System Architecture | Feature Catalog | TEC Summary | KPI Standards Overlay

DeepSigma

DeepSigma prevents decision amnesia in AI systems.

Log every agent decision. Detect when it drifts. Prove what happened.

Quickstart

pip install deepsigma

# Log an agent decision
coherence agent log decision.json

# Audit all logged decisions
coherence agent audit --json

# Coherence score
coherence agent score

60-Second Proof

coherence demo

BASELINE   90.00 (A)
DRIFT      85.75 (B)   red=1
PATCH      90.00 (A)   patch=RETCON  drift_resolved=true

Three states, deterministic every run:

1. BASELINE — sealed episode, coherence scored
2. DRIFT — data changed, drift detected automatically
3. PATCH — governed retcon applied, coherence restored

Machine-readable: coherence demo --json

What It Does

DeepSigma is the institutional memory layer that makes AI decisions reconstructable.

Every agent decision becomes a sealed, hash-chained episode. Drift between decisions is detected automatically across 8 types. Authority is captured cryptographically, not implied. The full "why" is retrievable in under 60 seconds.

In practice:

• the "why" is retrievable (not tribal)
• authority is explicit (not implied)
• changes are patched, not overwritten
• drift is detected early and corrected consistently

Editions

One product line, one version, two editions:

• CORE edition: minimal, demo-first, deterministic (pip install deepsigma)
• ENTERPRISE edition: extended adapters, dashboards, and ops surfaces (repo-native under enterprise/)

Edition boundary ledger: EDITION_DIFF.md

Operating Modes

Core Mode

Use Core mode when you need fast adoption and low cognitive load.

Active Core surface at repo root:

• run_money_demo.sh
• src/core/
• docs/examples/demo-stack/
• tests/test_money_demo.py

Enterprise Mode

Use Enterprise mode when you need connectors, dashboards, extended security, broader telemetry, and integration-heavy workflows.

Dependency note:

• pip install "deepsigma[enterprise]" installs enterprise runtime extras used by telemetry/radar tooling.
• Full enterprise code surfaces are repository-native under enterprise/ and are run from source in this repo.

Enterprise surfaces are first-class under:

• enterprise/README.md

Examples of parked modules:

• enterprise/dashboard/
• enterprise/docker/
• enterprise/release_kpis/
• enterprise/schemas/
• enterprise/scripts/
• enterprise/src/ (non-core packages)
• enterprise/docs/ (full enterprise docs)

Run the enterprise wedge:

make enterprise-demo
make test-enterprise

Primitive Loop: CERPA

Claim -> Event -> Review -> Patch -> Apply

CERPA is the foundational adaptation loop for the platform. Every governance flow — across IntelOps, ReOps, FranOps, AuthorityOps, and ActionOps — follows this cycle:

1. Claim — an asserted truth or commitment
2. Event — an observable occurrence
3. Review — evaluate the claim against the event
4. Patch — corrective action if drift is detected
5. Apply — execute the patch and update state

python -m src.core.examples.cerpa_contract_demo
python -m src.core.examples.cerpa_agent_supervision_demo

Model Exchange Engine

Deep Sigma is the reactor, boundary, and memory system. Models are interchangeable cognitive thrusters. Models produce exhaust. Deep Sigma produces judgment.

The Model Exchange Engine (MEE) standardises how external or local models plug into Deep Sigma. Adapters can draft, reason, summarise, and disagree — but they cannot commit canon, approve themselves, or bypass AuthorityOps.

python -m core.cli mee demo          # Run all adapters
python -m core.cli mee health        # Check adapter health

Available adapters: APEX (Cognis-APEX-3.2), Mock, OpenAI, Claude, GGUF.

See docs/architecture/model_exchange_engine.md for full details.

Release Artifacts

Build both edition artifacts from one version line:

make release-artifacts

Outputs in dist/:

• deepsigma-core-vX.Y.Z.zip
• deepsigma-enterprise-vX.Y.Z.zip

Full Platform References

For the full-platform docs and architecture map, use parked docs directly:

• enterprise/docs/positioning/positioning_manifesto.md
• enterprise/docs/positioning/executive_briefing_one_page.md
• enterprise/docs/release/
• enterprise/docs/security/
• enterprise/docs/mermaid/

Repo Snapshot (auto-generated 2026-03-07 17:40 UTC)

• 1,799 files | 331,813 lines of code
• 41 CI workflows | 178 test files | 5 pyproject.toml
• 21 EDGE modules

LOC by extension: .py 116,372 .html 59,838 .json 57,437 .md 49,252 .svg 28,054 .jsonl 5,278 .tsx 2,942 .ttl 2,804 .patch 1,438 .ts 1,244

Repo Intent

• Keep root focused on a reliable first proof.
• Keep enterprise depth available without deleting capability.
• Expand from Core into Enterprise intentionally, not by drift.

License

MIT