The Reasoning Chain

A real, verified pipeline for tracking research state — not narration in a chat, a callable chain. Four links, three of which have been run with real output shown, not just described.

PurposeThe Observe → Research → Decide chain that answers "where is this specific piece of work right now" and "what's the one highest-value next step" as callable functions instead of judgment calls buried in conversation.

StatusVerified — 3 of 4 links executed directly with real output; the fourth (decision kernels) has one component blocked on a missing dependency file

Built fromverification_registry.py, observation_kernels.py, research_state_kernel.py, decision_kernels.py

Superseded by—

EvidenceReal execution output on each linked page below.

The verified chain

Verification Registry Tracks which invariants each claim was checked against — the measurement layer everything below reads from.

↓

Observation Kernels Turns registry data into a real evidence-completeness number. Run and confirmed.

↓

Research State Kernel Classifies a work item into one of 8 real stages from those numbers. Run and confirmed on 3 real cases.

↓

Decision Kernels Turns a stage into a graduated action — not a menu, one recommendation.

What this chain is not

This is not the whole of "General Reasoning Operating System" as originally named — that claim listed six layers as equally "already-tested." Only the four in this chain have been directly verified here. Front Gate, Multi-Stage Pipeline, and governance selection are real per other documents but haven't been executed in front of a reader on this site yet — see the Master Kernel Library for the honest, current status of everything, not just this chain.

Exploratory — real, but not part of the verified chain

Kept visibly separate on purpose — these are real files, but their own docstrings say they're not something else depends on yet. Mixing them into the chain above would bury what's actually verified under what's still being played with.

planner_fuzz_harness.py

1000-trial archetype-sampling fuzz test for the execution-kernel planner. Depends on ncfca_execution_kernel_v4.py, which hasn't been verified on this site — described here, not independently executed.

multi_engine_hyperweave_playground.py

Its own docstring: "explicitly not production code... play with it, break it, see what's interesting — not a component anything else depends on." Left exactly that way here.