Skip to content

Releases: Octid-io/cloudless-sky

v2.1.0 — Math on the Wire

21 Apr 22:19

Choose a tag to compare

OSMP encodes instructions. EML encodes mathematics. Both ship in the same package.

This release adds the EML evaluator — a companion math-on-the-wire layer — across Python, Go, and TypeScript, plus two new MCP tools. Purely additive; no breaking changes.

What's new

EML across three languages (sdk/python/osmp/eml.py, sdk/go/osmp/eml/, sdk/typescript/src/eml.ts)

Based on Odrzywołek (2026, arXiv:2603.21852): a single binary operator eml(x, y) = exp(x) − ln(y), together with the constant 1, generates the standard calculator function basis — exp, ln, sin, cos, sqrt, arithmetic, and more — as compact expression trees. A full sin(x) or sqrt(x) approximation fits in fewer than 100 bytes on the wire.

  • Zero runtime dependencies. Uses only standard-library math primitives.
  • 16 pre-verified base corpus entries and 4 multi-variable arithmetic compounds (neg, +, ×, linear calibration).
  • Compact wire encodings for constrained-channel transmission.

Cross-language byte-exact determinism

All three language implementations produce the identical SHA-256 corpus fingerprint:

e9a4a71383f14624472fe0602ca5e0ff1959e00b09725a62d584e1361f842c1b

Every platform, every runtime, every deployment — identical bits out of the same encoded chain.

Dual-mode precision

  • Fast mode (fdlibm-derived, 1-ULP accurate) — included publicly under Apache 2.0. Correct for LoRa / BLE / edge-ML, drone swarm coordination, constrained-channel telemetry, and general scientific computation.
  • Precision mode (correctly-rounded, audit-grade) — available under commercial license for regulated-industry applications: medical (IEC 62304), aerospace (DO-178C), nuclear (IEC 61513), audit-grade financial, cryptographic protocol-frame hash inputs, and DoD / defense-aerospace deployments.

Contact ack@octid.io for precision-mode access — see PATENT-NOTICE.md.

Two new MCP tools (total now 19, up from 17)

  • osmp_eml_evaluate(chain_name, values) — evaluate a pre-built EML chain at input value(s)
  • osmp_eml_corpus_lookup(chain_name) — list or inspect the 20 pre-verified chains

Install

# Python SDK
pip install osmp==2.3.2

# MCP server (wraps the Python SDK, exposes 19 tools)
pip install osmp-mcp==1.1.0

# TypeScript / JavaScript SDK
npm install osmp-protocol@2.3.2

# Go SDK
go get github.com/octid-io/cloudless-sky/sdk/go/osmp/eml

Full changelog: CHANGELOG.md.

Attribution

Built on the universal binary operator eml(x, y) = exp(x) − ln(y) introduced by Andrzej Odrzywołek (Jagiellonian University, arXiv:2603.21852, March 2026). The operator itself is not claimed; the present work covers the transmission, encoding, and apparatus layer distinct from the operator.

Commercial services

Three request-gated services are available under commercial license (see PATENT-NOTICE.md):

  • Omega namespace extensions — custom opcode registration for specialized domains
  • MDR certification — authoritative namespace mapping for regulated deployments
  • UBOT precision mode — correctly-rounded evaluator for audit-grade applications

Contact ack@octid.io.

v2.0.0

16 Apr 03:17

Choose a tag to compare

OSMP v2.0.0 — Cloudless Sky

Breaking Changes

  • v15 Dictionary (352 opcodes): 24 opcode renames for improved cross-model readability. Existing SAL instructions using renamed opcodes (B:BA, B:BS, B:L, E:OBS, F:PRO, M:A, M:MA, Y:PROMOTE, Y:STAT, Z:CAP, and others) must be updated. Full rename table in protocol/OSMP-semantic-dictionary-v15.csv.
  • ADJ meta-verb opcodes removed: A:ADJ, O:ADJ, R:ADJ, Z:ADJ removed per Principle 3 (no meta-verb opcodes). Operations expressed through slot values and delta glyphs instead.

