The scale rung — find ripwire's next super-linear path on the largest public codebases

## What is missing

Some of ripwire's worst defects cannot be seen below a certain tree size. Nobody in this project
routinely runs it on the Linux kernel, Chromium or a large monorepo, so those defects go unmeasured
until someone stumbles into one. The job is to run the ladder deliberately: measure cold and warm
CPU carefully, profile what bends, and reduce each finding to a generated fixture with a red-first
ratio arm.

## The evidence

Two defects of exactly this shape have already been found and fixed, and neither was visible on a
standard corpus:

- **#127**: indexed tree-sitter child walks were O(children²). They cost llvm-project **a fifth of
its cold parse**.
- **#83**: a single quadratic dedup took **143 of 154 seconds** in one phase of an llvm `--grep`.

A carefully measured ladder that finds nothing is a valuable result too, and the kit says so.

The kit also carries the instrument-artifact checklist drawn from the 2026-09-10 audit's own
mistakes — a cache that evicted itself and faked super-linearity, and a cap sweep run over a
population of zero — because a measurement that fools its own author is the common failure here, not
a missed defect.

## Size

**Medium**, mostly machine time:

- **One rung measured and reported**: small to medium.
- **A defect found, with a fixture, a red-first ratio arm and a byte-identical fix**: medium.

The rungs are independent, so several people can take different trees.

## Prerequisites

Disk and cores — llvm-project alone is 2.9 GB, and Linux and Chromium are larger. A sampling
profiler (`perf` on Linux; `sample` or Instruments on macOS). For fixes: comfort reading C++ and
tree-sitter walks, and the ASan build.

## Where to start

`prompts/help-wanted/find-the-next-superlinear.md` is a self-contained prompt for a coding agent. It
carries the measurement protocol (pinned corpora, cold kept apart from warm, CPU rather than wall,
interleaved arms, a placebo arm, output identity, a quiet process table), how to find a super-linear
path (per-unit cost across a corpus ladder, a fixture that grows one dimension, an isolation pair),
how to rule out instrument artifacts, the no-timing-budget constraint, and a fill-in report format.
Like every prompt in `prompts/`, **it ends by writing a plan and stopping** — a maintainer agrees
the plan before any code is written.

Post the ladder report even if nothing bends.

Comment here to claim it, naming the corpora and host available.

Read prompts/help-wanted/find-the-next-superlinear.md first. Choose an available corpus and host, follow its cold/warm CPU measurement and artifact-checking protocol, then write the required ladder report. Done means a measured rung is reported, whether or not it reveals a super-linear path; any defect additionally needs a generated fixture, red-first ratio arm, and byte-identical fix.

Written by the indexing model from the issue text.