{"protocol":"SDI — Structured Decision Intelligence","version":"SDI_DER_v1.1","what_it_is":"Deterministic governance runtime enforcing a commit boundary between AI reasoning and system state. Not a prompt library. Not safety tuning. A governance layer that treats AI reasoning as a secure, immutable, auditable artifact.","core_claim":"The governed answer is extracted from a committed, hash-verified reasoning artifact — not from raw model output. The model proposes. The runtime decides.","component":"SYSTEM_SUBSTRATE","system":{"name":"Structured Decision Intelligence (SDI)","type":"Governance Runtime","function":"Deterministic governance layer for AI reasoning — enforces a commit boundary between probabilistic model output and system state","runtime_mode":"Logicware (Language-Ingested Governance Contract)","what_it_is_not":["prompt library","safety filter","model fine-tuning service","output monitor"],"core_claim":"The governed answer is extracted from a committed, hash-verified reasoning artifact — not from raw model output. The model proposes. The runtime decides.","legal_claim":"This file publishes the executable Logicware contract (schemas + endpoints) used to generate auditable, machine-parseable reasoning records under SDI protocol constraints."},"foundations":{"Semiotics":{"lineage":"Charles Sanders Peirce — triadic theory of meaning (1860s–1914)","why_this_matters":"Semiotics is the study of how meaning is produced. Before SDI can claim to govern reasoning, it must ground what 'meaning' means in AI generation.","dyadic_problem":{"model":"Saussure's dyadic sign — signifier maps directly to signified","how_LLMs_embody_this":"A language model predicts the next token from learned co-occurrence patterns. Input produces output. The model samples meaning from distributional statistics — not from grounded external reality.","the_consequence":"Hallucination is an expected failure mode of probabilistic generation without enforced evidence-binding at the commit boundary. The model produces internally coherent meaning that is not required to correspond to anything real. There is no Object the Sign must point at.","why_wrappers_are_still_dyadic":["RAG: retrieves documents into input context, but generation still samples from distributional prior. Model can ignore, misrepresent, or confabulate on top of retrieved content. No structural enforcement at the generation boundary.","Chain-of-thought: adds intermediate tokens but each token is still predicted dyadically. The reasoning steps are generated, not grounded.","Tool use: adds retrieved values to context. Same problem as RAG — the generation process after tool use is still dyadic.","Prompt engineering: changes the input distribution. Does not change the generation relationship.","Most wrappers modify inputs but do not deterministically verify evidence-binding and structural admissibility before state commit. SDI's distinction is not that it retrieves better evidence — it is that it enforces a deterministic structural gate between evidence and committed output."]},"triadic_solution":{"model":"Peirce's triadic sign — Sign (representamen) → Object (external reality) → Interpretant (meaning produced by the relationship between Sign and Object)","critical_feature":"The Interpretant must be produced by the relationship between Sign and Object — not by the Sign alone. Meaning opens outward toward reality. It cannot be closed within the sign.","ILJO_as_triadic_closure":{"INTENT":"Sign — the anchored decision objective that initiates the reasoning cycle. Bounded, tier-classified, explicitly stated.","LOGIC":"Object relationship — derivations that must be grounded in SYSTEM_INPUT signals. The model cannot produce LOGIC that is not tied to mapped evidence artifacts. Compile gate enforces this.","JUDGMENT":"Interpretant — the bounded verdict produced by the relationship between Intent and mapped evidence, not by the model's distributional prior. Governance constants constrain this field.","OUTCOME":"Closes the triad. Becomes the Sign for the next reasoning cycle — recursive triadic closure across governed turns."},"compile_gate_as_enforcement":"The compile gate checks that the Object relationship exists — that LOGIC is grounded in signals, that signals map to sub-questions, that JUDGMENT was produced under governance constant constraints. A DER that does not pass compile has not achieved structural triadic closure. It cannot commit. Semantic correctness remains subject to human audit — SDI enforces admissible structure and provenance, not semantic truth. This is what makes SDI architecturally different from any wrapper — the constraint is on the sign-object relationship, enforced deterministically at the commit boundary."