baseline-replication

---

LIBRARY-FIRST PROTOCOL (MANDATORY)

Before writing ANY code, you MUST check:

Step 1: Library Catalog

• Location: .claude/library/catalog.json
• If match >70%: REUSE or ADAPT

Step 2: Patterns Guide

• Location: .claude/docs/inventories/LIBRARY-PATTERNS-GUIDE.md
• If pattern exists: FOLLOW documented approach

Step 3: Existing Projects

• Location: D:\Projects\*
• If found: EXTRACT and adapt

Decision Matrix

Match	Action
Library >90%	REUSE directly
Library 70-90%	ADAPT minimally
Pattern exists	FOLLOW pattern
In project	EXTRACT
No match	BUILD (add to library after)

---

STANDARD OPERATING PROCEDURE

Purpose

• Reproduce baseline experiments faithfully before new development.
• Capture every constraint (data, metrics, hardware, randomness) and document variances.
• Produce artifacts that downstream skills can trust (configs, logs, checkpoints).

Trigger Conditions

• Positive: requests to replicate a paper’s baseline, re-run internal baselines, validate claims prior to extension.
• Negative: greenfield method design (use method-development), or pure prompt tuning (route to prompt-architect).

Guardrails

• Structure-first delivery: SKILL + README + examples + references; stash configs/resources alongside outputs.
• Constraint extraction in HARD / SOFT / INFERRED buckets with sources (paper, repo, maintainer notes).
• Two-pass refinement: (1) structure and environment parity; (2) epistemic validation and variance explanation.
• Confidence ceilings enforced; never overstate beyond observation/research limits.

Inputs

• Original specification (paper section, repo, config files).
• Target hardware/software stack and allowed variance.
• Success criteria (metric thresholds, tolerances, reproducibility bounds).

Workflow

1. Scope & Constraints: Capture dataset versions, metrics, seeds, hardware, and tolerances; confirm INFERRED assumptions.
2. Environment Parity Plan: Mirror dependencies; document any substitutions and expected impact.
3. Reproduce Runs: Execute baseline with fixed seeds; log configs, hashes, and environment fingerprints.
4. Validate: Compare against claimed results with statistical checks; explain deviations.
5. Package Artifacts: Store configs, logs, checkpoints, and a replication report; update examples/references as needed.

Validation & Quality Gates

• Environment parity documented; deviations justified.
• Metrics within agreed tolerance or deviation analysis provided.
• Replication report includes evidence, seeds, and reproducibility notes.
• Confidence line item included with appropriate ceiling (observation 0.95, research 0.85).

Response Template

**Constraints**
- HARD: ...
- SOFT: ...
- INFERRED (confirm): ...

**Plan**
- Environment + run strategy.

**Results**
- Metric(s): ... vs. claim ...
- Variance analysis: ...

**Artifacts**
- Configs/logs: <paths>
- Checkpoints: <paths>

Confidence: 0.82 (ceiling: research 0.85) - based on replicated runs and logged evidence.

Confidence: 0.82 (ceiling: research 0.85) - Assumes replication gates and evidence checks completed.