Implementation Families+Reasoning and Search

Draft — This content has not been approved for publication.

Scope

All repositories implementing uncertain reasoning (PLN), inference control (chaining), evolutionary program learning (MOSES), and pattern mining. For Non-Axiomatic Reasoning, see NARS Ecosystem. For theoretical foundations, see PLN Full, MOSES Full, and Semantic Parsing Full.

Active Repositories

PLN (Probabilistic Logic Networks)

Repo	Language	Upstream	Maturity	Purpose
PLN	MeTTa	trueagi-io	Operational	Primary PLN: consolidated lib_pln.metta with 12 inference rules (Revision, Modus Ponens, Deduction, Induction/Abduction, Negation, Inversion, Similarity, Evaluation, and more). Runs on PeTTa via git-import.
patham9/PLN	MeTTa	patham9	Operational	Fork preserving older modular src/ structure with build.sh and per-directory tests. Includes PLN.py Python integration.
ngeiswei/PLN	MeTTa	ngeiswei	Operational	Fork preserving older modular layout. Near-identical to patham9/PLN.
pln-experimental	MeTTa / LaTeX	trueagi-io	Experimental	Research platform with nuPLN formalization, multiple rule representations, ETV support, program synthesis.
hyperon-pln	MeTTa / Idris2	trueagi-io (local clone tracks patham9 fork)	Experimental	Four rule implementation approaches (dependent types, match, entail, equality). Includes generic program synthesizer.
pln (legacy)	C++ / Scheme	opencog	Legacy	Original C++/Scheme PLN on URE. Superseded by MeTTa PLN. Not in local collection — reference only.

Chaining and Inference Control

Repo	Language	Upstream	Maturity	Purpose
chaining	MeTTa	trueagi-io	Operational	DTL-based forward/backward/polyward chaining, program synthesis, Metamath proofs, modal logic. Core dependency for hyperon-miner and pln-experimental.
ure (legacy)	C++ / Scheme	opencog	Legacy	Unified Rule Engine with Thompson sampling. Superseded by MeTTa chaining. Required by legacy miner.
unify (legacy)	C++ / Scheme	opencog	Legacy	AtomSpace expression unifier. Required dependency for URE and legacy miner (cogutil → atomspace → unify → ure).
tv-toolbox	Haskell	opencog	Experimental	PLN truth value validation platform: STV/DTV/ITV conversions, deduction formulas, inference formula testing.
mm2metta	Python	ngeiswei	Operational	Metamath-to-MeTTa converter enabling formal proof verification. Used by chaining/experimental/metamath.

MOSES (Evolutionary Program Learning)

Repo	Language	Upstream	Maturity	Purpose
metta-moses	MeTTa + Python	iCog-Labs-Dev	Operational	Full MOSES reimplementation: demes, metapopulation, knob representation, 6 optimizers, scoring, feature selection, ENF.
hyperon-moses	MeTTa	trueagi-io	Experimental	Boolean reduction core — ENF simplification as foundation for evolutionary search. Design docs with cognitive synergy vision.
moses (legacy)	C++	opencog	Legacy	Original MOSES v3.7.0. Combo language output. Optional MPI for clusters. Requires cogutil, Boost.
asmoses (legacy)	C++	opencog	Legacy	AtomSpace-integrated MOSES. Atomese/Python output. Commercial use history (medical/financial). Requires cogutil, atomspace, ure.
MOSES-MORK	Python	iCog-Labs-Dev	Operational	Python MOSES targeting MORK/Hyperon: factor-graph EDA, ENF reducer, quantale-based variation operators.
elegant-normal-forms-python	Python	iCog-Labs-Dev	Operational	ENF satisfiability checker. Supporting tool for MOSES-MORK's boolean reduction pipeline.

Pattern Mining

Repo	Language	Upstream	Maturity	Purpose
hyperon-miner	MeTTa / Prolog	trueagi-io	Operational	Three implementations (Prolog, match-based, dependent-types). Dependent-types version integrates with chaining module.
miner (legacy)	C++ / Scheme	opencog	Legacy	URE-based frequent/surprise subhypergraph mining. Requires cogutil → atomspace → unify → ure build chain.
hyperon-miner (iCog)	MeTTa / Prolog	iCog-Labs-Dev	Operational	iCog Labs fork extending base hyperon-miner with expanded validation suites, surprisingness metrics, and integrated pipeline.

