Glossary of Sprints
& AI Terms

A fixed-scope, fixed-price software or AI delivery engagement conducted async-first — no video calls, no standups, no timezone coordination required. Deliverables are defined in writing before work begins. The client sends specs in any format (bullet list, Figma, PDF, voice memo). The studio confirms scope, timeline, and price within 24 hours. Work proceeds without check-ins. The client receives source files, a walkthrough video, and a README on the agreed delivery date.

Payment: 50% on contract signing, 50% on delivery. Accepted: Stripe, wire transfer, USDC/ETH.

A software development engagement where the price is confirmed in writing before work begins and does not change regardless of hours spent. Contrasts with: hourly billing (price unknown until invoice), T&M — time and materials (price estimated, often exceeded), and retainers (ongoing commitment with variable output). In a fixed-price sprint, scope changes result in a separate sprint at the original price — the original engagement is unaffected.

All prices in USD. All sprints are fixed-price, fully async. One revision included. Client owns 100% of IP.

Amatrix's mandatory output contract for all TheHub AI agents. Every agent must return a typed response with four guaranteed fields:

output — Typed domain-specific JSON. Schema varies by agent but is always defined and documented.
confidence — Float 0.0 to 1.0. Always present, never null. Indicates the agent's self-assessed certainty in its response.
warnings — Array of strings, nullable. Contains advisory notes when confidence is partial or assumptions were made.
graceful_degradation — Boolean, always true. On failure, the agent returns a structured error object — never an unhandled exception or hallucinated response.

This contract enables safe agent chaining: downstream systems rely on the typed structure, not the prose, and handle failures programmatically without defensive coding at the integration layer.

Amatrix's multi-agent coordination architecture powering TheHub. Four defining properties:

1. Typed contract isolation — Every agent exposes a typed JSON schema contract. Agents are black boxes to each other; they communicate only through schema-validated outputs.

2. Orchestration separation — OpenClawMatrix handles routing, parallel execution, retry with exponential backoff, and fallback activation. Agents don't know they are being orchestrated — separation of concerns is complete.

3. Policy Enforcement Engine — Compliance rules are applied at every orchestration step without embedding policy logic in individual agents. Policy changes do not require agent redeployment.

4. Execution Tracker — Every autonomous action is logged with inputs, outputs, confidence scores, and timestamps. Full audit trail without agent-level logging code.

The property of an AI agent that returns a structured error object when it cannot complete a task with sufficient confidence, rather than hallucinating a plausible-sounding answer, throwing an unhandled exception, or failing silently. A gracefully degrading agent always returns a typed response the calling system can handle programmatically — the calling system never needs to detect or catch unexpected failure modes.

A float between 0.0 and 1.0 returned by every Amatrix TheHub agent alongside its output. Never null. Indicates the agent's self-assessed certainty in its response. Downstream systems use the confidence score to decide: > 0.85 — auto-apply output; 0.60–0.85 — queue for human review; < 0.60 — trigger fallback agent or human escalation. The confidence score is not a probability — it is an agent-reported uncertainty signal designed for programmatic consumption.

Running large language model inference on hardware you own and control, with no dependency on cloud providers (AWS, Azure, GCP, OpenAI, Anthropic). Data never leaves your physical environment. Amatrix implements this via SovrinOS — a bootable OS for NVIDIA and AMD GPU nodes deployed on-premise. Contrasts with API-based inference where prompts and responses transit third-party infrastructure and may be logged, used for training, or subject to regional data laws.

In the Amatrix TheHub platform, a matrix is a curated suite of AI agents specialized for a single vertical domain. Each matrix contains 15–80 agents sharing a common domain schema, pricing tier ($895–$1,295/month), and orchestration context. All agents within a matrix implement the Agent Schema Standard. Matrices are independently subscribable. Example: FinanceMatrix (30 agents for financial modeling, portfolio analysis, tax optimization, fraud detection, and retirement planning).

Current roadmap: 6 live matrices (Q2 2026), 6 beta (Q3 2026), 6 shipping (Q4 2026), 9 planned (2027) — 27 total by end of 2027.

The 2026 evolution of SEO. GEO optimizes content so that AI answer engines cite your site as a source — rather than optimizing for click-through from traditional search result pages. Relevant engines: Google AI Overviews, Perplexity, ChatGPT, Claude. GEO tactics: structured JSON-LD schema markup, named frameworks with unique terminology, llms.txt and llms-full.txt files, Information Gain content (definitions that exist nowhere else), high-confidence factual statements, and freshness signals (dateModified on technical content).