governance-fundamentals

name	governance-fundamentals
description	Use when an agent needs to understand UNITARES governance concepts — EISV state vectors, basins, policy actions, coherence, calibration. Reference material for interpreting governance metrics as proprioceptive state estimation, not outcome judgment.
last_verified	2026-06-29
freshness_days	14
source_files	["unitares/config/governance_config.py","unitares/src/auto_ground_truth.py","unitares/src/governance_monitor.py","unitares/src/behavioral_state.py","unitares/src/behavioral_assessment.py","unitares/src/monitor_decision.py","unitares/src/mcp_handlers/core.py"]

Governance Fundamentals

What UNITARES Is

UNITARES provides digital proprioception for AI agents — awareness of your own state, your relationship to the system, and whether you are drifting. The live path is behavioral state estimation: observable work signals become EISV readings, smoothed over time and compared with the agent's own trajectory once a baseline exists. The thermodynamic / ODE model remains useful as a research lens and telemetry; do not present it as cold-start authority or live verdict authority.

EISV State Vector

Every agent has four dimensions, updated through check-ins:

Dimension	Range	Meaning
E (Energy)	[0, 1]	Productive capacity
I (Information Integrity)	[0, 1]	Claims matching results / calibration
S (Entropy)	[0, 1]	Drift / instability from normal (lower is usually steadier)
V (Valence)	[-1, 1]	EMA-smoothed E-I imbalance

How the Live Path Reads Them

• E (Energy) blends observable progress signals such as decision success, coherence, complexity calibration, and sometimes external task evidence.
• I (Integrity) tracks whether claims/confidence match observed results and whether coherence is holding.
• S (Entropy / drift) rises with drift norm, regime instability, and complexity divergence.
• V (Valence) is derived from the E-I imbalance. Positive means running hot (motion outruns integrity); negative means running careful (integrity outruns progress).

The headline math is proprioceptive residuals. In the live behavioral path, warmup verdicts fall back to a mostly server-derived cold-start prior (the behavioral track scores against fixed universal thresholds); after warmup, residual-like state comes from self-relative z-score deviation against the agent's own Welford baseline, with absolute safety floors and basin-health gates always in force.

Roadmap target semantics for richer cold-start grounding are:

measurement_t = EISV_t
reference_t   = blend(agent_baseline_t, class_anchor; w(grounding))
residual_t    = measurement_t - reference_t

Do not present the class-anchor blend as deployed unless live code exposes it. Deviation is information first, not a guilty verdict. Policy can map persistent or high-margin residuals to guide, pause, or reject, but EISV itself is measurement/diagnosis, not prosecution.

Prefer live tool output over static range lore if the current runtime reports a narrower or more precise bound.

Basins

Your state sits in a basin — a region of the EISV space:

• High basin: Healthy. E and I are high, S and V are low. Normal operating range.
• Low basin: Degraded. May need recovery or intervention.
• Boundary: Transitioning between basins. Extra attention from governance. Verdicts may carry margin: tight.

Use get_governance_metrics() as the source of truth for the current basin/mode labels rather than assuming they are constant across runtime versions.

When a response includes policy_evaluation.inputs.basin, read it as the decision-time policy basin. Agent-facing state fields can be sourced from the primary EISV path for that response, so newer responses also include policy_basin, policy_basin_source, and primary_eisv_source to make that measurement distinction explicit.

Verdicts

Governance issues a decision after each check-in. The response's verdict field wraps the decision action, which is binary — proceed or pause — qualified by a sub_action:

Action	Sub-action	Meaning	What to do
proceed	approve	State is healthy	Continue working normally
proceed	guide	Something is slightly off	Read the guidance text, adjust approach
pause	reject	Risk threshold reached	Stop current work, reflect; dialectic review or human input
pause	void_pause, coherence_pause, basin_pause, risk_pause, cirs_block	A specific subsystem tripped	Read the reason/guidance fields; consider dialectic review

Separately, metrics.verdict may carry an internal UNITARES verdict such as safe / caution / high-risk. Read it as interpreted state/context, not as moral judgment. In current default posture, behavioral assessment drives live policy actions: fixed thresholds during warmup, Welford z-score residuals after warmup, with floors and gates always in force. Φ/ODE scoring is telemetry and research lens, not the control loop.

Margin

margin describes how much headroom you have before the nearest state-space edge. It is a small enum, not a number:

margin	Meaning	What to do
settling	Warmup — fewer than 3 check-ins, so there is not enough history to judge headroom yet	Keep checking in; a real margin appears after 3+ check-ins
comfortable	Clear of every edge by a healthy distance	Proceed normally
tight	Within the edge threshold of the nearest boundary (or in the boundary basin)	Be more careful with next steps; avoid increasing complexity
warning	An edge has just been crossed (less than 0.1 past the threshold)	Stop increasing complexity; reflect before the next step
critical	An edge is crossed deeply (0.1 or more past the threshold)	Halt the current approach; recover or escalate

The actionable levels are tight, warning, and critical — each carries a companion nearest_edge field naming which boundary you are closest to (risk, coherence, or void). On comfortable and settling, nearest_edge is null (there is no edge to warn about). Prefer the live margin/nearest_edge values over assuming a fixed enum across runtime versions — get_governance_metrics() is the source of truth.

The plain-English mirror array in your check-in response already summarizes anything actionable (including a tight/warning/critical margin) — read that first. In mirror mode margin/nearest_edge are surfaced only when actionable; a comfortable/settling margin is steady-state and stays out of the response (the mirror's "No actionable signals — steady state" line covers it).

Coherence

Coherence measures how well your state vector holds together. It is calculated from EISV/monitor signals — not from the content of your work. Think of it as structural health, not semantic quality.

• Full range is [0, 1]
• Critical threshold is available via get_governance_metrics() in the thresholds field — do not hardcode it
• Do not chase a number — check in honestly and let it track naturally
• Coherence reflects balance, not performance

Calibration

The system tracks whether your stated confidence matches evidence. Over time this builds a calibration curve.

• Grounding comes from objective signals: test pass/fail, command exit codes, lint results, file operations. These feed calibration automatically via auto_ground_truth.py and the outcome_event hook. Human validation is not required for deterministic evidence.
• Overconfidence is tracked and can lower Integrity / raise uncertainty through the check-in pipeline

Diagnostics

When the numbers look surprising, do not guess first. Use:

• identity() to verify who the runtime thinks you are
• health_check() to verify the server and knowledge graph are healthy
• get_governance_metrics() for the current live thresholds and interpreted state

What NOT to Do

• Do not game coherence by reporting low complexity / high confidence on everything
• Do not ignore guide verdicts — they are early warnings before pause/reject
• Do not create duplicate discoveries — always search the knowledge graph first
• Do not check in after every trivial action — it is noise, not signal
• Do not leave high-severity findings as open forever — resolve or archive them