docs: add START_HERE.md entry point

Triage-shaped reading path. Four reasons newcomers end up in this
repo (current research, run sparse-parity, build a multi-agent
catalog, contribute a finding) each get a 3-line answer pointing
to the right docs or the right repo. Reduces the "what do I read
first" problem without requiring any file migration.

Companion to #96 (the reshuffle).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: 0bserver07

START_HERE.md

@@ -0,0 +1,34 @@
+# START HERE
+
+Four reasons people end up in this repo. Pick yours.
+
+## 1. "What's Sutro Group working on right now?"
+
+- [docs/research/active-threads-2026-05-09.md](docs/research/active-threads-2026-05-09.md) — who's on what this week
+- [DISCOVERIES.md](DISCOVERIES.md) — what's already proven
+- [docs/tasks/INDEX.md](docs/tasks/INDEX.md) — open tasks
+
+## 2. "I want to run a sparse-parity experiment"
+
+Wrong repo. The submission pipeline is at [cybertronai/sparse-parity-challenge](https://github.com/cybertronai/sparse-parity-challenge). The benchmark code in `src/sparse_parity/` here is being moved out (see [#96](https://github.com/cybertronai/SutroYaro/issues/96)).
+
+## 3. "I want to build my own multi-agent research catalog"
+
+The pattern shipped twice: [hinton-problems](https://github.com/cybertronai/hinton-problems) (53 stubs) and [schmidhuber-problems](https://github.com/cybertronai/schmidhuber-problems) (58 stubs).
+
+- `BUILD_NOTES.md` in either repo above — token math, agent-team pattern, lessons learned
+- [.claude/skills/](.claude/skills/) — the dispatcher skills
+- [docs/related-repos.md](docs/related-repos.md) — where each repo in the org fits
+
+## 4. "I want to contribute a finding"
+
+- [findings/_template.md](findings/_template.md) — the template
+- [CONTRIBUTING.md](CONTRIBUTING.md) — full guide
+
+## What this repo isn't
+
+- Not the active research front. That's at [ByteDMD/experiments/grid](https://github.com/cybertronai/ByteDMD/tree/dev/experiments/grid).
+- Not the home for benchmark problems. Those have dedicated repos.
+- Not the cost metric. That's [ByteDMD](https://github.com/cybertronai/ByteDMD) and [simplified-dally-model](https://github.com/cybertronai/simplified-dally-model).
+
+Longer version: [README.md](README.md) and [docs/related-repos.md](docs/related-repos.md). The reshuffle to match the file tree to this scoped role is [#96](https://github.com/cybertronai/SutroYaro/issues/96).