Use Cases¶

Entropy-Gated AI Framework¶

Use cases demonstrate the framework applied to real, concrete tasks. Each entry maps a task category to its corresponding charged vial, explains what the session governs, and describes what a correct output looks like under that process mode.

All use cases on this page are executable with the pre-built charged vials in shelf/vials/charged/. Load the vial in the Gaffer or paste the manual variant into any frontier chatbot interface.

Security configuration analysis¶

Vial: VIAL_SP_SECURITY_ANALYSIS
Mode: SINGLE_PASS
Process: Atomic, one-shot. The artifact is submitted once. One governed report is produced.

What it governs¶

A firewall ruleset, IAM policy, network configuration, server configuration, or security log excerpt is submitted as the artifact. ROLE_ASSISTANT analyzes it within the declared scope and reports findings. No new content is generated. No remediation outside the declared scope is permitted.

Artifact types in scope¶

Firewall rules and ACLs
IAM and RBAC policy documents
Network topology configurations
Server and service hardening configurations
Security log excerpts
Policy documents

What the output contains¶

A structured report with three sections: Artifact Summary (type and completeness), Findings (severity-classified table), and Risk Summary (one-paragraph overall posture assessment).

Severity levels: CRITICAL / HIGH / MEDIUM / LOW

Why SINGLE_PASS¶

Security analysis against a declared artifact is a bounded, fully-defined task. All inputs are present upfront. One governed output is required. Retries under this mode require editing the E_T and resubmitting — not continuing a conversation.

Gaffer path¶

shelf/vials/charged/gaffer/VIAL_SP_SECURITY_ANALYSIS.md

Manual path¶

shelf/vials/charged/manual/VIAL_SP_SECURITY_ANALYSIS.md

Code review¶

Vial: VIAL_SP_CODE_REVIEW
Mode: SINGLE_PASS
Process: Atomic, one-shot.

What it governs¶

A code artifact is submitted. ROLE_ASSISTANT evaluates it across five dimensions — correctness, security, readability, error handling, and performance — and produces a structured report. No rewriting occurs unless explicitly requested in a custom E_T.

Dimensions evaluated¶

Correctness — logic errors, off-by-one errors, unhandled edge cases
Security — injection vulnerabilities, unsafe input handling, exposed secrets, insecure defaults
Readability — naming clarity, comment quality, structural complexity
Error handling — uncaught exceptions, missing null checks, silent failures
Performance — obvious inefficiencies, unnecessary loops, memory concerns

What the output contains¶

A structured report with three sections: Summary (overall assessment), Findings (severity-classified per-issue entries), and Verdict (PASS / PASS WITH NOTES / FAIL).

Gaffer path¶

shelf/vials/charged/gaffer/VIAL_SP_CODE_REVIEW.md

Manual path¶

shelf/vials/charged/manual/VIAL_SP_CODE_REVIEW.md

Structured comparison¶

Vial: VIAL_SP_COMPARISON
Mode: SINGLE_PASS
Process: Atomic, one-shot.

What it governs¶

Two artifacts — ARTIFACT_A and ARTIFACT_B — are submitted together. ROLE_ASSISTANT performs a structured side-by-side comparison and reports differences across five dimensions. No recommendation is made about which artifact is better unless explicitly requested.

Artifact types in scope¶

Documents and specifications
Code files
Configuration files
Policy texts
Data schemas

Comparison dimensions¶

Structure — organizational differences, hierarchy, section presence or absence
Content — substantive differences in meaning, logic, or data
Additions — elements present in B but not in A
Removals — elements present in A but not in B
Conflicts — contradictory statements or values between A and B

What the output contains¶

Artifact Identification, Comparison Table (dimension-by-dimension delta), Conflict Register, and Summary.

Gaffer path¶

shelf/vials/charged/gaffer/VIAL_SP_COMPARISON.md

Manual path¶

shelf/vials/charged/manual/VIAL_SP_COMPARISON.md

Document drafting¶

Vial: VIAL_MT_DOCUMENT_DRAFT
Mode: MULTI_TURN
Process: Iterative, staged. One section per turn. Each section is accepted before the next begins.

What it governs¶

A document is drafted section by section across multiple turns. ROLE_USER provides the title, audience, section outline, and tone on the first turn. Each subsequent turn produces exactly one section. Accepted sections become the input for the next. The session closes when all sections are accepted and the full document is assembled.

What it prevents¶

Drafting ahead of accepted output
Accumulating unreviewed content that influences later sections
Tone or voice drift across long documents
Implicit acceptance of sections the user has not reviewed

What the output contains¶

Per turn: section heading, section content, word count. Final turn: complete assembled document.

Why MULTI_TURN¶

Long-form documents require staged review. Each section builds on the last. MULTI_TURN ensures that only accepted sections influence subsequent drafting — preventing the document from drifting across turns.

Gaffer path¶

shelf/vials/charged/gaffer/VIAL_MT_DOCUMENT_DRAFT.md

Manual path¶

shelf/vials/charged/manual/VIAL_MT_DOCUMENT_DRAFT.md

Governed general assistant¶

Vial: VIAL_IX_GOVERNED_CHAT
Mode: INTERACTIVE
Process: Governed conversational mode. Open scope. Passive entropy monitoring.

What it governs¶

A governed general-purpose session. ROLE_USER may ask questions, request analysis, or work through problems conversationally. All OIL_CONTRACT constraints remain active. ROLE_ASSISTANT monitors for constraint pressure and issues SOFT_WARN when detected. Authority boundaries are enforced throughout.

How it differs from ungoverned chat¶

In ungoverned chat, the model fills gaps, infers intent, and carries forward unvalidated outputs silently. In INTERACTIVE mode under EGAF, all of those behaviors are prohibited. ROLE_ASSISTANT surfaces ambiguity rather than resolving it and will not expand scope beyond what ROLE_USER has declared.

Appropriate for¶

Exploratory research with defined scope boundaries
Iterative analysis where governance assurance matters
General assistance where accountability is required
Any session where the user wants governance without rigid structured output

Gaffer path¶

shelf/vials/charged/gaffer/VIAL_IX_GOVERNED_CHAT.md

Manual path¶

shelf/vials/charged/manual/VIAL_IX_GOVERNED_CHAT.md

Building a custom use case¶

All five use cases above use pre-built charged vials. For tasks that don't match a pre-built use case, a custom E_T can be authored and loaded into any primed vial or after the pimed vial.

Gaffer: Download the E_T Template from the toolbar, fill all fields, and upload it. The Gaffer loads it into the active primed vial automatically.

Manual: Download the E_T template from spec/E_T_MANUAL.md, fill all fields, and paste it as the next turn after seal acknowledgment.

See Quick Start — Load an E_T for the full procedure, and For Developers — Writing an E_T for structural rules.