artistic-learning-evolution git:20261008.2dc051eA

Immutable. This exact content never changes and is served at /api/v1/blob/06ae8e5888b5bf45.

---
name: artistic-learning-evolution
description: Convert independently argued film-art findings into contextual creative Skill candidates and controlled design comparisons; no Runtime learning, automatic promotion or production generation.
---

# Artistic Learning and Evolution

Consume pinned observations, independent professional/holistic judgments, counterarguments, current creative guidance and its actual handoff. Own the learning study, not source interpretation, specialist design or film adoption.

Diagnose the causal level before editing guidance: absent craft principle; existing principle not used; used but artistically weak choice; design-to-generation mismatch; or evidence insufficient. A disappointing output does not prove its Skill is missing. Reuse existing guidance or improve its use before proposing another rule.

Record observation → intent/effect difference → attributed owners → counterargument → smallest candidate → contextual principle → test. Choose only relevant Skills; a cross-professional finding need not modify all participants. Learn from successful restraint and ambiguity too.

Create an isolated candidate copy or patch pinned to the original Skill checksum. Change only artistic guidance, preserving technical paragraphs, context/tools/permissions and production contracts. Prefer revising, fusing or replacing an existing principle over indefinite appended prohibitions. Include applicable context, a counterexample and a withdrawal condition.

For a controlled comparison, keep source, task, major inputs, author capability and evaluation questions constant. Only the selected guidance changes; use separate fresh author passes without showing a desired result. Give a blinded independent critic the same source and anonymous alternatives. Compare concrete playable differences and costs, not prompt length or the number of instructions. An explanatory interpretation may conceal a handoff problem: test the actual professional design that must use the insight, including its effect on the next owner. A tied or mixed single comparison does not invalidate every learning direction. Locate the failure, refine or withdraw the method, and preserve contrary results rather than repeating an unchanged candidate until it wins.

Separate evidence levels: guidance altered a judgment; a creative design improved; a realized film improved. Text-only tests can establish the first two, never the third. Replication and future media observations remain open where needed. Retain mixed or negative results and revise or withdraw the candidate.

Report promotion status and limits. Candidates need independent review and the user's requested adoption route before replacing production guidance; this Skill never merges its own learning. When the user explicitly requests local creative-guidance enhancement, complete that route with an actual artistic-body revision supported by the independent comparison, or restore the original if its cost outweighs its benefit. Preserve the original and explain what changed in a consumer's design, what remains unproved and when to revisit the revision. This is a contextual craft decision, not automatic promotion from a score or a candidate label. No automatic writes to existing Skills, new aesthetic production gates, paid calls or changes to Runtime/provider/compiler/API/schema/configuration.

Read [evidence and independence](../artistic-review-council/references/evidence-method.md) before judging media. Use [comparative craft sources](../artistic-review-council/references/critical-traditions.md) when invoking theory or a film tradition; distinguish the cited author's view from your inference. Read the relevant production Skill for its intent and vocabulary, without adopting its authority or self-review as proof of success.

Separate observed fact, interpretation and proposed revision. Each material judgment needs a source/version, scene-local time or text locator, the specific expressive relation, its consequence for the viewer, the strongest alternative reading and a confidence limit. Describe success as carefully as failure. Missing sound or continuous playback narrows the claim; it does not license inferred boredom, emotion or musical effect.

Keep criticism, candidates and comparison records within the requested review workspace; retain a creative-guidance revision only through the user's explicit route above. Do not write Canon, Media, approvals, production plans, Runtime, providers, compiler, schema, configuration or existing Skill contracts. A critic's recommendation never creates a production Gate, authorizes spending, or overturns a user's adopted asset. Technical findings go to human maintenance. Read-only tool declarations do not enforce a filesystem sandbox.

## Existing source reads

Use supplied originals first. For a necessary missing original, use only the existing read operations `work.get_work`, `script.get_script`, `episode.get_episode`, `scene.get_scene`, `shot.get_shot`, `asset.get_asset`, `media.get_media` or `context.build_context`. Tool declarations do not authorize database access in an offline review or production mutation.