enforces the three required evidence-backed contradictions and missing-information findings.
• X1 rejects invented references, validates the executive brief and handoff priorities, and authorizes final publication.

The application also calculates the final numeric risk score deterministically. GPT-5.6 never chooses the score or bypasses the evidence gates.

How I used Codex and GPT-5.6

Codex was my primary development collaborator throughout the build. It helped turn the original concept into a narrow demo contract, PRD, technical specification, and 32-step implementation checklist. It then implemented the synthetic evidence package, strict schemas, deterministic validators, five specialist adapters, orchestrator, streamed interface, recovery path, accessibility coverage, and release evidence.

I made the key product and engineering decisions: focus the story on the three-to-seven responsibility reveal; use synthetic data only; bind the architecture to one orchestrator and five non-autonomous specialists; keep scoring deterministic; require citations for every major finding; and preserve an explicit cached recovery path rather than disguising a failed model call.

Live benchmark work used five complete GPT-5.6 workflows to tighten output envelopes without weakening deterministic authority. All five benchmark runs completed with five specialist invocations, all five gates passed, zero retries, and the same validated 93 High result.

Testing and reliability

The repository includes:

• 291 deterministic unit, contract, integration, and acceptance tests;
• four Playwright browser journeys covering live results, recovery, keyboard operation, and responsive layouts;
• lint, type-check, production build, secret scanning, and patch-integrity checks;
• five recorded successful GPT-5.6 benchmark workflows; and
• a validated cached recovery path that requires no API key and is always visibly labeled as cached.

Judges can clone the public repository and follow the README's no-key cached-demo path to inspect the complete interface, citations, inventory, and handoff plan without incurring API charges.

Built for a focused public demo

ContinuityIQ is intentionally a narrow hackathon MVP. It does not include arbitrary document uploads, production Microsoft 365 or HRIS integrations, enterprise authentication, automatic account deactivation, employee-performance scoring, departure prediction, or production-grade security claims.

The goal is to make a costly organizational blind spot visible, explainable, and actionable in under three minutes.

Built With

• codex
• gpt-5.6
• next.js
• openai
• openai-agents-sdk
• playwright
• react
• typescript
• vitest
• zod