A methodology for working with AI systems: the disciplines that keep the work on track as it develops through a loop with the model.
Lodestone names how disciplined work with AI fails, and the practice that catches each failure before it ships. It is one of two open reference artifacts published by Clarethium:
- Touchstone validates work against quality standards.
- Lodestone orients practice.
Lodestone defines the components of disciplined work with AI systems: the stance, the loop, the calibration, the altitude, the failure shapes, the quality levels, the surface protocols, and the compound practice.
This is a reference, not a tutorial. Read it as you would read a field manual or a published standard. The audience is builders ready to adopt the practice.
Senior builders whose work develops continuously through a loop with AI. The quality of that loop determines the quality of the output. Engineers, researchers, designers, writers, analysts.
See OUTLINE.md for the manuscript table of contents.
A forward-looking annex on collective practice extends Lodestone's discipline into territory not yet validated by a pilot across multiple builders. The annex is provisional, held separately from the canonical sections, and evolves at its own pace as pilot evidence accumulates.
Reference integrations that deliver pieces of the methodology into your workflow are in tools/:
- A Claude Code
PreToolUsehook that surfaces Lodestone protocols at the moment of action (authentication surface, database surface, git write operations, deletion, Pattern Study triggers). Composes with cma's reference hook. - A cross-reference document mapping Lodestone concepts (Verified/Not verified, Quality theater, Assumption over verification, Default implementation) to Touchstone measurement layers, with runnable examples.
- A runnable
lodestone-scoreCLI that takes a work file and a source file, runs Touchstone, and emits a Lodestone score report with threshold-based concern flags (exit code 1 if any concept tripped, 0 if clean; usable in CI).
See tools/README.md for installation and details.
specimens/ seeds the recognition library that Klein's RPD lineage (Section V) calls for. Each specimen instantiates one failure shape in a form that lets an external reader recognize the pattern in their own work. The v1 corpus starts with one illustrative specimen as format anchor; abstracted-from-capture specimens contributed by builders running cma are the intended growth path. See specimens/README.md for format and contribution shape.
Lodestone composes with the other open Clarethium artifacts:
- Touchstone: third-person measurement of AI outputs. The substrate that pairs with Lodestone's first-person discipline.
- cma: executable compound-practice loop. The terminal-side instantiation of Lodestone's Section VIII.
The methodology lives in Lodestone. cma is what running that methodology looks like in a terminal.
CC-BY 4.0. See LICENSE.
L. Lucic. Published under Clarethium.