Org mirror note: this STATUS is grafted from personal tip for Spec 007 continuity. License SoT on this repo remains Proprietary (LICENSE). Train SoT remains personal scrimshawlife-ctrl/Hyperlex.
Naming: repo Hyperlex is the transitional monorepo shell. Public products: Hyperlexical (Spec 007 model / train / eval / E2 / name_gate claim) and ne0l0gist (slang ingest — operator-named 2026-09-24, repo spelling). name_gate true for seed-morph78 (Danny flip name_gate, 2026-09-24, amendment A6).
Version: 0.4.0
Observed: 2026-09-24
Posture: Hermes skill = current operator surface. Spec 007 = model path (SHADOW). Spark BEST = seed-morph78 (soft_ceiling PROMOTE_BEST 2026-09-24; broad 0.9883 n=256). Ingest product named ne0l0gist (as in repo). Naming lock: docs/NAMING.md + Notion Naming lock. #101 merged to main. Org naming graft #14 merged. morph78 settled have fun staying poor + fr fr no cap; force 236/hard 277. name_gate true for seed-morph78 (Danny flip name_gate, 2026-09-24, amendment A6). Card/package IDs stay hyperlex-encoder-*; Hub and T13 not authorized.
Install: bash install.sh → ~/.hermes/skills/hyperlex
Claude (optional): bash install.sh --claude → ~/.claude/skills/hyperlex
Track: Phases 0–4 complete · Phase 5.0–5.3 · Pages static run history · Spec 007 SHADOW encoder on main
This file is the operator snapshot. The docs site copies it to status. Do not treat it as a Hub card or a Brier score.
| Layer | Role | State |
|---|---|---|
| Hermes skill | What you run today (SKILL.md, CLI, src/hyperlex/) |
Ready (v0.4.0) |
| T0 | Encoder baseline; card hyperlex-encoder-* |
Specified. Not named Hyperlexical. |
| T1 | First artifact that may be called Hyperlexical | seed-morph78: trained E2 PASS + Danny name_gate yes (2026-09-24) — named Hyperlexical |
name_gate |
Name wall | true (seed-morph78, A6) · publish wall (Hub) still closed |
| Hub | Operator upload | Not published |
Classify volume is ready. Volume did not flip name_gate; Danny's sentence did. Seed smoke ≠ T1.
python3 scripts/hyperlex.py doctor
python3 scripts/release_preflight.py
python3 scripts/hyperlex.py simulate --term rizz --mode scenario
python -m hyperlex inbox list
PYTHONPATH=scripts/shadow python3 -m hyperlexical.infer --text rizz --offlineSHADOW / advisory. Not on API_V1. Pin seed-morph78 may be called Hyperlexical (A6). Do not call any other checkpoint, stub, or seed smoke Hyperlexical.
Operator scoreboard 2026-09-10 PT evening (Danny-locked; matches Notion Operator Hub):
| Surface | n | Notes |
|---|---|---|
Local SoT ~/.hyperlex/hyperlexical/ingest_candidates.jsonl |
4333 | 402 OBSERVED / 3931 INFERRED. Not in git. |
Export --include-live (operator machine) |
6506 | classify family 2437 · unbind 1345 · negatives 208 |
Tracked specs/007-hyperlexical-model/exports/civilian.v0.1.jsonl |
883 | Seed/snapshot only. Do not treat as the train SoT. |
Danny ~2500 candidate bar: met. Hermes 913 / “gap to 2500” is superseded — not current SoT status. Afternoon store family-labeled 1789 (stretch 2000 not reached) is the blanket-yes receipt figure; export classify 2437 is the harvest gate. Moltbook row counts (for example ~360 ai-native) are a Moltbook subset, not the global SoT.
| Gate | State |
|---|---|
| Classify volume | Ready — operator --include-live family classify 2437 (≥2k). name_gate gaps 0 / 0 / 0 on that surface. |
name_gate |
true for seed-morph78 — Danny flip name_gate 2026-09-24 (A6; receipt specs/007-hyperlexical-model/receipts/20260924-name-gate-yes-morph78.md). E2 PASS alone did not flip it. |
| Spark BEST | seed-morph78 — PROMOTE_BEST via soft_ceiling ceiling_escape: broad OBSERVED 0.9883 > morph65 0.8867 n=256; E2 PASS; force fair 1.0 n=164 (advisory). LAST=8. Upsample freeze 11+. |
| E2 vs Spec 004 | Stub still FAIL (expected). Trained trunk-forward E2 PASS on morph19 (unbind_exact=1.0). Seed smoke ≠ T1. |
| Hub publish | No — skeleton in-repo; weights stay on Spark. |
| T1 name | Hyperlexical approved for seed-morph78. Card/package IDs still hyperlex-encoder-* until a separate card-rename change. |
| Lineage families | 8 only. No ninth family. |
| Brier | null on every 007 packet. |
| Crawl | Crawl4AI 0.9.3 default. --source firecrawl aliases to crawl4ai. No paid Firecrawl without Danny yes. |
Spark trains from the local SoT / export --include-live, not from the tracked seed alone.
