Use this chooser when you need a strict “facts-only” output contract but must decide what evidence is allowed.
How-to guides
Purpose
This hub is the procedure catalog for the site. Use it when you already know you need a step-by-step guide.
Who this is for
- People using AI assistants/tools for daily work (drafting, summarizing, analysis).
- Builders and reviewers of agentic workflows who need repeatable procedures (verification, evidence boundaries, engineering gates).
How to use this page
- If you are new to the site, start from the homepage.
- Use On this page to jump directly to a procedure category, or scan the full How-to catalog.
- If you need prerequisites or definitions, open Before you run.
If you need an overview of content types (How-to vs Reference vs Articles), use the Content map.
On this page
Jump directly to a procedure category.
How-to catalog (scan)
Evidence boundaries
Use only user-provided files/logs/screenshots; no web browsing.
Browsing allowed; claims require citation-grade sources.
Verification workflows
A concrete procedure to reduce factual errors and improve auditability.
Open guide
Policies (4) Hide policies
Use this procedure to run a structured verification loop before emitting a final answer.
Open guide
Read policy
Prompt templates (8) Hide prompt templates
-
System
chain-of-verification.system.txt -
System
facts-only-artifacts-only.system.txt -
System
facts-only-authoritative-sources-required.system.txt -
System
web-verification-and-citations.system.txt -
User
web-browsing.user.txt -
User
citations-output-contract.user.txt -
System
confidence-score.system.txt -
System
instruction-hierarchy-and-evidence-boundary.system.txt
Verification procedure (claims).
Architecture classification, layering/boundary analysis, and minimal structural remediation.
Source-backed review of code or configuration changes against official guidance.
Terminology consistency + prevent overclaims.
Web verification & citations
This guide provides copy/paste templates to request web browsing / web search from an agent.
Modes & reporting
This guide explains how to run EGAM as a profile layered on authoritative sources.
Ensure every response includes an evidence-based analytic confidence score.
Scholarly review
Run the scholarly literature review stack to produce one evidence-gated scholarly review output from inspected sources.
Tooling & prompting
Route day-to-day prompting to one of two starter stacks: files-only or web-verified.
Open guide
Prompt templates (8) Hide prompt templates
-
System
facts-only-artifacts-only.system.txt -
Component
deep-read.user.txt -
Component
deep-scan.user.txt -
System
facts-only-authoritative-sources-required.system.txt -
System
web-verification-and-citations.system.txt -
User
web-browsing.user.txt -
System
chain-of-verification.system.txt -
System
confidence-score.system.txt
Memory & context management
Procedure + checklist for making “memory” behavior predictable and auditable.
Baseline (recommended for all runs) Baseline (hide)
Read/apply these before running any procedure (they define the default constraints and enforcement posture).
Terminology Terminology (hide)
- Evidence boundary: which sources are allowed (e.g., artifacts-only vs authoritative sources).
- Artifacts: files/logs/screenshots/excerpts you provide in the request.
- Authoritative sources: externally published, citable sources (standards, official docs, peer-reviewed papers).
- Claim: any non-trivial statement that should be verified before output.
- Verification procedure: a repeatable sequence that checks claims before output.
- Fail-closed: if required evidence is missing, stop and report what’s missing (don’t fill gaps).
How the pieces fit How the pieces fit (hide)
- Policies = rules (evidence boundaries, citation requirements, fail-closed behavior).
- Prompt templates = copy/paste prompt files you load into your runtime (typically as system/developer/user messages).
- How-to guides = the procedure steps for applying the policy + templates to a task.