What does this skill do?

The Recursive Decision Log Skill provides a rigorous framework for documenting recursive decision-making processes, high-dimensional searches, and stochastic optimization. It preserves the usefulness of repeated cycles—previous memory, fresh information, explicit markers—while eliminating the risk of assuming that repetition guarantees certainty. Each iteration is recorded in an auditable format in JSONL or Markdown.

Optimization Rollouts
Perform multiple iterations to force deeper computation or explore local optima with full traceability.
Comparison of Ensembles
Compare model ensembles in stochastic optimization processes and record deviations from the previous winner.
Allocation of Capital
Document capital allocation decisions, production deployments, or migrations with strict traceability.
Promotional Doors
Verify security gates—repetition, freshness, metrics—before authorizing destructive or live actions.

Usage examples

🔄 Rollout Iteration
Run rollout 15 with 50 trials, compare the top candidate with the previous winner, Model_A, and make the decision.
📊 Compare assemblies
Compare the new ensemble with the previous winner and verify the consistency gates before allowing promotion.
🚦 Validate promotion for live
Verify whether the current candidate can be promoted to production: check for repetition, data freshness, and metrics.
📋 Iteration Summary
Generate a summary of the rollout, indicating the status, whether the previous winner remains in the lead, and what the next hurdles are.

Features

Explicit decision markers Label each result as "accept," "monitor," "reject," "expired," or "requires retesting."
Security Doors Verify repetition, freshness, and metrics before authorizing destructive or live actions.
Records in JSONL and Markdown Saves artifacts in an auditable JSONL format for machine-readable logging and in Markdown for human readers.
Degradation Detection Automatically demote candidates if there is a deviation, tail risk, or outdated data compared to the previous winner.
Signs of Coherence Check whether the assembly, recursion, and last rollout match the previous winner before allowing promotion.

Frequently asked questions

The skill generates artifacts in JSONL for machine-readable logging and in Markdown for human reading. Both include the rollout_id, timestamp, candidates, and decision labels.
These are checks for duplication, data freshness, and metrics that must be passed before any destructive action or promotion to a live environment is authorized.
Yes. The skill is designed for capital allocation, deployments, and migrations where strict traceability is required before a decision is rolled out to production.
Each iteration explicitly records the candidate's status and downgrades the results if there is a deviation or outdated data, avoiding the assumption that repetition guarantees certainty.

Related skills

More skills in the same category:

Recursive Decision Log — Optimization Traceability with Claude AI

¿Prefieres escuchar el contenido? Genera la narración de audio con un clic.