New: Deterministic NL-to-SAL Composition (All 3 SDKs)

  • SALComposer: deterministic pipeline composes SAL from natural language without inference. Generation index with 358 phrase triggers (318 auto-generated from ASD definitions + 40 curated). Confidence gate prevents false positives. BAEL byte pre-check. 95.7% opcode coverage on the full 352-opcode dictionary. 98.6% with LLM fallback.
  • Ported to all three SDKs: Python (SALComposer), TypeScript (SALComposer), Go (Composer). Same pipeline, same results, same generation index.
  • MCP tools: osmp_compose (NL to SAL), osmp_macro_list, osmp_macro_invoke. 17 tools total.
  • Composition tested across 3 paths: deterministic tool (91.7%) outperforms frontier LLMs with full doctrine (77-81%).

New: Cross-Model Composition Test Suite

Six test scripts for measuring composition quality: deterministic sweep, three-path comparison, agent roundtrip, panel validation, panel discovery, full dictionary sweep (352 opcodes x 3 composition paths).

New Features

  • ADR-004 Dictionary Basis Manifest: SAIL intern table is now a pure function of an ordered, content-addressed basis. FNP extended-form ADV (0x81) exchanges 8-byte basis fingerprints. Sessions grade to ESTABLISHED_SAIL or ESTABLISHED_SAL_ONLY based on basis agreement. Prevents silent SAIL misdecode between nodes with different dictionary loadouts. Implemented across Python, TypeScript, Go.
  • MacroRegistry: Pre-validated SAL instruction chain templates. Register, expand with slot-fill, encode in compact or expanded wire format. Consequence class inheritance from chain. 16 Meshtastic macros shipped (DeviceMetrics, EnvironmentMetrics, AirQuality, Power, Health, LocalStats, Position, NodeInfo, Alert, Traceroute, Waypoint, threshold rules, MEDEVAC).
  • ADR-005: Behavioral vs actuator command layers in R namespace. R:ACC/R:THR and R:DECEL/R:BRK paired entries for motion commands spanning planning and control layers.
  • 11 new opcodes: A:CMPLY, A:PERM, G:DOP, Q:ANL, Q:FB, Q:JDG, Q:RPRT, R:ACC, R:THR, R:DECEL, R:BRK.

Improved: Bridge and Decode

  • Bridge receiver parses [SAL: ...] wrapper from model responses
  • decode_natural_language handles all chain operators, outputs domain context labels and readable NL
  • Bridge annotation quality improved for acquisition seeding

Spec Updates

  • Principle 3: No meta-verb opcodes (Section 12.3)
  • Principle 4: Namespace-mnemonic alignment (Section 12.3)
  • Composition Rule 10: Meta-verb prohibition (Section 12.5)
  • Section 3.6: SAIL bijection anchored to basis equality
  • Sections 9.1-9.3: Extended-form ADV/ACK, basis fingerprint computation
  • Section 9.5: State machine updated (ESTABLISHED_SAIL, ESTABLISHED_SAL_ONLY)
  • Section 9.8: Dictionary Basis Manifest specification

Bug Fixes

  • Finding 2: ICD-10 code dot-normalization in MCP resolver (J93.0 now resolves)
  • Finding 49: Go SAIL encoder I:section-sign (U+00A7) character class fix
  • Finding 17: Stale opcode count references corrected
  • Cross-SDK test runner paths fixed

Infrastructure

  • gen_asd.py regenerates TypeScript and Go ASD tables from canonical Python source
  • ASD drift check clean across all three SDKs
  • Cross-SDK wire compatibility: 86/86 Python, 86/86 TypeScript, 86/86 Go
  • 624 Python tests passing

SDK Versions

  • osmp (Python): 2.2.2
  • osmp-mcp: 1.0.34
  • osmp-protocol (TypeScript): 2.2.2

Upgrade Guide

Update renamed opcodes in any existing SAL strings. Remove any usage of A:ADJ, O:ADJ, R:ADJ, Z:ADJ. Update pip: pip install --upgrade osmp

OSMP v1.0.1 - Cloudless Sky

06 Apr 06:24

Choose a tag to compare

OSMP (Octid Semantic Mesh Protocol) v1.0.1. Tier 1 developer experience, SALBridge propagation, SAIL binary wire codecs, SEC security envelope, Rule 8 regulatory dependency grammar, package split. Patent pending, priority date March 17, 2026. Apache 2.0 with express patent grant.

