AURORA Stage-1 artefact lift 2026-05-02 — git 329cb110

README.md

@@ -9,40 +9,10 @@ pinned: true
 
 # Anthril
 
-**Anthril** is the research org publishing artefacts from the **AURORA** programme — a sparse, event-first, energy-bounded learning system whose Stage-1 falsifiability programme tests six pre-registered hypotheses (H1–H6) against a LoRA-tuned 8B-class transformer baseline.
+Independent AI research lab building **sparse, event-first, energy-bounded** learning systems.
 
-## What lives here
-
-| Repo | Type | Role |
-|---|---|---|
-| [`Anthril/aurora-workflow-1`](https://huggingface.co/datasets/Anthril/aurora-workflow-1) | Dataset | Rule grammars + spec for the primary Stage-1 structured-workflow dataset (60 000 generator-defined episodes across 6 domains). |
-| [`Anthril/lora-llama8b-aurora-baseline-2026-05-02`](https://huggingface.co/Anthril/lora-llama8b-aurora-baseline-2026-05-02) | Model | LoRA adapter for Meta-Llama-3-8B-Instruct, fit on AURORA-Workflow-1. The H1 baseline anchor (ADR-0033). |
-| [`Anthril/aurora-m0-fit-2026-05-02-pubgrade`](https://huggingface.co/Anthril/aurora-m0-fit-2026-05-02-pubgrade) | Model | Publication-grade AURORA-M0 fit snapshot (ADR-0037). All seven Stage-1 microfield archetypes activated. |
-
-## The H1 hypothesis (verbatim)
-
-> On structured workflow-apprenticeship tasks, an event-first AURORA-M0 model will achieve task success within 5 percentage points of a LoRA-tuned 8B-class transformer baseline while using **at least 40 % less measured energy per successful task**.
-
-The repos in this org are the **load-bearing artefacts** for that comparison.
-
-## Reproducibility posture
-
-Every artefact published here is a mirror of a content-addressed local copy whose SHA-256 manifest, git HEAD, push timestamp, and HF commit OID are recorded in an on-disk `hf-publish-manifest.json` sidecar. Local artefacts remain canonical per ADR-0020 (Backblaze B2 WORM) and are anchored in the AURORA Provenance Ledger (Ed25519-signed, Merkle-checkpointed every 10 000 entries).
-
-## Pre-registration
-
-The Stage-1 hypotheses are pre-registered on OSF:
-- Parent project: [u8wph](https://osf.io/u8wph/) — *AURORA: Adaptive Unified Resonance Organism for Relational Autonomy*
-- Stage-1 component: [jerwm](https://osf.io/jerwm/) — *Stage 1 - Falsifiability Programme (H1-H6)*
-
-## Replication invitation
-
-H1, H3, and H6 require independent two-platform replication. If you want to run the verdict on your hardware, see the LoRA repo's model card for the exact runner invocation. **Apple-Silicon hosts** (Mac mini M4 and similar) are publication-eligible per ADR-0043 (Apple Silicon `powermetrics` collector); **NVIDIA-equipped Linux/Windows hosts** are publication-eligible per ADR-0035 (NVML + psutil).
+We publish models, datasets, and reproducibility artefacts as open research — under permissive licenses where we can, and with full provenance trails so partners can replicate and audit our claims.
 
 ## License
 
-Artefacts are CC-BY-4.0 unless an upstream license applies (the LoRA adapter requires Meta's [Llama 3 Community License](https://llama.meta.com/llama3/license/) for its base model).
-
-## Citation
-
-If you use these artefacts, please cite the AURORA Stage-1 master pre-registration (OSF [u8wph](https://osf.io/u8wph/)) and any specific ADR(s) the artefact's model card cites (ADR-0033, ADR-0037, ADR-0035, ADR-0043).
+Artefacts are CC-BY-4.0 unless an upstream license applies.