epistemic-grounding

name	epistemic-grounding
description	Use when authoring any artifact that makes claims — skill content, documentation, audit findings, architecture proposals, code review comments. Covers the discipline of grounding every claim to a verifiable source, distinguishing verified-by-evidence from inferred-from-context, using normative vocabulary precisely (RFC 2119 MUST/SHOULD/MAY), and structuring arguments so the warrant from data to claim is visible to a reader. Do NOT use for verification protocol mechanics in this repo (use the verification-protocol rule file), for output-completeness enforcement (use methodology), or for self-scoring on a 1-5 scale (use self-evaluation).
license	MIT
allowed-tools	Read Grep
metadata	{"metadata":"{\"schema_version\":6,\"version\":\"1.0.0\",\"type\":\"capability\",\"category\":\"foundations\",\"domain\":\"foundations/epistemics\",\"scope\":\"reference\",\"owner\":\"skill-graph-maintainer\",\"freshness\":\"2026-05-15\",\"drift_check\":\"{\\\\\\\"last_verified\\\\\\\":\\\\\\\"2026-05-15\\\\\\\"}\",\"eval_artifacts\":\"planned\",\"eval_state\":\"unverified\",\"routing_eval\":\"absent\",\"comprehension_state\":\"present\",\"stability\":\"experimental\",\"keywords\":\"[\\\\\\\"epistemic grounding\\\\\\\",\\\\\\\"claim grounding\\\\\\\",\\\\\\\"source citation\\\\\\\",\\\\\\\"RFC 2119 modality\\\\\\\",\\\\\\\"Toulmin argument\\\\\\\",\\\\\\\"evidence-based assertion\\\\\\\",\\\\\\\"hallucination prevention\\\\\\\",\\\\\\\"normative vocabulary\\\\\\\",\\\\\\\"warrant\\\\\\\",\\\\\\\"epistemic hedge\\\\\\\"]\",\"triggers\":\"[\\\\\\\"ground this claim\\\\\\\",\\\\\\\"cite a source\\\\\\\",\\\\\\\"MUST vs SHOULD\\\\\\\",\\\\\\\"is this verified\\\\\\\",\\\\\\\"how do you know that\\\\\\\"]\",\"examples\":\"[\\\\\\\"before stating that this library supports X, confirm against the actual docs\\\\\\\",\\\\\\\"rewrite this finding so each assertion either cites a file or is marked as inference\\\\\\\",\\\\\\\"should this be a MUST or a SHOULD? what's the strength of the claim?\\\\\\\",\\\\\\\"the agent reported 'fix works' but no test was run — flag the gap in grounding\\\\\\\"]\",\"anti_examples\":\"[\\\\\\\"verify every step of an audit task with concrete evidence (use methodology)\\\\\\\",\\\\\\\"decide which lint rule to add for a specific kind of drift (use skill-infrastructure)\\\\\\\",\\\\\\\"evaluate a finished SKILL.md against the comprehension grader (use evaluation)\\\\\\\"]\",\"relations\":\"{\\\\\\\"related\\\\\\\":[\\\\\\\"methodology\\\\\\\",\\\\\\\"semantics\\\\\\\"],\\\\\\\"boundary\\\\\\\":[{\\\\\\\"skill\\\\\\\":\\\\\\\"methodology\\\\\\\",\\\\\\\"reason\\\\\\\":\\\\\\\"methodology enforces output-level completeness and step-level evidence receipts; epistemic-grounding is the upstream discipline that decides what counts as evidence in the first place.\\\\\\\"},{\\\\\\\"skill\\\\\\\":\\\\\\\"semantics\\\\\\\",\\\\\\\"reason\\\\\\\":\\\\\\\"semantics owns the rules for naming and meaning-making; epistemic-grounding owns the rules for grounding a claim to a verifiable source.\\\\\\\"}],\\\\\\\"verify_with\\\\\\\":[\\\\\\\"methodology\\\\\\\",\\\\\\\"evaluation\\\\\\\"]}\",\"mental_model\":\"|\",\"purpose\":\"|\",\"boundary\":\"|\",\"analogy\":\"Epistemic grounding is to claims what double-entry bookkeeping is to financial transactions — every assertion has a corresponding source on the other side of the ledger, and any entry without its pair is a red flag in the audit.\",\"misconception\":\"|\",\"concept\":\"{\\\\\\\"definition\\\\\\\":\\\\\\\"Epistemic grounding is the discipline of binding every assertion to a verifiable source, marking the modality (strength) of the claim, and making the warrant (the inference from source to claim) explicit. It is the practice that turns a generated statement into a defended statement.\\\\\\\",\\\\\\\"mental_model\\\\\\\":\\\\\\\"|\\\\\\\",\\\\\\\"purpose\\\\\\\":\\\\\\\"|\\\\\\\",\\\\\\\"boundary\\\\\\\":\\\\\\\"|\\\\\\\",\\\\\\\"taxonomy\\\\\\\":\\\\\\\"|\\\\\\\",\\\\\\\"analogy\\\\\\\":\\\\\\\"|\\\\\\\",\\\\\\\"misconception\\\\\\\":\\\\\\\"|\\\\\\\"}\",\"skill_graph_source_repo\":\"https://github.com/jacob-balslev/skill-graph\",\"skill_graph_protocol\":\"Skill Metadata Protocol v5\",\"skill_graph_project\":\"Skill Graph\",\"skill_graph_canonical_skill\":\"skills/epistemic-grounding/SKILL.md\"}","skill_graph_source_repo":"https://github.com/jacob-balslev/skill-graph","skill_graph_protocol":"Skill Metadata Protocol v4","skill_graph_project":"Skill Graph","skill_graph_canonical_skill":"skills/epistemic-grounding/SKILL.md"}

