Kronos Capital OS — Full Autonomous Quantitative Institution

Version: 1.1.0 — final audited release
Upstream Kronos: shiyu-coder/Kronos
Pinned upstream commit: 67b630e67f6a18c9e9be918d9b4337c960db1e9a

Kronos Capital OS (KCOS) is the full package developed across the project chain: a persistent, context-aware, cross-asset autonomous quantitative operating system that can observe markets continuously, maintain institutional memory, map cross-asset relationships, generate hypotheses and strategies, validate them through fixed research gates, allocate capital, pass all trades through an independent deterministic CRO/risk kernel, execute through configured venues, reconcile state, attribute outcomes, detect degradation, and keep learning.

It is not only Genesis. Genesis is merely the onboarding/birth sequence.

System loop

STREAM MARKET + MACRO + FUNDAMENTAL EVENTS
                ↓
        REAL-TIME HOT STATE
                ↓
      SIX-SECOND WORLD SYNC
                ↓
   TREND / REGIME / MARKET GRAPH
                ↓
      CONTEXT + MEMORY RETRIEVAL
                ↓
 KRONOS + FACTOR + VOLATILITY MODELS
                ↓
         ALPHA ENSEMBLE
                ↓
     CURIOSITY / HYPOTHESES
                ↓
     STRATEGY FACTORY + TESTS
                ↓
        ALPHA MARKETPLACE
                ↓
         AUTONOMOUS CIO
                ↓
      PORTFOLIO CONSTRUCTION
                ↓
    INDEPENDENT AUTONOMOUS CRO
                ↓
        DETERMINISTIC RISK
                ↓
        EXECUTION ROUTER
                ↓
      RECONCILIATION / FILLS
                ↓
       P&L ATTRIBUTION / DRIFT
                ↓
     MEMORY + MODEL GOVERNANCE
                ↓
              REPEAT

What is included

1. Genesis / owner onboarding + simple GUI

The primary setup path is now a browser-based Genesis Control Center. It captures jurisdiction, deployable capital, broker/exchange credentials, market-data credentials, macro-data credentials, hard risk limits, and whether validated strategies may graduate automatically to live capital. Secrets are encrypted in a persistent local vault and never returned to the browser. A CLI Genesis flow remains available as a fallback.

2. Six-second real-time brain

• event-driven feed ingestion
• global heartbeat every 6 seconds or less
• market-state versioning
• connector watchdogs
• stale-state protection
• latency classification
• Prometheus metrics
• emergency-stop API

3. Context and institutional memory

• Redis hot state
• in-process hot/warm memory
• PostgreSQL durable institutional memory
• bounded context packets instead of loading full history into a reasoner
• relevant market-graph neighbor retrieval
• decision and audit history

4. Cross-asset market map

• rolling return graph
• correlation edges
• lead/lag estimates
• event memory
• multi-horizon trend state
• cross-asset signal component

5. Model plane

• pinned upstream Kronos adapter
• asynchronous Kronos inference service
• factor model
• EWMA/realized volatility
• regime classifier
• forecast calibration tracker
• multi-model forecast ensemble

6. Curiosity / self-improvement

• surprise detection
• expected-information-value research priority
• hypothesis plus counter-hypothesis generation
• automatic candidate-strategy creation
• persistent experiment/strategy lineage
• model drift controls

7. Research integrity and promotion

• leakage guard
• walk-forward split engine
• persistence/momentum/random baselines
• transaction-cost model
• Monte Carlo/bootstrap testing
• generic validation metrics
• strategy registry
• fixed promotion chain:

RESEARCH → WALK_FORWARD → PAPER → CANARY → LIVE → SCALED

• automatic demotion/retirement on decay, data-quality failure, execution mismatch, or risk breach

8. Alpha marketplace

Strategies compete for capital using risk-adjusted live/OOS performance, expectancy, stability, drawdown, and decay instead of receiving permanent allocation.

9. Portfolio construction

• signal-to-order-intent layer
• covariance shrinkage
• risk-adjusted optimizer
• asset-class/venue exposure aggregation
• concentration-based hedge planner
• options intelligence layer with implied volatility, Greeks, term/skew summaries and defined-risk option-expression proposals; live option contracts still require venue-specific contract IDs and validation

10. Independent deterministic risk

The AI/research layer cannot override:

• max risk per trade
• aggregate open risk
• daily/weekly loss breakers
• drawdown stop
• gross leverage ceiling
• single-asset notional ceiling
• venue exposure ceiling
• stale-state prohibition
• emergency stop

Tail-risk and stress-test modules are included separately from Kronos forecast uncertainty.

