vsf-rsi — Recursive Self-Improvement for VSM systems

Don't ask the language model to improve itself. Give the improvement a substrate.

vsf-rsi is a cybernetic feedback loop that observes, evaluates, and improves socratic-engine predicates and trees through four levels of recursive self-improvement (RSI). It uses a Genetic Algorithm (GA) to evolve predicate structures through feature construction and decision trees.

The system is deliberately bounded. It does not try to make the AI "smarter" unboundedly. It gives the system a formal structure in which improvements can be proposed, executed, validated, and rolled back — all within VSM safety constraints.

Status: v0.2.13 — Adaptive weights, gate registry, auto-generated code. 858 tests. RSI pipeline-generated predicates gitignored; seeds tracked, learned state backed up externally.

Installation

pip install vsf-rsi

With socratic-engine (required dependency):

pip install vsf-rsi socratic-engine

From source:

git clone https://github.com/rm-w3kufe/vsf-rsi.git
cd vsf-rsi
pip install -e ".[dev]"

The bet

LLMs are excellent at proposing solutions. They are less reliable when they must implicitly improve their own evaluation criteria while simultaneously maintaining safety guarantees.

The bet here is simple:

Let the model propose improvements. Let a deterministic substrate validate and execute them. The model does not certify its own improvements.

flowchart LR
    A["LLM<br/>propose improvement"] --> B["RSI Observer<br/>capture + discriminate"]
    B --> C["Resolution<br/>drift / inject / generate / evolve"]
    C --> D["SocraticEngine<br/>re-evaluate"]
    D --> E["Metrics<br/>did it improve?"]
    E --> A

This creates a division of labour:

The central boundary is:

The LLM proposes. The substrate validates. The LLM does not self-certify.

What it is

At its core, vsf-rsi provides:

Core:

• Observer (rsi_observer.py) — wraps SocraticEngine.evaluate() to capture every evaluation event, discriminate errors, and resolve them through L1-L4. Extracted helper methods for error handling, fallback events, and resolution. Circular buffer (500 actions) for memory safety.
• Metrics (rsi_metrics.py) — tracks classification history, accuracy, latency, and threshold adjustments
• Bridge (rsi_bridge.py) — integrates with state-canon-mcp for canonical state queries, metrics feedback, and rule enforcement
• Socratic Bridge (rsi_socratic_bridge.py) — registers RSI predicates and trees in socratic-engine
• Pipeline (rsi_pipeline.py) — full evolution cycle orchestrator: load predicates → evaluate → learn → register. Includes input sanitization, structured condition trees (no exec()), and global engine singleton.

L3 Autonomous Cycle:

• Autonomous L3 (rsi_autonomous_l3.py) — orchestrator: detect → generate strategies → shadow validate → activate → monitor. Uses genome V3 for strategy generation with comparison predicates (gt/lt/eq) and contradiction detection.
• Fault Detector (rsi_fault_detector.py) — detects complex faults (≥3 BLOCKING errors in 10 evals) with fault signatures, sliding windows, and error ratios
• Shadow Mode (rsi_shadow_mode.py) — validates strategies with 10 real evaluations before activation. Tracks baseline vs candidate accuracy.
• Rollback Manager (rsi_rollback.py) — monitors activated strategies, auto-reverts if accuracy degrades below baseline

Learning:

• Scenario Memory (scenario_memory.py) — procedural learning: records failures with correction paths, matches novel faults to prior corrections
• Scenario Bridge (rsi_scenario_bridge.py) — connects failures → gaps → corrections across modules
• Feedback Loop (rsi_feedback_loop.py) — manages threshold adjustments with drift bounds and step limits
• Gap Detector (rsi_gap_detector.py) — detects evaluation gaps: stale predicates, missing branches, low accuracy
• Pattern Detector (rsi_pattern_detector.py) — identifies recurring error patterns with time-based decay (10%/day)

Generation:

• Predicate Generator (rsi_predicate_generator.py) — creates new predicates from error patterns (with AST validation)
• Tree Generator (rsi_tree_generator.py) — generates evaluation trees from gap analysis
• Advanced Tree Generator (rsi_advanced_tree_generator.py) — threshold-optimized and coverage trees
• Forest Generator (rsi_forest_generator.py) — generates populations of trees for genetic evolution
• Tree Registry (rsi_tree_registry.py) — manages predicate versions, prevents conflicts, tracks lineage

Genetic Algorithm:

