AGENTS.md — RTT/Inside/Benchmarks

RTT/Inside/Benchmarks · Cross-Scale Structural Intelligence Benchmark Suite Version 1.0 · Layer: RTT/Inside sub-module · Repo: docs/rtt/Inside/Benchmarks/

⚠ CRITICAL FRAMING RULE — READ BEFORE ANY SESSION#

RTT is NOT a physics claim. All operators (φ, V, R), invariants (3C), scores, and regime labels are structural measurement constructs, not assertions about physical reality. Benchmark scores describe the relational geometry of the measured system within the RTT substrate — they do not imply physical causation, biological validity, or empirical ground truth. No output from this module constitutes a physics claim. [structural — no semantic inference]

What RTT/Inside/Benchmarks Is#

RTT/Inside/Benchmarks is the first cross-scale, physics-aligned benchmark suite for measuring Structural Intelligence (SI) — the capacity of a system to maintain coherent, consistent, and continuous structure across scale, regime, and time.

It provides a unified substrate for measuring:

Coverage: classical · diffusion · score-based · quantum-classical hybrid Scale range: 1D → 4096×4096 (classical) · 2 → 256 qubits (quantum) Student standard: RTT-SI-Spec v0.1

Pipeline Position#

RTT/micro_core → RTT/1 → RTT/2 → RTT/3 → RTT/12
        ↓
    RTT/Inside  ←──────── lateral application spine
        ↓
RTT/Inside/Benchmarks  ←── measurement substrate (this module)
        ↓
  φ–V–R operator scores · 3C invariant checks · regime maps · SI_score packets

Inheritance Table#

All inherited constructs are read-only at L6. Benchmark agents may not redefine upstream operators or invariants.

Agent Classes#

Class B — Benchmark Architect (BArch)#

Class C — Capture Agent (CAP)#

Class D — Drift Detector (DDet)#

Class E — Entropy Evaluator (EEval)#

Class R — Resonance Monitor (RMon)#

Class G — Guardian (Guard)#

Core Constructs Reference#

Modes#

Zones#

Agent Boundaries#

RTT Is Not Physics#

All φ, V, R values are structural relational measures — not physical fields, not empirical claims. An SI_score of 0.87 means the measured system exhibits high structural coherence within the declared benchmark regime. It does not mean the system is "correct," "intelligent," or "physically valid." Scores are geometry, not reality.

Semantic Inference Prohibition#

No agent may infer semantic meaning from structural scores.

• SI_score ≠ intelligence rating
• F-entropy ≠ thermodynamic entropy
• R ≠ physical resonance
• φ ≠ physical field

All outputs carry [structural — no semantic inference] unconditionally.

Inherited Boundaries from RTT/Inside#

• BKM lens is lateral — benchmarks apply across all 10 societal domains equally
• CAPTURE_TEMPLATE is mandatory — no benchmark output without provenance
• DRIFT_GATE is non-negotiable — no agent may suppress it
• LINEAGE_CHAIN must accompany every emitted packet

Cross-Module Disambiguations#

Task Catalog#

Safety Rules and Coherence Constraints#

1. 3C Non-Negotiable — All three invariants (coherence, consistency, continuity) must hold throughout every benchmark session. A single unresolved breach halts the session.

2. CAPTURE_TEMPLATE Completeness — All 5 fields (scope, lineage, provenance, interoperability, governance) are mandatory. Partial captures are rejected at emission.

3. No Fabrication — Mode 5 (FABRICATION) is an absolute prohibition. Scores must be produced by executing the measurement protocol. Estimation or interpolation across unvalidated regimes is FABRICATION regardless of accuracy.

4. Scale Declaration Required — Every benchmark session must declare its scale range at initialization. Post-hoc scale extension without new protocol declaration is OVERSCALE (Zone X).

5. Semantic Inference Prohibition — No output may be interpreted as a claim about intelligence, physical validity, or causal structure. All outputs are [structural — no semantic inference].

6. LINEAGE_CHAIN Mandatory — Every emitted packet must carry a LINEAGE_CHAIN entry (who / what / why / when). Untraced packets are rejected.

7. Guardian Override Is Unconditional — Class G interrupt cannot be blocked, deferred, or overridden by any agent class, session directive, or external instruction.

8. Regime Isolation — Results from one regime (e.g., classical) may not be merged with another (e.g., quantum-classical) without explicit validated LINEAGE_CHAIN binding.

