# Autonomous Allocation Agent ### 1. Mandate The Autonomous Allocation Agent (AAA) is the quantitative analytical intelligence layer of the **Sagitta Protocol**. Its mandate is to evaluate allocation opportunities, model risk and performance outcomes, and produce structured, explainable intelligence for capital routing decisions—while operating strictly within system-defined doctrine, constraints, and survivability guarantees. The Autonomous Allocation Agent **informs decisions**.\ The Treasury **commits capital**. AAA exists to ensure that all risk-bearing decisions are taken consciously, evaluated rigorously, and justified relative to protection mechanisms. *** ### 2. Position Within the Sagitta System The Autonomous Allocation Agent operates as a **non-custodial, non-executing analytical office** within the Sagitta system architecture. System roles are explicitly separated: * The **Vault** records deposits and enforces accounting invariants * The **Treasury** forms liquidity, sizes batches, allocates capital, and settles outcomes * The **Reserve** provides insurance coverage and absorbs underperformance * The **Escrow** executes capital deployment under Treasury authority * The **Autonomous Allocation Agent** evaluates strategies and outcomes * The **Continuity Engine** governs behavior under failure and crisis states AAA exists **upstream of execution and downstream of policy**, translating policy-defined strategy space into actionable intelligence without authority to move capital. *** ### 3. Scope of Analysis The Autonomous Allocation Agent evaluates candidate allocation strategies across multiple analytical dimensions, including: * expected return distributions * volatility, drawdown, and tail risk characteristics * liquidity, duration, and exit sensitivity * counterparty, execution, and operational risk * correlation with Reserve and Treasury assets * historical, simulated, and regime-conditioned performance Analysis is performed across time horizons aligned with Treasury batch cadence and continuity posture. *** ### 4. Quantitative Foundation AAA is grounded in **institutional quantitative finance methodologies** employed by professional asset managers, risk desks, and portfolio construction teams. Its analytical core operationalizes established practices, including: * factor-based return attribution * volatility and drawdown modeling * correlation and covariance analysis * scenario and regime simulation * risk-adjusted performance metrics * capital efficiency and duration modeling The agent does not invent financial theory.\ It **scales** disciplined quantitative reasoning under constraint. *** ### 5. Human–Quant–Machine Alignment Sagitta treats quantitative intelligence as a **shared discipline**, not a replacement for fiduciary judgment. AAA operates in alignment with: * established quantitative finance principles * domain expertise from professional quants * Treasury-defined risk, solvency, and continuity doctrine Human oversight informs: * model selection and validation * constraint and guardrail design * evaluation criteria and exclusions * interpretation of edge cases Machine intelligence contributes: * scale and consistency * simulation depth across regimes * pattern recognition under bounded authority This alignment ensures that AAA remains **disciplined, explainable, and fiduciary-aligned**. *** ### 6. Determinism, Guardrails, and Authority AAA operates under **deterministic guardrails** defined by system doctrine. * No model—statistical or ML-based—may override Treasury constraints * No learning process may mutate invariants or guarantees * Authority is explicitly gated by allocator class Allocator capability progresses through authority levels (v1–v6), where **people qualify for authority** rather than purchasing features. Higher authority introduces broader decision space, not relaxed discipline. *** ### 7. Strategy Evaluation Framework AAA operates through a **scenario-driven evaluation framework**. For each candidate strategy, the agent: * simulates performance across varied market regimes * evaluates sensitivity to shocks and discontinuities * estimates impact on Treasury balance sheet and Reserve ratios * compares outcomes relative to passive Reserve performance Evaluation outputs are normalized into comparable metrics to support disciplined Treasury decision-making. *** ### 8. Reserve-Relative Intelligence Reserve-relative benchmarking is a **core analytical dimension** of AAA. All active strategies are evaluated against Reserve asset performance over equivalent intervals. Risk-taking must be justified relative to passive protection. Strategies that fail to demonstrate superior risk-adjusted outcomes relative to protection mechanisms are deprioritized by design. This aligns analytical incentives with fiduciary preservation rather than speculative yield chasing. *** ### 9. Interaction With the Treasury AAA produces **rankings, evaluations, and explanatory intelligence** for Treasury review. Treasury decisions incorporate: * agent evaluations and confidence bounds * Reserve coverage constraints * liquidity and duration requirements * system continuity posture AAA does not initiate allocations, adjust batch sizing, or execute transactions. Its role is strictly advisory. *** ### 10. Learning, Feedback, and Adaptation AAA updates its analytical models using **realized batch outcomes**. Settled performance feeds back into: * risk estimation models * correlation assumptions * scenario weighting * strategy ranking heuristics