11. Cross-asset execution

Execution routing supports asset-class mapping to:

• Interactive Brokers
• Coinbase Advanced Trade
• OANDA
• paper execution

The router can map equities, ETFs, options, futures, rates, commodities and indices to IBKR; FX to OANDA/IBKR; crypto to Coinbase/IBKR where configured and supported by the account.

12. Reconciliation and execution quality

• persistent idempotent order manager
• strategy-level position/cost-basis ledger
• target-position execution (trade only the delta, never re-buy merely because the six-second signal is unchanged)
• duplicate-fill protection
• broker/order-status reconciliation
• position reconciliation
• slippage model
• venue health abstraction
• audit events for risk vetoes, orders and execution errors

13. Treasury

• deployable-capital calculation
• cash-buffer policy
• venue-cap policy
• no autonomous withdrawal authority by default

14. Learning layer

• P&L attribution
• postmortems
• drift detector
• model-governor allocation haircut/disable logic
• institutionalized lessons

15. 24/7 operations and recovery

• Docker Compose
• PostgreSQL + pgvector
• Redis persistence
• durable market-state/bar recovery after restart
• durable desired-state revisions + encrypted secret vault
• autonomous runtime process
• separate research worker so research cannot consume the six-second live-decision budget
• systemd boot service
• health, status, metrics and emergency-stop endpoints
• loopback-first GUI security plus operator token for non-local access

Connectors you should provide

Recommended starting configuration:

System	Purpose
Interactive Brokers	Broad multi-asset execution
Coinbase Advanced Trade	Crypto execution
Databento	Primary live + historical market data
FRED	Macro data
SEC EDGAR	Public filings/XBRL
OANDA	Optional dedicated FX execution/feed
PostgreSQL	Durable memory/audit/strategy lineage
Redis	Hot state/event cache
Vault/KMS/Secrets Manager	Production secret isolation
Linux host	Continuous 24/7 operation

See docs/CONNECTORS.md and docs/24_7_DEPLOYMENT.md.

Simple setup

On a Linux host with Docker Engine + Docker Compose:

make launch

Then open http://127.0.0.1:8080 and complete Genesis in the browser. The Docker build fetches and verifies the pinned upstream Kronos commit automatically; make bootstrap is only needed for a local non-Docker development checkout.

The GUI provides Overview, Setup, Capabilities, Strategies, Research and Audit views. KCOS continuously compares desired state with actual connector/runtime state and reports repair actions instead of relying on agent/chat memory.

For a headless/CLI setup instead:

python -m venv .venv
source .venv/bin/activate
pip install -e ".[dev,marketdata,coinbase]"
make genesis

After validating the host, enable reboot recovery:

sudo bash scripts/install_systemd.sh

The default Docker port is bound to loopback only. If you intentionally expose the dashboard through a non-loopback hostname or reverse proxy, KCOS requires the operator token printed by:

kcos admin-token

Safety constitution

The intelligence is allowed to become smarter, not less constrained.

Self-improvable under validation:

• strategies
• features
• model/ensemble weights
• market-graph knowledge
• research code
• execution algorithms

Not self-modifiable by the trading/research agents:

• raw credential permissions
• absolute risk ceilings
• validation-gate authority
• audit history
• emergency-stop semantics
• withdrawal/transfer authority

Reality of the package

KCOS is a complete operating-system architecture with runnable paper-mode infrastructure and live venue adapter boundaries. It does not ship with a pre-proven profitable strategy, and no claim is made that Kronos or any dynamically generated strategy has a durable edge. Live strategies only become eligible after the configured evidence gates are satisfied. Provider-specific authentication, KYC, market-data entitlements, instrument identifiers, and account permissions must still be valid.

Automated trading can lose capital. The purpose of the package is to make autonomous research/execution disciplined, observable, recoverable, and bounded by deterministic risk controls.

Final audit

The release includes scripts/release_audit.py, scripts/build_release.py, FILE_HASHES.json, and docs/FINAL_AUDIT.md. make audit verifies the declared release contract before packaging.