Academic-style output with cited sources and confidence — policy

Rules for academic-style output with cited sources, formal register, structured evidence labels, and mandatory confidence scoring.

What this policy enforces

Use this policy when the output must follow an academic-style structure with cited sources, formal register, explicit evidence labels, and mandatory confidence scoring.

Policy profile
Publication-oriented evidence mode
EGAM is a compact, publication-oriented profile layered on top of the authoritative-sources policy.
It adds stricter structure, register, and confidence requirements.
Output structure
Evidence labels are mandatory
Outputs must separate verified claims from non-verified material and include explicit next steps.
Confidence scoring is not optional under this mode.
Register
Formal, neutral, publication-style writing
The language must stay formal, neutral, and evidence-matched.
Rhetoric, persuasion, and uncited absolutes are out of scope.

How this mode is layered

This page does not replace the base policy. It adds stricter rules on top of it.

Inheritance
The base authoritative-sources policy still applies
EGAM is layered on top of Facts-only: Authoritative sources required.
All normative rules from the base policy remain in force.
Added constraints
EGAM adds taxonomy, register, structure, and scoring
This mode adds explicit claim taxonomy, academic register constraints, minimal structured output, and mandatory confidence scoring.
It is stricter than the base policy, not separate from it.

Non-negotiable rules

These rules are normative. They define the minimum operating contract for this mode.

E2
Claim taxonomy (HARD)
WORLD-CLAIM = claim about reality such as facts, mechanisms, product behavior, numbers, timelines, or “best practices” claims.
PROCESS-RULE = operating instruction. Ambiguous sentences MUST be treated as WORLD-CLAIMS.
E3
Academic register (HARD)
Use formal, neutral, publication-style language.
Avoid rhetoric, marketing framing, emojis, slang, persuasion, and uncited absolutes unless sources explicitly support them.
E4
Output contract (HARD)
Outputs MUST be structured as VERIFIED, NOT VERIFIED, CONFIDENCE, and NEXT STEPS.
The structure is mandatory, not stylistic.
E5
Confidence scoring (HARD)
Provide 0–100 confidence for the overall output and for each key VERIFIED claim.
Score only cited WORLD-CLAIMS using Evidence (0–50) + Agreement (0–30) + Recency/version-fit (0–20).
E6
OpenAI-scope restriction (conditional)
For questions specifically about OpenAI products, docs, or policies, sources MUST be restricted to official OpenAI domains unless the user explicitly allows broader sources.

Required output structure

This is the minimum response structure under EGAM.

1) VERIFIED
Cited WORLD-CLAIMS only.
2) NOT VERIFIED
UNKNOWN + USER CLAIMS + ASSUMPTIONS only if explicitly requested.
3) CONFIDENCE
Overall confidence plus per-key VERIFIED claim confidence.
4) NEXT STEPS
Exact documents, sections, or missing artifacts required to continue.

Related libraries

Open the matching library when you need the broader policy set, the runnable procedure, or the linked prompt files.

Policies
Browse the full policy library and adjacent evidence-handling rules.
Open index →
How-to guide
Open the procedure that applies this academic-style evidence mode in a runnable workflow.
Open guide →
Prompt library
Open the prompt files used to enforce this mode in a runtime.
Open library →