methodology

Files about how we work — reusable procedure, process discipline, the craft of running hard projects. Process is craft, not science; these are living playbooks that accrete.

• Method — a library of good procedure — the collection
• The Anchor Method — drift is the enemy; the gate is the anti-drift anchor that proves grounding; install the strongest anchor your grounding allows; the director runs a recalibration meta-check via divergent-approach tournaments
• Specimen & Thesis — The Two-Altitude Ledger — separate the dated source-anchored specimen from the portable thesis it feeds; connect with a two-way ledger whose downward half (thesis→specimens) rots; promote harvests, don’t trap them, and don’t let auto-memory substitute for the page
• Author Web, Derive Hierarchy — generative work discovers its structure, so keep the substrate write-optimized (web) and derive the read-optimized hierarchy (indexes, ledgers, audit) with tooling; a tree severs lateral radiation, the web holds it; difficulty of scope = cost of compounding
• The Dominance-Frontier Lens — dominance pruning as a reducer; the analytical frame for competitive systems (games + markets) — map cost vs effect, read the frontier off the graph
• Repairing LLM Code — The Two Oracles — generator/judge independence; calibrate deterministic rewrite rules with a transfer-oracle instead of optimizing a goto-count proxy
• The Rosetta Stone Method — learn new hardware by anchoring the unknown machine against a portable program you already understand (KOEI’s cross-platform VM as the exemplar)
• Absolutes and Differentials — Frame Selection as the Political Crux — the 4-step diagnostic for arguments that will not resolve despite both sides citing accurate figures — check level-vs-gap first, argue the numbers last.
• Watching the Solver Learn — objective/reward bugs are BEHAVIOURAL: nothing crashes, no test fails, metrics climb, agent useless. Build the observation layer (live view + training telemetry) before the tuning loop; run the curriculum with a from-scratch control; a benchmark the agent can shrink isn’t measuring its label
• Media Pipeline — (draft) controlled md→video harness; three-specimen thesis that output quality tracks HARNESS quality more than model quality, plus a boundary table naming what stays hand-run
• Vault Heartbeat — 2026-08-26
• Building Swarms — Token Rules as Architecture — Nate B.
• The Substrate Is the Governing Mechanism — Agents coordinating through a shared store is read as the thing that makes them ungovernable.
• Vault Multi-Machine Sync — Branching, Divergence, and Merge Resolution — The vault runs on two machines and commits straight to master on both.
• Comments and the Distance to an Oracle — the fixed-context A/B that repairs the transfer-fidelity test by adding a cost term
• The Contract Model vs. the Substrate Model — a per-delivery contract (delivery / constraints+guardrails / proof artifact / outside-verification / owner) is the vault’s architecture with the accumulation stripped out — contracts don’t compound, substrate does; plus the artifact-vs-reader split in verification independence, and tests written only to pass as oracle collapse
• Oracles Are Objective Functions — supplies the missing retire clause: a standard decays in proportion to the pressure on it, so audit checks by who can move them, and read all-pass as the decay signal