Shadow Analytics

Binding sources

• docs/FINAL_SYSTEM_VISION.md — Layer 6 learning/optimization diagnostics; strategy quality first-class; shadow-first in this repo unless explicitly extended.
• AGENTS.md — passive/diagnostic; no silent runtime mutation; L7 not bypassed.

Purpose

Use this skill to keep the shadow analytics layer coherent, operator-readable, and structurally rich.

Aligns with docs/FINAL_SYSTEM_VISION.md Layer 6 (learning and optimization) and reporting surfaces — diagnostic / shadow-first in this repo unless explicitly extended.

This layer exists to provide passive diagnostics from paper-trading data. It must never alter runtime behavior.

Use When

Use this skill when:

• editing app/learning/analytics.py shadow insight structures
• editing app/shadow_learning.py summary or persistence helpers
• extending shadow diagnostic dimensions (stability, calibration, regime transitions, strategy matrix)
• surfacing shadow analytics in the operator panel
• adding new insight persistence or trend tracking structures
• extending AI advisory comparison signals
• improving confidence calibration diagnostics

Do Not Use When

Do not use this skill for:

• changing strategy scoring or thresholds
• changing risk policy
• changing execution behavior
• auto-enabling learning or AI modes
• adding adaptive runtime behavior

Safety Contract

• Shadow analytics remain descriptive and advisory only.
• Strategy×regime matrix is an observation tool, not a strategy selector.
• Regime transitions describe observed market behavior, not predictions.
• AI comparison signals describe alignment, not AI authority.
• Confidence calibration is a review aid, not an automatic tuner.
• Insight persistence is bounded (max 200 entries) and local-only.
• No shadow output may trigger trades, change risk, or alter runtime modes.

Primary Workflow

1. Read AGENTS.md.
2. Inspect current shadow insight structures.
3. Extend descriptive dimensions without changing runtime paths.
4. Verify the new insight does not alter execution behavior.
5. Add bilingual labels (EN/TR) for any new surface.

Required Checks

• New diagnostics remain purely heuristic and descriptive.
• Strategy×regime cells derive from realized closed records only.
• Calibration bands use paired outcome data only.
• AI comparison collects only when AI_MODE != OFF.
• Persistence helpers are bounded and fail safely.
• Labels exist in both EN and TR.

Expected Output

• Shadow Analytics: what diagnostic dimension changed or was added.
• Operator Surface: how the new data appears in the panel.
• Safety Check: explicit note that execution behavior was not changed.