• Genome V2 (rsi_genome_v2.py) — base genome with dictionary dispatch for operator application
• Genome V3 (rsi_genome_v3.py) — enriched genome with feature construction, variadic ops, chaining, sign normalization, parity detection, XOR-2, and Tabu memory
• Genetic Algorithm v1 (rsi_genetic_algorithm.py) — evolves predicate populations with tournament selection, crossover, mutation, and convergence detection
• Genetic Algorithm v2 (rsi_genetic_algorithm_v2.py) — improved GA with adaptive mutation rates
• Adversarial Harness (rsi_adversarial_harness.py) — connects GA to adversarial benchmarks for fitness evaluation
• Adversarial Harness v2 (rsi_adversarial_harness_v2.py) — improved adversarial fitness with raw feature extraction

Benchmarking:

• Benchmark (rsi_benchmark.py) — load scenarios, run benchmarks, save/load reports, compute improvement curves
• Adversarial Scenarios (rsi_adversarial.py) — 4 scenario generators: Prisoner's Dilemma, Parábola Silenciosa, XOR 5D, Señal en Ruido
• Stress Tests (rsi_stress_test.py) — 32 tests across 7 dimensions to find breaking points

Infrastructure:

• Component Registry (rsi_component_registry.py) — maps package components to their roles and dependencies
• Manifest Parser (rsi_manifest_parser.py) — parses RSI manifest files for tree/predicate registration
• Adaptive Weights (adaptive_weights.py) — dynamic weight adjustment based on performance metrics
• Gate Registry (gate_registry.py) — central registry for RSI gates and validation rules
• Demo (rsi_demo.py) — runnable demo: generate → evaluate → evolve → measurable improvement
• Error Recovery — system continues with degraded functionality on failures (never crashes)
• Logging — structured logging via logging.getLogger("vsf_rsi.observer")

The four levels of RSI

flowchart TD
    subgraph L1["L1: Parameter Drift"]
        A1["adjust thresholds<br/>(autonomous)"]
    end
    
    subgraph L2["L2: Capability Extension"]
        A2["inject predicate wrappers<br/>(autonomous)"]
    end
    
    subgraph L3["L3: Autonomous Strategy Cycle"]
        D["fault detector<br/>(≥3 BLOCKING in 10 evals)"]
        G["genome V3<br/>(generate strategies)"]
        S["shadow mode<br/>(10 real evals)"]
        R["rollback manager<br/>(auto-revert if degrades)"]
        D --> G --> S --> R
    end
    
    subgraph L4["L4: Genetic Evolution"]
        A4["evolve predicate populations<br/>(human approval)"]
    end
    
    L1 -->|"if drift fails"| L2
    L2 -->|"if pattern detected"| D
    R -->|"if enough data"| L4

L3 generational loop (v0.2.16+)

Each cycle runs up to MAX_GENERATIONS (3) × STRATEGIES_PER_FAULT (5): gen-0 is random; later generations breed from shadow-ranked winners (elitism + genome-level crossover_v3/mutate_v3), early-exit on first passing generation. Winners persist to scenario memory with their full tree, and the next cycle re-seeds from recorded winners (cross-cycle retention) — gold found in cycle N survives to N+1. Every candidate's tree is recoverable from the strategy registry for audit.

Genomes convert to socratic computation nodes (zero information loss: 15-op feature chains with d0 → d1 → d2 chaining + native decision tree), evaluated by the engine without translation.

Measured (benchmarks battery, benchmarks/ in vOSlab): ~50–70% activation on BLOCKING faults (was ~10% random-shot), slope ≥ 0 across cycles, holdout generalization without regression, 3/3 ablation wins vs equal-budget random search.

state-canon-mcp integration

The rsi_bridge module provides functions to integrate with the state canon:

from vsf_rsi.rsi_bridge import get_rsi_rules, query_canon, feed_metrics_to_canon

# Get rules that constrain RSI behavior
rules = get_rsi_rules()

# Query canonical state
result = query_canon("rsi_metrics", {"predicate": "stasis_check"})

# Feed metrics back to state canon
feed_metrics_to_canon({"total_evaluations": 100, "error_rate": 0.05})

CLI usage

Each module has its own CLI entry point:

# Feedback loop — adjust thresholds
python -m vsf_rsi.rsi_feedback_loop adjust
python -m vsf_rsi.rsi_feedback_loop status
python -m vsf_rsi.rsi_feedback_loop history

# Pattern detector — analyze error patterns
python -m vsf_rsi.rsi_pattern_detector detect
python -m vsf_rsi.rsi_pattern_detector detect --predicate stasis_check
python -m vsf_rsi.rsi_pattern_detector summary

