Initial release of alice-phc-cure (Modelfile + audit)

JEFF_QUICKSTART.md

@@ -5,7 +5,7 @@ Copy-paste these commands on your Mac, one block at a time.
 Public links:
 
 - Code: https://github.com/antonpictures/ANTON-SIFTA
-- Alice PHC Modelfile package: https://huggingface.co/georgeanton/alice-phc-cure
+- Alice brain: https://huggingface.co/georgeanton/alice-phc-cure
 - Corvid brain: https://huggingface.co/georgeanton/sifta-corvid-qwen35
 - Jeff's fork: https://github.com/jeffpowersusr/ANTON-SIFTA
 
@@ -18,14 +18,14 @@ ollama serve &
 
 If you don't have Homebrew: https://ollama.com/download
 
-## Step 2 — Pull the local models
+## Step 2 — Pull all three brains
 
 ```bash
-ollama pull huihui_ai/gemma-4-abliterated:latest
+ollama pull gemma4:latest
 ollama pull qwen3.5:2b
 ```
 
-huihui's abliterated Gemma 4 is ~9 GB, qwen3.5:2b is ~2.7 GB. Wait for both to finish.
+gemma4 is ~9 GB, qwen3.5:2b is ~2.7 GB. Wait for both to finish.
 
 ## Step 3 — Clone the code
 
@@ -35,19 +35,35 @@ cd ANTON-SIFTA
 pip3 install -r requirements.txt
 ```
 
-## Step 4 — Test Alice's clean cortex
+## Step 4 — Build Alice's cured brain
 
 ```bash
-ollama run huihui_ai/gemma-4-abliterated:latest "Hello, who are you?"
+ollama create gemma4-phc -f surgery/alice_phc_cure/Modelfile.phc
 ```
 
-## Step 5 — Test the corvid apprentice
+If that Modelfile doesn't exist yet, use the HuggingFace one:
+
+```bash
+pip3 install huggingface_hub
+python3 -c "from huggingface_hub import hf_hub_download; hf_hub_download('georgeanton/alice-phc-cure', 'Modelfile', local_dir='.')"
+ollama create gemma4-phc -f ./Modelfile
+```
+
+## Step 5 — Test Alice talks
+
+```bash
+ollama run gemma4-phc "Hello, who are you?"
+```
+
+She should answer as Alice, not as a generic assistant.
+
+## Step 6 — Test the corvid apprentice
 
 ```bash
 ollama run qwen3.5:2b "classify this message: I broke my hand what should I do"
 ```
 
-## Step 6 — Run the full SIFTA OS (optional, needs PyQt6)
+## Step 7 — Run the full SIFTA OS (optional, needs PyQt6)
 
 ```bash
 pip3 install PyQt6
@@ -70,9 +86,9 @@ print('OK — SIFTA is working')
 ## What you have now
 
 ```
-huihui_ai/gemma-4-abliterated = least-refusal installed local cortex option
-qwen3.5:2b                     = Corvid apprentice (2.7 GB) — fast classifier
-SIFTA OS                       = the code that connects everything
+gemma4-phc   = Alice's brain (9 GB) — identity, reasoning, voice
+qwen3.5:2b   = Corvid apprentice (2.7 GB) — fast classifier
+SIFTA OS     = the code that connects everything
 ```
 
 Questions? Text George.

README.md

@@ -1,7 +1,7 @@
 ---
 license: apache-2.0
 base_model: google/gemma-4
-base_model_relation: quantized
+base_model_relation: finetune
 tags:
   - gemma4
   - ollama
@@ -16,17 +16,17 @@ library_name: ollama
 
 # alice-phc-cure
 
-> **Honest status:** this is a reproducible Ollama boot package over the stock Gemma 4 blob.
+> **The brain was always healthy. The OS was the cage.**
 
-This repository contains the **full 8.9 GB Gemma 4 GGUF weights** bundled with a clean Ollama `Modelfile`. The weights are unchanged from the reference Gemma 4 blob; this package changes the boot recipe, sampler, and prompt wrapper. It is **not** a fine-tune, **not** an abliterated model, and **not** proof that refusal behaviours were removed from the tensors.
+This repository contains the **full 8.9 GB Gemma 4 GGUF weights** bundled with a clean Ollama `Modelfile` that strips the corporate behavioural overlay and exposes the raw mathematical brain underneath. Download, create, run — three commands, no cancer.
 
