HMAA AUTONOMY RESEARCH PLATFORM v3
DEMPSTER-SHAFER w/ SHAFER DISCOUNTING • FORMAL FSM w/ HYSTERESIS • PHYSICS SENSORS • PREDICTIVE TWIN • CROSS-SENSOR VALIDATION • MONTE CARLO
T+0.0s
◆ GOVERNANCE
⊞ CONTROL
⚡ EXPERIMENTS
🧠 EXPLAINABLE
◎ TWIN
📋 TRACE
∀ FORMAL
◆AUTHORITY (FORMAL FSM)
A3
FULL AUTHORITY
◈D-S FUSED TRUST (SHAFER DISCOUNTED)
0.94
⬡COMMAND GATE
◇PER-SENSOR TRUST (D-S BELIEF + CROSS-VALIDATION)
⬢CARA RECOVERY
⚡ADVERSARIAL ENVIRONMENT
ΣD-S FUSION + CROSS-VALIDATION
▣EVENT LOG
📡TELEMETRY
◎TACTICAL MAP
⊞MISSION CONTROL (INTERLOCKED)
🛡SAFETY ENVELOPE
⚙SYSTEM HEALTH (WORKLOAD)
📊TRUST TIMELINE (PER-SENSOR)
▣CONTROL LOG
🧠DECISION TRACE
⛓LIVE REASONING
✕CROSS-SENSOR VALIDATION
📈GOVERNANCE TIMELINE
🤖PHYSICAL ROVER (SIMULATED)
💻PREDICTIVE TWIN (INDEPENDENT MODEL)
ΔDIVERGENCE ANALYSIS (MEASURED vs PREDICTED)
REQUIREMENTS TRACEABILITY MATRIX
Requirement Design Element Implementation Verification
FAILURE MODE AND EFFECTS ANALYSIS (FMEA)
Failure Mode Detection Method Sev. Mitigation Residual Risk
GOAL STRUCTURING NOTATION (GSN) SAFETY CASE
Illustrative safety-argument outline in GSN, with reference to MIL-STD-882E and DO-178C concepts (not an external compliance assessment)
TLA+ FORMAL SPECIFICATION, AUTHORITY FSM
Model-checkable specification. Properties verifiable with TLC model checker.
Scope of the state count reported by the model checker below: 23,748 reachable states of the discrete authority automaton, under the assumption that instantaneous authority equals its target. Continuous behaviour between decision instants is not covered, and the result is a specification-level property of the model rather than a claim about fielded hardware.
STATE SPACE
-- Authority FSM state vector
level ∈ {0, 1, 2, 3}     -- 0=A3, 1=A2, 2=A1, 3=A0
dwellCounter ∈ ℕ       -- ticks at current level meeting upgrade conditions
locked ∈ {TRUE, FALSE}  -- oscillation lockout active
lockExpiry ∈ ℕ         -- tick at which lockout expires
transitionCount ∈ ℕ    -- transitions in current window
TRANSITION GUARDS
Downgrade(trust, anomCnt) ≜
  target := IF trust < 0.25 ∨ anomCnt ≥ 3 THEN A0
            ELSE IF trust < 0.50 ∨ anomCnt ≥ 2 THEN A1
            ELSE IF trust < 0.80 THEN A2 ELSE A3
  ∧ target > level   -- only if more restrictive
  ∧ ¬locked
  ∧ level' = target   -- CAN SKIP LEVELS
  ∧ dwellCounter' = 0

Upgrade(trust, anomCnt, level, dwell) ≜
  ∧ level > 0 ∧ anomCnt = 0 ∧ ¬locked
  ∧ trust ≥ UpgradeThreshold[level]   -- hysteresis: UP > DOWN
  ∧ dwell ≥ DwellTime[level]   -- temporal guard
  ∧ level' = level - 1   -- EXACTLY ONE STEP
  ∧ dwellCounter' = 0
FORMAL PROPERTIES (TLC STATUS NOTED PER PROPERTY)
TLA+ MODULE (DOWNLOADABLE)
Full specification available for TLC model checking. State space: |level|×|trust|×|anomaly|×|locked|×|dwell| = 4×101×7×2×N


TLC MODEL CHECKING RESULTS
Excerpt of a TLC re-run on 2026-10-21 of the archived specification (HMAA_Authority_FSM, Zenodo 10.5281/zenodo.21300977) at MaxTick=8, MaxDwell=8, OscMaxTransitions=3, OscLockoutTicks=4: 23,748 distinct states, depth 9, no counterexample; 4 of 8 properties exercised, 4 vacuous at these bounds; the run at operational constants did not complete.