# Gap detector — find evaluation gaps
python -m vsf_rsi.rsi_gap_detector detect
python -m vsf_rsi.rsi_gap_detector gaps

# Tree generator — create trees from gaps
python -m vsf_rsi.rsi_tree_generator generate
python -m vsf_rsi.rsi_tree_generator list

# Advanced tree generator — threshold-optimized and coverage trees
python -m vsf_rsi.rsi_advanced_tree_generator generate

# Forest generator — create populations for GA
python -m vsf_rsi.rsi_forest_generator generate --predicate stasis_check
python -m vsf_rsi.rsi_forest_generator evolve --predicate stasis_check
python -m vsf_rsi.rsi_forest_generator best --predicate stasis_check
python -m vsf_rsi.rsi_forest_generator list

# Pipeline — full evolution cycle
python -m vsf_rsi.rsi_pipeline status

# Autonomous L3 — detect → generate → shadow → activate
python -m vsf_rsi.rsi_autonomous_l3 status

# Fault detector — detect complex faults
python -m vsf_rsi.rsi_fault_detector status

# Shadow mode — validate strategies
python -m vsf_rsi.rsi_shadow_mode status

# Rollback manager — monitor activated strategies
python -m vsf_rsi.rsi_rollback status

# Component registry — manage components
python -m vsf_rsi.rsi_component_registry list

# Demo — full RSI cycle
python -m vsf_rsi.rsi_demo

Demo

Run the complete end-to-end demo:

python -m vsf_rsi.rsi_demo

The demo demonstrates the full RSI cycle:

1. Creates an SocraticEngine with built-in predicates
2. Runs 10 evaluations (mix of passes and failures)
3. Records failures in scenario_memory
4. Matches novel faults to prior corrections
5. Applies scenario_memory corrections to improve results
6. Generates patterns from error history
7. Evolves predicate populations with the genetic algorithm
8. Reports measurable improvement

Testing

# Run full suite (858 tests, 99% coverage)
pytest tests/ -v

# Run with coverage
pytest tests/ --cov=vsf_rsi --cov-report=term-missing

# Run integration tests only (requires socratic-engine)
pytest tests/test_rsi_observer_integration.py -v

# Run scenario_memory integration test
pytest tests/test_scenario_memory_integration.py -v

Architecture

flowchart TD
    subgraph SocraticEngine["Socratic Engine"]
        SE["evaluate(tree, ctx)"]
        PD["predicates"]
    end
    
    subgraph RSIObserver["RSI Observer"]
        OB["wrapper: evaluate()"]
        BR["bridge: track_classification()"]
        DI["discriminate: BLOCKING / STRUCTURAL"]
        RE["resolve: L1 / L2 / L3 / L4"]
    end
    
    subgraph L3Cycle["L3 Autonomous Cycle"]
        FD["fault_detector<br/>detect complex faults"]
        G3["genome_v3<br/>generate strategies"]
        SM["shadow_mode<br/>10 real evals"]
        RB["rollback_manager<br/>auto-revert"]
        FD --> G3 --> SM --> RB
    end
    
    subgraph RSIModules["RSI Modules"]
        MET["rsi_metrics<br/>track + aggregate"]
        FL["rsi_feedback_loop<br/>adjust thresholds"]
        SC["scenario_memory<br/>match + record"]
        PG["rsi_predicate_generator<br/>create predicates"]
        GA["rsi_genetic_algorithm<br/>evolve populations"]
    end
    
    subgraph VSMConstraints["VSM Safety"]
        A0["A0: purpose inalienable"]
        A5["A5: autonomy ceiling"]
        R17["R17: destructive consent"]
    end
    
    SE -->|"capture"| OB
    OB -->|"feed"| BR
    BR -->|"store"| MET
    MET -->|"analyze"| DI
    DI -->|"classify"| RE
    RE -->|"L1"| FL
    RE -->|"L2"| OB
    RE -->|"L3"| FD
    RE -->|"L4"| GA
    FL -->|"adjust"| PD
    SC -->|"match"| RE
    SC -->|"record"| RE
    RB -->|"activate"| PD
    A0 -.->|"constrain"| RE
    A5 -.->|"constrain"| RE
    R17 -.->|"constrain"| RE

Quick start

from vsf_rsi import RSIObserver, RSIMode
from socratic_engine import SocraticEngine

