Skip to content

CLAUDE.md: inherited design is a hypothesis, not terrain #1054

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
InauguralPhysicist merged 1 commit into from
Aug 28, 2026
Merged

CLAUDE.md: inherited design is a hypothesis, not terrain #1054

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
31 changes: 31 additions & 0 deletions CLAUDE.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,3 +172,34 @@ model — don't work around a gap, surface it).
- **Infra**: **eigen-site** (inauguralsystems.com landing + the self-owned HTTP
attack target), **homebrew-eigenscript** (tap), **eigs-package-template**,
**awesome-eigenscript** (curated index — a list, not a registry).

## This design is pre-v1 and mostly written by older models — question it

**90% or more of this code was designed by Claude sessions running models
many generations old.** Newer models keep arriving that are substantially
stronger, the project is **pre-v1**, and there are no external consumers to
break. So a decision you find in the tree carries **no authority from
seniority** — it is a hypothesis an earlier session committed, and it is
fair game.

The tell is writing, or thinking, *"X must be true because the design does
Y."* Stop there and ask whether Y is a **law** — language semantics, a
physical constant, an external contract — or a **decision**. If it is a
decision, price the alternative before designing around it.

Bought 2026-08-28 (ouroboros#127 / DMG). The AOT compiles the main file to
C but emits `load_file` as a runtime call, so loaded modules are parsed and
interpreted by the linked VM. A real bug lived in that seam and was found,
minimised, fixed and verified. It was also reported as "unlocking the AOT
multiplier for DMG" — until the design itself was questioned. Measured:
DMG is 3,288 lines, of which 818 are compiled and 2,470 interpreted,
including the 837-line, 128-function opcode dispatch. Every emulated
instruction dispatches into interpreted code, so the fix makes DMG *run*
and cannot make it meaningfully *faster*.

An entire investigation cycle had treated "loaded modules are interpreted"
as terrain, including rejecting the alternative as unbuildable. The
capability to do it the other way already existed elsewhere in the same
tree (`obs_gate_resolve_static_loads`, used by the #915 observer gate).
**When a design looks forced, check whether the alternative is already
implemented somewhere else for another purpose.**
Loading