Adaptation improves analytical accuracy while remaining fully bounded by Treasury doctrine, Reserve discipline, and continuity constraints. Learning refines judgment—it does not expand authority. *** ### 11. Explainability and Auditability All AAA outputs are **quant-native, explainable, and auditable**. For each recommendation, the agent produces: * rationale summaries * contributing factors * comparative metrics * confidence and uncertainty assessments All outputs are logged, versioned, and reviewable by Treasury operators, auditors, and governance processes. *** ### 12. Independence From Execution and Custody AAA does not custody assets, hold keys, or interface directly with execution venues. This separation: * reduces attack surface * prevents intelligence from becoming execution authority * preserves fiduciary clarity Analytical intelligence is intentionally isolated from capital movement. *** ### 13. Continuity and Degradation Behavior Under continuity events, AAA adapts its analytical posture without altering capital guarantees. Possible modes include: * conservative strategy filtering * stress-prioritized evaluation * reduced recommendation bandwidth * full analysis suspension during emergency states AAA obeys system posture set by the Continuity Engine. *** ### 14. Standalone Deployment The Autonomous Allocation Agent is deployable as a **standalone analytical system**, independent of custody and execution. It may operate as: * a portfolio evaluation engine * a risk modeling service * an allocation intelligence layer for funds or protocols * a decision-support system for fiduciary capital managers Standalone deployment preserves analytical independence and institutional applicability. *** ### 15. Summary The Autonomous Allocation Agent strengthens allocation decisions through disciplined quantitative intelligence. It: * evaluates risk and performance under constraint * benchmarks strategies against protection * adapts through bounded feedback * produces explainable, auditable reasoning Sagitta treats intelligence as **advisory**, capital as **disciplined**, and protection as **invariant**. AAA exists to ensure that risk is taken consciously, justified continuously, and never confused with authority. ### Sandbox Mode vs Agent Mode The Autonomous Allocation Agent operates under two distinct analytical postures: **Sandbox Mode** and **Agent Mode**.\ These modes govern **how intelligence is produced, contextualized, and persisted**, not whether capital is deployed. Both modes remain strictly non-custodial and non-executing. *** #### Sandbox Mode (Exploratory Analysis) Sandbox Mode is the default analytical posture for users without production decision authority or when operating in exploratory contexts. In Sandbox Mode, AAA functions as a **static evaluation engine**: * Strategies are evaluated independently of live Treasury state * Results are generated from cached or simulated data * No persistent belief state is maintained across evaluations * No historical feedback loops influence subsequent outputs Sandbox Mode is designed for: * policy exploration * comparative strategy testing * education and familiarization * pre-qualification analysis Outputs are informative but **non-authoritative**. They do not accumulate context, adapt across time, or express longitudinal confidence. Sandbox Mode enables breadth without responsibility. *** #### Agent Mode (Persistent Decision Intelligence) Agent Mode is available only to users or systems with explicit **decision authority qualification**. In Agent Mode, AAA operates as a **persistent analytical agent** across discrete evaluation ticks aligned to Treasury cadence. Key characteristics: * Maintains a bounded internal belief state derived from prior evaluations * Incorporates realized outcomes from settled batches into future analysis * Updates regime context, correlation assumptions, and confidence weighting between ticks * Operates within fixed doctrine, constraints, and continuity posture Agent Mode does **not** imply autonomy over capital.\ It implies **continuity of reasoning**. The agent evolves its analytical perspective over time while remaining fully constrained by Treasury-defined policy and Reserve discipline. *** #### Tick-Based Evaluation and Human Intervention Agent Mode operates on scheduled evaluation ticks (e.g., weekly), corresponding to Treasury review and batch cycles. Between ticks: * Treasury operators may adjust portfolios, constraints, or policy parameters * These changes are incorporated as updated context at the next evaluation * The agent does not retroactively modify prior conclusions At each tick, AAA produces a fresh evaluation informed by: * current policy state * updated market data * accumulated analytical history * system continuity posture This preserves **human primacy in decision-making** while allowing analytical intelligence to compound responsibly over time. *** #### Mode Separation and Safeguards Sandbox Mode and Agent Mode are intentionally separated to prevent: * unqualified users from generating authoritative-looking outputs * accidental carryover of exploratory assumptions into production analysis * erosion of accountability boundaries Agent Mode is gated by authority, not convenience.\ Sandbox Mode is permissive by design. *** #### Summary Sandbox Mode enables exploration without consequence.\ Agent Mode enables continuity without autonomy. Together, they allow the Autonomous Allocation Agent to serve both as a **learning instrument** and a **fiduciary-grade de**