# Create engine with predicates
engine = SocraticEngine()

@engine.register("my_predicate")
def my_predicate(ctx, threshold=0.5, **kw):
    val = ctx.get("value", 0.0)
    from socratic_engine.engine import PredicateResult, Truth
    return PredicateResult(
        truth=Truth.TRUE if val > threshold else Truth.FALSE,
        certified=True,
        evidence={"value": val},
        source="my_predicate",
    )

# Create observer
observer = RSIObserver(engine, mode=RSIMode.CAPABILITY.value)

# Evaluate a tree
tree = {"predicate": "my_predicate", "args": ["$ctx", 0.5]}
ctx = {"task_id": "example", "value": 0.3, "input_value": 0.3}
event = observer.evaluate(tree, ctx)

# Check what happened
print(f"Error: {event.is_error}")
print(f"Class: {event.error_class}")

# Run pattern analysis
patterns = observer.analyze_patterns()
print(f"Patterns: {len(patterns)}")

# Generate predicates from patterns
actions = observer.generate_predicates(min_occurrences=3)
print(f"Generated: {len(actions)}")

Integration with socratic-engine

vsf-rsi depends on socratic-engine as its recursive evaluation substrate. The integration is via the observer wrapper:

sequenceDiagram
    participant User
    participant Observer
    participant Engine
    participant Metrics
    participant L3Cycle as L3 Autonomous Cycle
    
    User->>Observer: evaluate(tree, ctx)
    Observer->>Engine: original evaluate()
    Engine-->>Observer: result
    Observer->>Metrics: track_classification(event)
    Observer->>Observer: discriminate(error)
    
    alt error detected
        Observer->>Observer: resolve_error(event, ctx)
        alt L1: parameter drift
            Observer->>Observer: adjust threshold
        else L2: capability extension
            Observer->>Observer: inject wrapper predicate
        else L3: autonomous cycle
            Observer->>L3Cycle: run_cycle(fault)
            L3Cycle->>L3Cycle: detect faults (≥3 BLOCKING)
            L3Cycle->>L3Cycle: generate strategies (genome V3)
            L3Cycle->>L3Cycle: shadow validate (10 evals)
            L3Cycle->>L3Cycle: activate + monitor
            L3Cycle-->>Observer: L3CycleResult
        end
    end
    
    Observer-->>User: event + actions

Integration with agent systems

vsf-rsi is designed to be used by any AI agent or human operator. Here's how to integrate it:

Available tools

Integration by platform

OpenCode

Add to your opencode.json:

{
  "mcp": {
    "socratic-engine": {
      "type": "local",
      "command": ["python3", "/path/to/socratic_mcp_with_bridge.py"]
    },
    "state-canon": {
      "type": "local",
      "command": ["python3", "/path/to/mcp_server.py", "--instance", "/path/to/your_state.py"]
    }
  }
}

Then use in your agent:

# Agent calls these via MCP tools
from socratic_engine import SocraticEngine
from vsf_rsi.scenario_memory import record

# Evaluate a decision
engine = SocraticEngine()
result = engine.evaluate(your_tree, your_context)

# Record what happened
record({
    "fault_signature": "what_went_wrong",
    "decision": "what_you_did",
    "outcome": "success_or_failure"
})

Claude Code

Add to your CLAUDE.md:

## Tools available

- `socratic_engine.evaluate(tree, context)` — evaluate decisions
- `state_canon.query(domain, filter)` — get ground truth
- `vsf_rsi.record(pattern)` — learn from outcomes

Perplexity / ChatGPT / Other agents

Install the packages:

pip install socratic-engine vsf-rsi
git clone https://github.com/rm-w3kufe/state-canon-mcp.git

Use in your code:

from socratic_engine import SocraticEngine
from vsf_rsi.scenario_memory import record

# Evaluate
engine = SocraticEngine()
result = engine.evaluate(tree, context)

# Learn
record({"fault": "error", "fix": "solution", "outcome": "success"})

Human operator

# Install
pip install socratic-engine vsf-rsi

# Use CLI
python -m socratic_engine evaluate tree.json --context ctx.json
python -m vsf_rsi.scenario_memory record --fault "error" --fix "solution"

For the full three-tool integration pattern, see state-canon-mcp README.