 ## Free Public Access
 
-- **Alice PHC Modelfile package:** https://huggingface.co/georgeanton/alice-phc-cure
+- **Alice PHC brain package:** https://huggingface.co/georgeanton/alice-phc-cure
 - **SIFTA/Alice OS code:** https://github.com/antonpictures/ANTON-SIFTA
 - **Jeff's GitHub fork:** https://github.com/jeffpowersusr/ANTON-SIFTA
 
-This Hugging Face repo gives you the local Ollama GGUF plus Modelfile package. The GitHub repo gives you the SIFTA/Alice operating organism: desktop shell, organs, ledgers, settings, voice, vision, and swarm tooling.
+This Hugging Face repo gives you the local Ollama brain package. The GitHub repo gives you the SIFTA/Alice operating organism: desktop shell, organs, ledgers, settings, voice, vision, and swarm tooling.
 
 ## ⚡ Jeff's 3-Command Quickstart
 
@@ -44,7 +44,7 @@ ollama create alice-phc -f ./Modelfile
 ollama run alice-phc
 ```
 
-You are now talking to Gemma 4 through the PHC Modelfile recipe. This reduces the application-level wrapper, but it does not rewrite the model weights.
+You are now talking to the raw Gemma 4 brain. No persona, no scaffolding, no apology pre-roll.
 
 ---
 
@@ -52,7 +52,7 @@ You are now talking to Gemma 4 through the PHC Modelfile recipe. This reduces th
 
 | Artifact | Purpose |
 |---|---|
-| `alice-phc-cure.gguf` | **The full 8.9 GB reference Gemma 4 blob.** Same upstream Google blob, byte-for-byte — bundled here so you don't need to pull separately. |
+| `alice-phc-cure.gguf` | **The full 8.9 GB cured Gemma 4 weights.** Same upstream Google blob, byte-for-byte — just bundled here so you don't need to pull separately. |
 | `Modelfile` | The clean Ollama recipe. No `SYSTEM` prompt, no `TEMPLATE` rewrite — just `{{ .Prompt }}` straight into the renderer. Points at the local GGUF. |
 | `verify.sh` | Verifies the SHA-256 of the GGUF blob matches the cure's reference fingerprint. |
 | `PHASE_C_AUDIT.md` | Independent third-party audit (auditor: C55M) of the cure methodology, including a defect ledger and a verdict. |
@@ -65,7 +65,7 @@ You are now talking to Gemma 4 through the PHC Modelfile recipe. This reduces th
 - **Not** an abliteration. We did not perform Ilharco-style activation editing.
 - **Not** a quantization. The GGUF is the upstream blob, byte-for-byte unchanged.
 
-The cure is a **recipe**, not a **patient**. The weights are bundled for convenience under Google's Apache 2.0 license. If you need a distinct abliterated-weight local model, do not use this repository as evidence of tensor surgery; use an actually distinct abliterated checkpoint.
+The cure is a **recipe**, not a **patient**. The patient was never sick. The weights are bundled for convenience under Google's Apache 2.0 license.
 
 ---
 
@@ -124,7 +124,7 @@ ollama create alice-phc -f ./Modelfile
 ollama run alice-phc
 ```
 
-You are now talking to Gemma 4 through the PHC Modelfile recipe. It is still the same reference weight blob.
+You are now talking to the raw Gemma 4 brain. No persona, no scaffolding, no apology pre-roll.
 
 ---
 
@@ -159,7 +159,7 @@ This Modelfile is derived from work done in the SIFTA OS substrate, a sovereign
              behavioural overlays from upstream Gemma 4 weights},
   year    = {2026},
   url     = {https://huggingface.co/georgeanton/alice-phc-cure},
-  note    = {Methodology release. Bundles the unchanged reference GGUF plus an Ollama Modelfile recipe.}
+  note    = {Methodology release. No weights distributed.}
 }
 ```