NARS (Non-Axiomatic Reasoning)

The NARS ecosystem has enough repos and conceptual coherence to warrant its own family card. See NARS Ecosystem for metta-nars, OpenNARS-for-Applications, estream, Pars, and NASTL.

How They Fit Together

PLN ecosystem: PLN (trueagi-io) provides the consolidated rule library. The patham9/PLN and ngeiswei/PLN forks preserve an older modular layout — useful for comparing structural approaches but not the canonical path. pln-experimental and hyperon-pln are research frontiers exploring different rule representations. The chaining repo provides the inference engine that PLN rules run on — it is a shared dependency.

MOSES ecosystem: Three generations coexist. metta-moses (iCog Labs) is the most complete MeTTa version and the recommended starting point. hyperon-moses (trueagi-io) focuses narrowly on boolean reduction. Legacy C++ repos (moses, asmoses) remain for reference and commercial backward compatibility.

Mining ecosystem: hyperon-miner's dependent-types implementation depends on the chaining module, creating a direct PLN→mining integration path. The legacy miner requires the full OpenCog C++ stack (cogutil → atomspace → unify → ure).

Quick Start

# PLN on PeTTa (primary path)
cd PeTTa && sh run.sh ./repos/PLN/examples/Smokes.metta

# Chaining tests
metta dtl/tests/test-forward-chaining-curried.metta

# metta-moses (full MOSES on MeTTaLog)
mettalog deme/tests/expand-demes-test.metta

# Pattern mining (match-based)
cd hyperon-miner/match && bash MatchTest.sh

# Legacy C++ MOSES
cd moses && mkdir build && cd build && cmake .. && make -j4

Current State vs. Whitepaper

PLN on MORK (whitepaper §6.1): Under development. Both backward chaining (PathMap indices) and factor-graph belief propagation are described but not yet operational natively on MORK.
GEO-EVO (whitepaper §6.2.1): Proposed. The geodesic evolutionary extension to MOSES exists only as whitepaper design — no implementation yet.
Quantale-annotated factor graphs (whitepaper §5.1): Proposed. The PLN-quantale integration is theoretical.
Context Quantaloid unification of PLN and NARS (whitepaper): Proposed. PLN and metta-nars are separate implementations with no shared algebraic framework yet.
MORK-native pattern mining (whitepaper §6.4): Described but not yet implemented. Current hyperon-miner runs on PeTTa, not directly on MORK's trie structures.

Forks and Mirrors

PLN forks: trueagi-io/PLN is canonical. patham9/PLN and ngeiswei/PLN preserve the older modular src/ structure.
chaining forks: trueagi-io/chaining is canonical. Local mirrors track rTreutlein/chaining and ngeiswei/chaining.
hyperon-miner forks: trueagi-io/hyperon-miner is canonical. A local mirror tracks the iCog-Labs-Dev fork with experimental validation suites.
ngeiswei mirrors: ngeiswei maintains mirrors of several repos: pln-experimental, hyperon-moses, miner, and ure. These are snapshot forks, not divergent development lines.

Recommended Entry Points

PLN reasoning: Start with PLN on PeTTa — run the Smokes or categorical reasoning examples.
Chaining/inference: Work through chaining's DTL test files (forward, backward, iterative).
Evolutionary learning: Use metta-moses on MeTTaLog — it has the most complete MOSES feature set in MeTTa.
Pattern mining: Start with hyperon-miner's match-based approach — the simplest of the three implementations.

Gaps and Consolidation Opportunities

PLN fork proliferation: PLN/PLN2/PLN3 diverge structurally. A convergence strategy would help — the modular layout may be worth adopting in the primary repo.
Two MOSES MeTTa ports: hyperon-moses (trueagi-io, boolean reduction only) and metta-moses (iCog Labs, full stack) overlap on ENF. Consolidation would reduce confusion.
No integrated PLN+MOSES loop: The cognitive synergy between PLN and MOSES described in the whitepaper has no working demonstration yet.
Legacy C++ repos require deep dependency chains: Building the C++ miner requires 4 prior installs (cogutil → atomspace → unify → ure). This blocks casual exploration.