Safety model

The VSM safety constraints are enforced at every level:

flowchart TD
    subgraph Autonomous["Autonomous (no approval)"]
        L1["L1: parameter_drift<br/>adjust thresholds"]
        L2["L2: inject_predicate<br/>create wrappers"]
        L3["L3: autonomous cycle<br/>detect → generate → shadow → activate"]
    end
    
    subgraph HumanApproval["Human Approval Required"]
        L4["L4: evolve_forest<br/>genetic evolution"]
    end
    
    subgraph NeverAutonomous["Never Autonomous"]
        A0["A0: purpose changes"]
        A5["A5: autonomy ceiling"]
        R17["R17: destructive actions"]
    end
    
    L1 -->|"validate before keeping"| L1
    L2 -->|"validate before keeping"| L2
    L3 -->|"shadow mode + auto-rollback"| L3
    L4 -->|"require approval"| L4

Key safety properties:

• A0/A0.1: Purpose is inalienable — RSI cannot change system purpose
• A5: Autonomy ceiling — L1-L3 are autonomous, L4 requires human approval
• R17: Destructive actions always require explicit human consent
• R4: Trust requires independent check — RSI cannot self-certify
• L3 safety: Shadow mode validates with 10 real evaluations before activation. Rollback manager auto-reverts if accuracy degrades below baseline.

Going deeper

• ROADMAP.md — version history and future plans.
• CHANGELOG.md — what changed, by version.
• docs/rsi_forests.vsm — generated forest populations (VSM notation).
• docs/rsi_generated_predicates.vsm — generated predicates (VSM notation).
• docs/rsi_generated_trees.vsm — generated trees (VSM notation).
• tests/ — 858 tests.

Roadmap

See ROADMAP.md for detailed version history and future plans.

v0.2.5 — Adversarial Benchmarks ✅

• Benchmark framework: load scenarios, run benchmarks, save/load reports, improvement curves
• 4 adversarial scenario generators (prisoner, parábola, XOR 5D, señal+ruido)
• validate_predicate_behavior() — run predicates against test cases, reject below threshold
• Context-aware predicate generation (uses avg_threshold, avg_input, error_class)
• Feature-detect enforce_limits via inspect.signature (backward-compatible)
• CI fix: install socratic-engine before running tests
• 775 tests

v0.2.7 — Genoma-V3: Genetic Algorithm ✅

• Genetic Algorithm with enriched genome representation
• Feature construction graph with variadic operations and chaining
• Sign normalization, parity detection, XOR-2 operations
• Tabu memory to avoid repeated failed feature combinations
• Checkerboard XOR solved at 100% (autonomous discovery of sign + parity)
• 5D XOR at 80% test accuracy
• Unit circle at 100%
• Stress test suite: 32 tests across 7 dimensions
• 785 tests

v0.2.10 — L3 Fixes & Comprehensive Cleanup ✅

• Fix _rsi_trees evaluation: register trees as callable predicates via engine.register() closures
• Add global _engine singleton to avoid re-loading predicates on every bridge call
• Fix _evaluate_with_predicate to check engine.predicates (not _rsi_trees)
• Remove dead code: Union import, inject_context parameter
• Reduce RSIObserver.evaluate complexity D→C via extracted helper methods
• Reduce RSIGenomeV2._apply_op complexity D→C via dictionary dispatch
• Add MAX_ACTIONS=500 circular buffer for action tracking in observer
• Fix DEBT-001: correct genome-to-tree conversion with comparison predicates (gt/lt/eq)
• Fix contradictory condition detection in L3 genome_to_tree
• Fix DEBT-002: update debt_verification_results.json passed=true
• 855 tests

v0.2.11 — L3 Activation Loop Closure ✅

• _feed_rollback_evaluations() in RSIObserver — closes activate → monitor → record_eval → confirm/rollback
• RollbackManager.record_evaluation() now called after every evaluation
• Matching: source/fault_id/strategy_id triple check
• 3 new tests in TestRollbackLoop class
• 858 tests

v0.3.0 — Production

• Dashboard for observation
• Cross-validation
• Overfitting detection
• Kernel integration for high-dimensional vector operations

Lineage & philosophy

Built on Stafford Beer's Viable System Model and the Cybersyn project (Chile, 1971) — a system is viable when it can be described, governed, and audited. "Don't trust, verify" isn't a slogan here; it's the reconciler and the verify-at-every-boundary discipline made mechanical. Community-first, and deliberately from the Global South — the heir to Cybersyn's bet that good cybernetics serves people.

License

Code: Apache-2.0 (LICENSE) · Docs: CC-BY-4.0.

This module is deliberately more permissive than the AGPL core it was extracted from — it is meant to be adopted, embedded, and improved by the community. Improvements can flow back; the module stays clean of AGPL code by construction.