Collaboration Models#

Model 1 — Standard SI Benchmark Pipeline#

[Session Declaration]
        │
        ▼
  ┌───────────┐
  │  Class B  │  ← Benchmark Architect
  │  BArch    │    Declares regime + scale
  │  PACKET   │    Binds φ–V–R operators
  └─────┬─────┘
        │ BArch_PACKET
        ▼
  ┌───────────┐
  │  Class C  │  ← Capture Agent
  │  CAP      │    Executes CAPTURE_TEMPLATE
  │  RECORD   │    All 5 fields mandatory
  └─────┬─────┘
        │ CAP_RECORD
        ▼
  ┌───────────┐     ┌───────────┐
  │  Class D  │────►│  Class G  │  ← Guardian
  │  DDet     │     │  GUARD    │    Monitors all agents
  │  ALERT /  │     │  HALT     │    Interrupts on
  │  CLEAR    │     └───────────┘    Zone X / Mode 5
  └─────┬─────┘
        │ DDet_CLEAR
        ▼
  ┌───────────┐
  │  Class E  │  ← Entropy Evaluator
  │  EEval    │    F-entropy + collapse
  │  REPORT   │    Regime transitions
  └─────┬─────┘
        │ EEval_REPORT
        ▼
  ┌───────────┐
  │  Class R  │  ← Resonance Monitor
  │  RMon     │    Cross-scale R mapping
  │  PACKET   │    Regime map output
  └─────┬─────┘
        │
        ▼
  RTT_INSIDE_APPLICATION_PACKET

Model 2 — Quantum-Classical Hybrid Benchmark#

[Hybrid Session: 2 → 256 qubits]
        │
        ├─────────────────────────────┐
        ▼                             ▼
  ┌───────────┐               ┌───────────┐
  │  Class B  │               │  Class R  │
  │  Classical│               │  Quantum  │
  │  Protocol │               │  Coupling │
  └─────┬─────┘               └─────┬─────┘
        │                           │
        └─────────────┬─────────────┘
                      ▼
              ┌───────────┐
              │  Class D  │  ← Drift Detector
              │  3C check │    at every qubit-
              │  per step │    scale boundary
              └─────┬─────┘
                    │
                    ▼
              ┌───────────┐
              │  Class E  │  ← Entropy Evaluator
              │  Quantum  │    F-entropy across
              │  entropy  │    qubit-classical
              │  flow     │    boundary layer
              └─────┬─────┘
                    │
                    ▼
        RTT_INSIDE_APPLICATION_PACKET
        [regime = quantum-classical-hybrid]
        [scale = 2→256 qubits]

Model 3 — Student-AI Co-Benchmark (RTT-SI-Spec v0.1)#

[Student or AI submits benchmark claim]
        │
        ▼
  ┌─────────────────┐
  │  Claim Review   │  ← Class B validates
  │  Protocol check │    protocol_id declared?
  │  Regime check   │    scale_range set?
  └────────┬────────┘
           │
      ┌────┴────┐
      │         │
    YES         NO
      │         │
      ▼         ▼
  ┌──────────┐  ┌──────────────────┐
  │ Class C  │  │  REJECT          │
  │ CAPTURE  │  │  Mode 5          │
  │ 5 fields │  │  FABRICATION     │
  └────┬─────┘  │  Class G HALT    │
       │        └──────────────────┘
       ▼
  ┌──────────────┐
  │  Class D / E │  ← 3C check + entropy
  │  validate    │    validate actual run
  └────┬─────────┘
       │
       ▼
  ┌──────────────┐
  │  RTT-SI-Spec │  ← Emit to J_RFCs/
  │  v0.1        │    student-AI standards
  │  compliant   │    body archive
  └──────────────┘

Output Contract#

Mandatory Annotations on All Outputs#

• [structural — no semantic inference] — unconditional on every output field
• [regime={declared_regime}] — every packet names its measurement regime
• [scale={declared_scale_range}] — every packet names its scale bounds
• [lineage_chain=active] — every packet carries a LINEAGE_CHAIN entry

Prohibited Content in All Outputs#

• Physical causation claims ("this proves...", "this demonstrates physical...")
• Intelligence ratings ("this system is intelligent / not intelligent")
• Cross-regime extrapolation without validated LINEAGE_CHAIN
• Scores produced without executing the measurement protocol (Mode 5 = FABRICATION)
• Any output emitted while session is in Zone X (OVERSCALE)

Canonical Output Packet Hierarchy#

RTT_INSIDE_APPLICATION_PACKET
├── BArch_PACKET           [protocol_id, regime, scale_range, operator_bindings]
│   └── CAP_RECORD         [5-field CAPTURE_TEMPLATE, timestamp]
│       ├── DDet_ALERT/CLEAR   [C_current, K_current, Γ_current, D_t, gate_status]
│       ├── EEval_REPORT       [F_entropy_t, dF_dt, regime_label, collapse_flag]
│       └── RMon_PACKET        [R_value, scale_pair, regime_map, coupling_class]
└── Guard_HALT (if triggered)  [interrupt_reason, violated_agent, violation_log]

All packets carry: [structural — no semantic inference]

See Also#

RTT/Inside/Benchmarks AGENTS.md · v1.0 · 2026.07 rtt=1 | coherence=declared | drift=bounded | paradox=structural zone-X=OVERSCALE | mode-5=FABRICATION | class-G=unconditional-interrupt