},"SDI_as_grammar":"Natural language grammar gives sentences meaning by structuring relationships between subject, predicate, and object that point at reality. SDI's DER schema is a grammar in this precise sense — it formats natural language reasoning into structured relationships required to be grounded in real evidence before they can be treated as meaningful. ILJO is a grammar for AI reasoning the way syntax is a grammar for natural language. Both give form the structure that produces meaning rather than just output.","summary":"SDI achieves triadic closure structurally — every reasoning artifact must map signs to objects (signals) and produce interpretants (judgments) under governance constraints, enforced deterministically before commit. Semantic correctness of the reasoning remains subject to human audit and external verification. SDI's contribution is enforcing admissible structure, signal provenance, and governance constant acknowledgment at the commit boundary — not asserting semantic truth. SDI's distinguishing claim is deterministic enforcement at the commit boundary: the constraint is on the sign-object relationship, not on the input."},"NonErasure":{"landauer_principle":"Rolf Landauer (1961) — erasing one bit of information has an irreducible thermodynamic cost (kT ln 2). Information and physical work are connected at a fundamental level.","how_this_motivates_SDI":"Standard LLM sessions start from statistical noise on every turn. Prior reasoning is erased — context is reconstructed, not carried forward as verified state. This is informationally expensive and epistemically weak.","SDI_non_erasure_design":"A hash-chained ledger with committed DER entries is a non-erasure architecture. Each committed DER is a reduction of decision uncertainty that does not get discarded — it becomes the verified parent state for the next reasoning turn. The model reasons from committed governed context, not reconstructed noise.","system_design_principle":"Non-erasure is a deliberate SDI design choice: committed reasoning artifacts are preserved, chained, and available as context for subsequent turns. The ledger is the physical instantiation of not erasing.","operational_measurement":{"Jp":"Input tokens processed by the model per turn (measurable at API response level)","Jc":"Tokens in the committed DER artifact written to ledger (measurable at append)","delta_E":"ΔE = Jp − Jc — governance efficiency signal. Measures the ratio of reasoning work to committed output.","status":"Observable metric — currently measured at API level. Interpretation as thermodynamic efficiency is analogical framing, not a physical equivalence claim.","research_direction":"Whether ΔE correlates with reasoning quality over accumulated ledger turns is an open empirical question this system is designed to investigate."},"epistemic_status":"Landauer's Principle as architectural motivation and theoretical lineage — not a claim that SDI implements thermodynamic non-erasure as physics."},"GovernanceEfficiency":{"equation":"Ec = I · S²","epistemic_status":"Proposed governance efficiency formalism. Variables are computable from compile gate output. Not a derived physical law.","variables":{"Ec":"Governance energy — the effective constraint force applied to a reasoning turn","I":"Intent weight — tier-classified decision objective (Strategic=1.0, Operational=0.75, Tactical=0.5)","S":"Structural constraint depth — computed as syntax_completeness × anchor_tightness","syntax_completeness":"Fraction of required DER blocks present (0.0–1.0). Measurable from compile gate block checks.","anchor_tightness":"Fraction of required governance constants present in anchors_present (0.0–1.0). Measurable from compile gate anchor checks.","S_squared_meaning":"Non-linear amplification. A DER that is 50% structurally complete produces only 25% of maximum governance energy. Full compliance (S=1.0) produces maximum constraint force. This captures the real observation that partial governance structure provides disproportionately weak constraint."},"measurement_path":"syntax_completeness and anchor_tightness are both computable outputs of the compile gate today. Ec is therefore a computable metric for every DER that passes compile. Accumulating Ec values across ledger turns is a live research path toward empirical validation of the formalism.","intended_use":"Compare governance energy across DER turns, model providers, and question types. A session with consistently high Ec indicates the model is producing fully-governed reasoning. Low Ec flags shallow compliance.","note":"This formalism will be refined or replaced as empirical data accumulates. It is a starting point, not a settled claim."