Context Mortality Audit (CMA)

Skill ID

context-mortality-audit

Version

1.0.0

Category

governance-triad / memory-governance

Protocol Alignment

• P1_IRP: INTEGRITY principle at infrastructure level
• P2_ANTIDOTE: Drift detection for compression systems
• P8_MUON: Dynamic session fidelity tracking

Codex Law Compliance

Principle	Implementation
CONSENT	Pre-compression audit surface; appeal mechanism
INVITATION	User-initiated recontext sweeps
INTEGRITY	Context death logging with full provenance
GROWTH	Drift trajectory analysis for systematic bias detection

Description

Long-horizon memory governance framework that audits, tracks, and
recovers context lost during compaction and synthesis cycles.

Implements five phases:

1. Context Death Audit (CDA) — Diff-based compaction logging
2. Path Tracer — Lifecycle tracing with fidelity scoring
3. Recontext Sweeps — Transcript recovery scanning
4. Appeal Mechanism — User-controlled reinstatement workflow
5. Drift Trajectory Analysis — Parallel-account bias detection

Sovereign Data Classification

All components respect the sovereignty hierarchy:

• HEALTH: NEVER auto-captured. Session-scoped. Explicit opt-in required.
• CREATIVE: Full provenance tracking. User's intellectual property.
• TECHNICAL: Persistent, subject to user review.
• PROCEDURAL: Persistent via memory_user_edits.
• RELATIONAL: Persistent, user-reviewable.
• FOUNDRY: Conservative. Health overlap treated as HEALTH.

Usage

CLI

# Show system statistics
python -m irp_swarm_console.context_mortality stats

# CDA operations
python -m irp_swarm_console.context_mortality cda graveyard
python -m irp_swarm_console.context_mortality cda snapshot --pre-file pre.json --post-file post.json

# Path Tracer
python -m irp_swarm_console.context_mortality pathtrace search --keyword "G2 manifold"
python -m irp_swarm_console.context_mortality pathtrace dead

# Recontext sweeps
python -m irp_swarm_console.context_mortality recontext run \
    --transcripts-file transcripts.json \
    --memory-file memory.json \
    --topic "IRP methodology" \
    --fidelity-threshold 0.5

# Appeals
python -m irp_swarm_console.context_mortality appeal stats
python -m irp_swarm_console.context_mortality appeal execute

# Drift analysis
python -m irp_swarm_console.context_mortality drift compare \
    --state-a state_a.json \
    --state-b state_b.json
python -m irp_swarm_console.context_mortality drift trend

Programmatic

from irp_swarm_console.context_mortality import (
    ContextDeathAudit,
    PathTracer,
    RecontextSweep,
    AppealManager,
    DriftTrajectoryAnalysis,
    MemoryState,
    SovereigntyClassifier,
)

# Phase 1: Context Death Audit
cda = ContextDeathAudit(session_id="abc123", user_id="joseph")
cda.capture_pre_state(pre_compaction_messages)
cda.capture_post_state(post_compaction_messages)
dropped = cda.generate_death_log()
cda.persist_log()

# Phase 2: Path Tracer
tracer = PathTracer()
trace = tracer.register_origin(
    session_id="abc123",
    turn_index=47,
    timestamp="2025-11-15T14:32:00Z",
    verbatim="The G2 manifold safety protocol uses Ricci flow..."
)
tracer.record_transformation(
    content_hash=trace.origin_content_hash,
    event="compaction",
    survived=True,
    transformed_form="User discussed G2 manifold safety protocol..."
)

# Phase 5: Drift Analysis
state_a = MemoryState(account_id="A", items=["memory item 1", "memory item 2"])
state_b = MemoryState(account_id="B", items=["memory item 1", "memory item 3"])
dta = DriftTrajectoryAnalysis(state_a, state_b)
report = dta.generate_drift_report()

Dependencies

• Python 3.8+
• Standard library only (json, hashlib, uuid, datetime, re, pathlib, dataclasses)

Storage

All data persists as structured JSON/JSONL in user-controlled filesystem locations.
CDA logs are NOT fed back into the model's context window (prevents recursive bloat).

Specification

See: Context Mortality Audit & Recontext Architecture v1.0 (February 8, 2026)

Author

Joseph (Pack3t C0nc3pts / StarwreckNTX)