Spark procedure (bring-up, not a product card):
- SPARK-BRINGUP.md (#28)
- AARON-SPARK-TRAIN.md
- HERMES-SPARK-RUN.md
- A5 milestones / engineering (#33): milestones.md
- Live-split coerce (#38) is on
main(lexical_splitin the export path)
Pages overview: SHADOW encoder (007)
| Area | Status |
|---|---|
| Skill contract + install | Ready |
| Mock offline analyze | Ready |
| Lineage (8 families + 2026 YTD leaves) | Ready |
YTD backfill packs (data/backfill/2026/) |
Ready |
| Lineage backpropagation (non-mutating) | Ready |
| Typology + community drivers | Ready |
| Virality prediction (SPECULATIVE) | Ready |
| Receipts + ledger + ledger-stats/diff | Ready |
| Forecasts → settle → Brier series | Ready (settlement required) |
| Rune relay + market connectors | Ready |
| Diagrams from history | Ready |
| Case study runner | Ready |
| MkDocs + Pages (enabled) | Ready |
| Pages static run history | Ready |
| Long-term analysis archive | Ready |
| Governed LLM (echo / openai_compatible) | Opt-in |
| Phase 5 cultural transmission / multi-agent / risk / phylogeny | Ready (SPECULATIVE) |
| Local vector DB + Chroma promote | Ready |
| Mutation prediction | Ready (SPECULATIVE) |
| Hybrid lineage re-rank | Ready |
| Domain phylogeny packs | Ready |
| Transmission calibrate / scenario library | Ready |
| Risk → scan/cron schedule | Ready (advisory) |
| Ingest routes + automatic pipeline | Ready |
| Atomic multi-term seeds | Ready |
| Analysis enrichment (compression_metrics, typology tags, signal_report, integrity header) | Ready |
Local attractor store (~/.hyperlex/signals/) |
Ready (`inbox list |
Attractor candidate rune (RUNE.HLX.ATTRACTOR_CANDIDATE) |
Ready (advisory only) |
| Spec 007 model path (T0→T1) | SHADOW · Spark BEST morph78 · broad 0.988 n=256 · force fair 1.0 n=164 · E2 PASS · LAST=8 · upsample freeze 11+ · morph69–78 closed · next climb needs new named-phrase card · name_gate true (morph78, A6) · named Hyperlexical · no Hub · T13 not authorized |
| 007 live ingest tap | SHADOW · pipeline/analyze/scan fail-open → ~/.hyperlex/hyperlexical/ingest_candidates.jsonl · INFERRED only |
| Public PyPI | Not planned |
| External system hard import | Never |
pipeline "rizz" | run "rizz"
→ hyperlexical tap (INFERRED candidates)
→ pending → settle → score-series
→ scan / risk-schedule
→ relay --push-inbox
→ PYTHONPATH=scripts/shadow python3 -m hyperlexical.ingest_tap
→ inbox list
→ vector-seed / vector-sync
→ archive-export
007 Spark (Aaron, not the daily loop): specs/007-hyperlexical-model/AARON-SPARK-TRAIN.md
~/.hyperlex/receipts/
~/.hyperlex/receipt_ledger.jsonl
~/.hyperlex/score_log.jsonl
~/.hyperlex/mutation_watch.jsonl
~/.hyperlex/cache/
~/.hyperlex/vector.db
~/.hyperlex/chroma/
~/.hyperlex/signals/inbox.jsonl
~/.hyperlex/hyperlexical/ingest_candidates.jsonl
~/.hyperlex/models/ # Spark dumps only; not git
data/backfill/2026/
- Spark BEST = morph78. Default HOLD. Operator review
specs/007-hyperlexical-model/HYPERLEXICAL-PRODUCT-PLAN.md; a new climb needs a new named-phrase acquire card (compare vs morph78 broad 0.9883 n=256).name_gateyes recorded (A6); next are separate card-rename / Hub / T13 decisions. Plan:specs/007-hyperlexical-model/NAME-GATE-AND-NAMED-PHRASES.md. - Burn-in offline runs + settle path (this is how Brier becomes real).
- Do not Hub-upload. Do not promote
scripts/shadow/hyperlexical/intosrc/hyperlex/(T13) without a separate sentence. Do not extendname_gateto any checkpoint other thanseed-morph78.
Operator front door expanded for stack parity with Athanor / Semion / Yggdrasil (2026-09-11). Changelog-style dumps stay in CHANGELOG / receipts — not the main page.