Epistemic Grounding

Coverage

The discipline of grounding every claim to a verifiable source, marking the modality (RFC 2119 MUST/SHOULD/MAY) of the claim, and making the warrant from source to claim explicit. Covers the six-primitive Toulmin argument structure (claim/data/warrant/backing/qualifier/rebuttal), the RFC 2119 normative vocabulary, the distinction between verified/inferred/hallucinated, how to hedge honestly without softening, and the failure modes (cargo-cult citation, stale grounding, vibe-based assertion). Applies to any artifact that makes claims: SKILL.md content, code review comments, audit findings, documentation, architecture proposals, agent output.

Philosophy

Confidence is not evidence. An LLM trained to be helpful produces fluent, confident text whether or not the underlying claim is true — RLHF rewards plausibility, not verification. The result is that ungrounded text is indistinguishable from grounded text by surface signals alone: tone, structure, and vocabulary look identical.

The countermeasure is not to ask the model to "be more careful" (it can't see its own confidence). It is to require structural signals that make grounding state visible: every claim has a source-warrant-qualifier triple in the text, or it carries an explicit hedge ("not verified", "as of 2026-05-15", "inferred from context"). A reader scanning the text can then tell at a glance which claims are grounded and which are not.

This discipline serves three downstream uses: (1) audit and review become tractable because the reviewer knows what to verify; (2) drift detection becomes possible because grounded claims expose their dependencies; (3) downstream agents reading the artifact can decide which claims to trust and which to re-verify.

The Six-Primitive Argument

Every grounded claim has six primitives, derived from Toulmin's argument structure (1958). Most agent output surfaces only the Claim and silently elides the rest.

Primitive	What it answers	Example
Claim	What is being asserted?	"HTTP DELETE is idempotent."
Data	What source supports the claim?	"RFC 9110 § 9.3.5"
Warrant	What rule connects data to claim?	"RFC 9110 normatively defines HTTP method semantics; § 9.3.5 specifies DELETE's idempotency."
Backing	What gives the warrant authority?	"IETF standards body; the RFC has been ratified and is the canonical HTTP/1.1 reference."
Qualifier