SHADOW topology specialist for the Abraxas model stack. Route atoms in. yggdrasil.route.v0 out.
Named for the world tree: namespaces, edges, and gates. This repo classifies topology. It does not replace the runtime spine in Abraxas-v2.0.
| Owns | route_class · scan_class · integrity · unknown-node gate |
| Honesty | OBSERVED / INFERRED / SPECULATIVE / NOT_COMPUTABLE |
| Shape | Spec 001–004 T0 rules live. Encoder name_gate false. Not a chatbot. |
| Anti | promotion · forecast mint · chat trunk · Hyperlex form · Athanor family · Semion triad · settled Brier |
| Lane | SHADOW — classify live · no train weights at T0 · Hub closed |
| Router | Spec 009 R009 / R010. Dispatch also classifies R011 construct packets (no Sigil-Forge import). Forecast stays R004 NOT_COMPUTABLE. |
| Version | 0.1.0 (see STATUS.md) |
| Ships now | Does not ship |
|---|---|
T0 classify / adapt / compat / dispatch |
Learned router / train weights |
| Eval mapping v2 builder + tests (106 rows when built) | Populated eval_mapping_v2.jsonl committed as gold (emit stub only) |
| Spec 009 bind flipped (R009/R010) | Replacement of Abraxas-v2.0 core/yggdrasil/ |
| Hero + OG rasters | Auto Settings social-preview write |
Train? Not for the current job. Yggdrasil is a deterministic route classifier. Eval gold checks classify honesty; it is not a training corpus. A learned router would be a later product decision.
| Area | State |
|---|---|
| Spec 000–004 | Specified + code |
| Encoder | name_gate false |
| Hub | closed |
| Spec 009 router bind | flipped (R009 / R010) on Abraxas-v2.0 |
| R011 construct classify | shipped in this repo's dispatch; no Sigil-Forge import |
Replaces Abraxas-v2.0 core/yggdrasil/ |
false |
| Eval mapping v2 | builder + tests; committed file is emit stub |
| Tests | 35 passed (pytest -q tests/) |
| When built | 106 rows (20 pipeline / 86 node) · R009/R010/R004 |
Live spine code stays in scrimshawlife-ctrl/Abraxas-v2.0 core/yggdrasil/. This specialist does not hard-import that tree.
Bind record on v2.0 main: specs/012-yggdrasil-topology/BIND.md (PR #1139). Local receipt: BIND_RECEIPT.md.
Raster card: assets/og-social.jpg (1280×640) · also assets/og-social.png.
Settings → Social preview is manual (no API/MCP).
spine / route packet → adapt() → atom → classify() → yggdrasil.route.v0
↓
dispatch()
Bare RUNE.YGGDRASIL.* (no namespace, no claimed lane) → runtime_gate + ALIGNED.
When a pipeline already has a Spec 012 lane, YGGDRASIL tokens set scan only (no fake runtime_gate conflict).
See fixtures/seed/README.md.
| Spec | Role | Path |
|---|---|---|
| 000 spine | Constitution | specs/000-yggdrasil-spine/ |
| 001 classifier | T0 route classify | specs/001-route-classifier/ |
| 002 adapt | Spine → atoms | specs/002-spine-adapt/ |
| 003 export | Compat export | specs/003-export/ |
| 004 dispatch | Shadow dispatch | specs/004-shadow-dispatch/ |
| Doc | Path |
|---|---|
| Status | STATUS.md |
| Bind receipt | BIND_RECEIPT.md |
| Constitution | constitution.md |
| Model card | MODEL_CARD.md |
| Seed / eval | fixtures/seed/README.md |
| Contributing | CONTRIBUTING.md |
| Security | SECURITY.md |
pip install -e ".[dev]"
python -m yggdrasil --version
python -m yggdrasil classify '{"namespace":"yggdrasil.ingest"}'
python -m yggdrasil adapt '{"name":"universal_ingest"}'
python -m yggdrasil dispatch '{"namespace":"yggdrasil.ingest"}'
pytest -q
# emit eval mapping locally (not committed):
PYTHONPATH=src python fixtures/seed/build_eval_mapping_v2.py| Layer | Where | In git? |
|---|---|---|
| Builder | fixtures/seed/build_eval_mapping_v2.py |
Yes |
| Emit stub | fixtures/seed/eval_mapping_v2.jsonl |
Yes (comments only) |
| Seed atoms | fixtures/seed/atoms.jsonl |
Yes (tiny smoke) |
| Built mapping | same path after builder/pytest | Local rewrite only |
Gold route_class is Spec 012 only. v2.0 node.lane is never a class. NAME_TO_LANE comes from yggdrasil.adapt.
Do not without Danny/operator yes:
- Treat this repo as a drop-in replacement for Abraxas-v2.0
core/yggdrasil/ - Flip
name_gateor open Hub for a learned router - Commit large populated eval dumps as if they were training weights
- Forecast mint, promotion, or phenomenal claims
Hyperlex (form / lexical) · Athanor (tradition structure) · Semion (sign relation) · Yggdrasil (route topology) · Abraxas-v2.0 (runtime spine)
Code: MIT unless otherwise noted. See repo license files.