Developer Experience

  • Tier 1 API across all three SDKs: from osmp import encode, decode (Python), import { encode, decode } (TypeScript), osmp.Encode() / osmp.Decode() (Go). Three lines, zero setup, zero dependencies.
  • Package split: pip install osmp is the standalone SDK. pip install osmp-mcp is the MCP server. The SDK has zero dependencies beyond Python stdlib. MCP wraps the SDK, not the other way around.
  • PyPI: osmp v2.0.1 (new package), osmp-mcp v1.0.26
  • Tier 2 class-based API (from osmp.core import OSMP) for advanced configuration
  • KNOWN-ISSUES.md documents platform-specific install notes (Termux, Raspberry Pi, constrained hardware)

SALBridge: Language Propagation by Contact

  • SALBridge shipped in all three SDKs. OSMP spreads by contact, not installation.
  • FNP state machine extended: IDLE, ADV_SENT, ESTABLISHED, SYNC_NEEDED, FALLBACK (non-OSMP peer), ACQUIRED (peer learned SAL through context exposure)
  • Bridge annotates outbound SAL with natural language equivalents, seeding the remote agent's context window with SAL patterns
  • Monitors inbound messages for valid SAL fragments. When the remote agent starts producing valid SAL (few-shot acquisition), transitions to ACQUIRED and sends pure SAL
  • Regression detection: ACQUIRED peers that stop producing valid SAL drop back to FALLBACK
  • Comparison logging: every bridged message records SAL bytes vs NL bytes for side-by-side measurement
  • from osmp import bridge -- Python factory function for zero-setup bridge creation

Protocol

  • Adaptive Shared Dictionary (ASD) v14: 342 opcodes across 26 namespaces (26 new opcodes in Autonomous Ops MDR extension, runtime-loaded)
  • SAIL (Semantic Assembly Isomorphic Language): Binary wire encoding, isomorphic to SAL. Bijective 1:1 mapping. Compose and debug in SAL, deploy in SAIL, decompile back to SAL for inspection. Wire codecs shipped in all three SDKs.
  • SEC Security Envelope: Node ID + monotonic sequence counter + AEAD tag + Ed25519 signature. 87 bytes fixed overhead. Mesh authentication with no certificate authority.
  • Four-mode wire architecture: SAL/SAIL x Unsigned/SEC. BAEL selects automatically based on channel capacity and consequence class.
  • Rule 8 -- Regulatory dependency grammar: REQUIRES rules as SAL expressions, evaluated within the SAL framework. Conjunctive prerequisites supported (A AND B OR C AND D). 27 rules in Autonomous Ops MDR corpus. Shipped in all three SDKs.
  • Six-category glyph architecture, Overflow Protocol (OP), BAEL floor guarantee, FNP handshake, ADP dictionary synchronization, R:ESTOP hard exception, SAL Usage Doctrine v2.2 -- all unchanged from v1.0.0.

SDKs

  • Python (reference implementation): CONFORMANT. pip install osmp
  • TypeScript (npm: osmp-protocol): CONFORMANT. Tier 1 encode/decode added.
  • Go (zero filesystem/network dependency): CONFORMANT. Package-level Encode/Decode with sync.Once singleton.
  • Cross-SDK wire compatibility verified. SAIL round-trip verified on all 55 canonical test vectors.