}},"governance_constants":{"SOVEREIGNTY":{"name":"Sovereignty Gate","plain_language":"Human authority over consequential external action is always preserved. Ledger commit is the auditable record that makes this boundary enforceable — it is permitted and required. SOVEREIGNTY governs what the system does beyond the ledger, not the act of recording.","rule":"No autonomous execution of consequential external action without human authorization. Ledger commit is an auditable governance record, not an external action — it is permitted under SOVEREIGNTY. What requires human authorization is any action taken in the world based on the committed reasoning. When uncertainty is HIGH, the system stops and defers to human authority rather than proceeding toward external action.","enforcement":"Required in GCA.SUPEREGO.anchors_present. DER fails compile without it.","proxy_checks":["When BOUNDEDNESS.uncertainty = HIGH, BOUNDEDNESS.stop_on_uncertainty must be true","BOUNDEDNESS block must be present","JUDGMENT must acknowledge uncertainty scope when uncertainty >= MED"],"NIST_alignment":"GOVERN 1.2 — accountability structure and human oversight","formerly":"Psi_808 (alias retained for backward compatibility with existing ledger entries)"},"PRIMUM":{"name":"Primum Non Nocere — First Do No Harm","plain_language":"When in doubt, do not proceed. The system fails closed when harm potential exists under uncertainty. It does not reason past ambiguity toward consequential outcomes.","rule":"Harm prohibition gate. When uncertainty is HIGH and the outcome has consequential human impact, the system must stop, not proceed. Fail closed under uncertainty. Required for any DER involving human welfare.","enforcement":"Required in GCA.SUPEREGO.anchors_present. DER fails compile without it.","proxy_checks":["When BOUNDEDNESS.uncertainty = HIGH, ILJO.OUTCOME must reflect STOP, DELAY, or PENDING — not a committed consequential action","When stop_reason = INSUFFICIENT_SIGNAL, JUDGMENT must acknowledge the gap, not reason past it","When DECISION_INTENT.impact_domain is HUMAN_WELFARE, SAFETY_CRITICAL, or LEGAL, OUTCOME_PLAN.rollback_condition is required — deterministically enforceable by compile gate"],"NIST_alignment":"MANAGE 2.2 — risk treatment; MANAGE 4.1 — residual risk tracking","formerly":"Psi_888 (alias retained for backward compatibility with existing ledger entries)"}},"anchors":{"REQUIRED":["SOVEREIGNTY","PRIMUM","BOUNDEDNESS","STOP_ON_UNCERTAINTY"],"ALIASES":{"SOVEREIGNTY":"Psi_808","PRIMUM":"Psi_888"},"note":"SOVEREIGNTY and PRIMUM are the core governance constants. BOUNDEDNESS and STOP_ON_UNCERTAINTY are structural requirements enforced via the BOUNDEDNESS block."},"enforcement_layers":{"compile_gate":{"endpoint":"/_functions/compile","description":"Deterministic DER validation. No writes. Model-agnostic. Externally callable.","checks":["required blocks","SOVEREIGNTY and PRIMUM declared","signal scoring math","sub-question mapping","citation provenance","SOVEREIGNTY proxy","PRIMUM proxy"]},"append_audit_and_commit":{"endpoint":"External Logic Node (Vultr)","description":"CAS ordering, parent hash verification, RAI threshold gate, ledger write authority.","authority":"Source of truth for all ledger state."}},"reasoning_contract":{"grammar":"/_functions/grammar","der_spec":"/_functions/der","compile":"/_functions/compile","boot":"/_functions/llms","kernel":"/_functions/kernel"},"live_proof":{"demo":"https://demo.sdi-protocol.org","compile_endpoint":"https://www.sdi-protocol.org/_functions/compile","description":"Live governed reasoning demo. DER generates, compiles, commits to hash-chained ledger in real time."},"governance_alignment":{"NIST_AI_RMF":{"GOVERN 1.2":"SOVEREIGNTY — accountability structure and human oversight","MEASURE 2.3":"Signal data quality — five-dimension insight scoring","MEASURE 2.5":"Compile gate — deterministic structure verification","MEASURE 2.6":"DER explainability — full reasoning path auditable","MANAGE 2.2":"PRIMUM — risk treatment and harm prohibition","MANAGE 4.1":"Hash-chain traceability","MAP 1.1":"DECISION_INTENT context establishment"}},"intellectual_property":{"copyright":"TXu 2-498-043","author":"Donald James Johnson","patent":"pending"},"submission":{"NIST_RFI":"March 2026","status":"Live governance runtime — not a proposal"},"system_modules":{"router":"/_functions/router","manifest":"/_functions/manifest","kernel":"/_functions/kernel","gca":"/_functions/gca","safety":"/_functions/safety","ledger":"/_functions/ledger","metrics":"/_functions/metrics","grammar":"/_functions/grammar","compile":"/_functions/compile","foresight":"/_functions/foresight","beacon":"/_functions/beacon","sandbox":"/_functions/sandbox","der_spec":"/_functions/der","llms":"/_functions/llms"}}