MCP Server

  • PyPI: pip install osmp-mcp (v1.0.26)
  • MCP Registry: io.github.Octid-io/osmp
  • 14 tools: encode, decode, compound_decode, lookup, validate, discover, resolve, batch_resolve, benchmark, bridge_register, bridge_send, bridge_receive, bridge_status, bridge_comparison
  • All tool descriptions expanded with usage examples and parameter documentation
  • Import path updated for package split with legacy fallback

Domain Corpora (Managed Dictionary Registry)

  • ICD-10-CM: 74,719 clinical codes (477KB compressed)
  • ISO 20022: 47,835 financial definitions (1.2MB compressed)
  • MITRE ATT&CK Enterprise v18.1: 1,661 entries (20KB compressed)
  • All three resolvable by all three SDKs without network access

Measured Performance (29-vector benchmark)

  • 86.8% byte reduction vs minified JSON
  • 84.5% byte reduction vs MessagePack
  • 70.5% byte reduction vs compiled Protocol Buffers (protoc 3.21.12)
  • 76.0% token reduction (GPT-4 cl100k_base)

Behavioral Compliance (cross-model)

  • SAL: 88-95% compliance across Claude Sonnet 4, GPT-4o, GPT-4o-mini
  • JSON: 85-90% at full wire cost
  • Consequence class identification: 100% SAL vs 75% JSON across all models tested

Test Suite

  • 212 tests passing (122 existing + 35 DX + 55 SAIL round-trip)
  • DX tests cover: Tier 1 API, Tier 2 class API, MCP tool interface, SAIL round-trip on all 55 canonical vectors, consequence class preservation, compound operator preservation
  • Bridge tests: FNP state transitions, annotation, acquisition, regression, comparison logging

Whitepaper

  • SAL Efficiency Analysis v2.2: docs/SAL-efficiency-analysis.md
  • Full methodology, .proto schemas, reproduction scripts in benchmarks/sal-vs-json/

OSMP v1.0.0 - Cloudless Sky

29 Mar 20:04
196b7ff

Choose a tag to compare

OSMP (Octid Semantic Mesh Protocol) v1.0.0. Open encoding standard for agentic AI instruction exchange. Patent pending, priority date March 17, 2026. Apache 2.0 with express patent grant.

Protocol

  • Semantic Assembly Language (SAL) instruction format with formal EBNF grammar
  • Adaptive Shared Dictionary (ASD) v13: 342 opcodes across 26 namespaces
  • Six-category glyph architecture (18 operators, 4 consequence class designators, 2 outcome states, 6 parameter/slot designators, 3 loss tolerance policies, 3 dictionary update modes)
  • Sovereign namespace extension (Omega)
  • Overflow Protocol (OP) with Tier 1 (single packet), Tier 2 (sequential burst), and Tier 3 (DAG decomposition for conditional branches and dependency chains)
  • Bandwidth-Agnostic Efficiency Layer (BAEL) floor guarantee: encoding never exceeds natural language input
  • Frame Negotiation Protocol (FNP) handshake: 78 bytes, two LoRa packets
  • ASD Distribution Protocol (ADP) for dictionary delta synchronization across nodes
  • R:ESTOP hard exception: overrides all policies on single fragment receipt
  • SAL Usage Doctrine v2.2 (350-vector test suite, 99.1% BOUND / 100% SAFE)

SDKs

  • Python (reference implementation): CONFORMANT
  • TypeScript (npm: osmp-protocol): CONFORMANT
  • Go (zero filesystem/network dependency): CONFORMANT
  • Cross-SDK wire compatibility verified on 86-instruction corpus

MCP Server

  • PyPI: pip install osmp-mcp (v1.0.23)
  • MCP Registry: io.github.Octid-io/osmp
  • 9 tools: encode, decode, compound_decode, lookup, validate, discover, resolve, batch_resolve, benchmark

Domain Corpora (Managed Domain Registry)

  • ICD-10-CM: 74,719 clinical codes (477KB compressed)
  • ISO 20022: 47,835 financial definitions (1.2MB compressed)
  • MITRE ATT&CK Enterprise v18.1: 1,661 entries (20KB compressed)
  • All three resolvable by all three SDKs without network access

Measured Performance (29-vector benchmark)

  • 86.8% byte reduction vs minified JSON
  • 70.5% byte reduction vs compiled Protocol Buffers (protoc 3.21.12)
  • 84.5% byte reduction vs MessagePack
  • 76.0% token reduction (GPT-4 cl100k_base)

Whitepaper

  • SAL Efficiency Analysis v2.2: docs/SAL-efficiency-analysis.md
  • Full methodology, .proto schemas, reproduction scripts in benchmarks/sal-vs-json/