Upload 5 files

README.md

@@ -1,17 +1,9 @@
----
-title: Paradigms
-emoji: 💬
-colorFrom: yellow
-colorTo: purple
-sdk: gradio
-sdk_version: 5.42.0
-app_file: app.py
-pinned: false
-hf_oauth: true
-hf_oauth_scopes:
-- inference-api
-license: cc-by-nc-4.0
-short_description: building a space for paradigm tests
----
+# ACE–CPT Context Agent (Hugging Face Space)
 
-An example chatbot using [Gradio](https://gradio.app), [`huggingface_hub`](https://huggingface.co/docs/huggingface_hub/v0.22.2/en/index), and the [Hugging Face Inference API](https://huggingface.co/docs/api-inference/index).
+A minimal ACE loop (Generator → Reflector → Curator) with CPT add‑ons.
+
+## Run locally
+```
+pip install -r requirements.txt
+python app.py
+```

app.py

@@ -1,70 +1,165 @@
+
 import gradio as gr
-from huggingface_hub import InferenceClient
-
-
-def respond(
-    message,
-    history: list[dict[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-    hf_token: gr.OAuthToken,
-):
-    """
-    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
-    """
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
-
-    messages = [{"role": "system", "content": system_message}]
-
-    messages.extend(history)
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
-        temperature=temperature,
-        top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
-
-        response += token
-        yield response
-
-
-"""
-For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
-"""
-chatbot = gr.ChatInterface(
-    respond,
-    type="messages",
-    additional_inputs=[
-        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
-        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
-        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
-        gr.Slider(
-            minimum=0.1,
-            maximum=1.0,
-            value=0.95,
-            step=0.05,
-            label="Top-p (nucleus sampling)",
-        ),
-    ],
-)
-
-with gr.Blocks() as demo:
-    with gr.Sidebar():
-        gr.LoginButton()
-    chatbot.render()
+import yaml, pandas as pd, os, re
+
+PLAYBOOK_PATH = "playbook.yaml"
+CEM_PATH = "cem.csv"
+
+def load_playbook():
+    if not os.path.exists(PLAYBOOK_PATH):
+        return {"playbook": {"metadata": {"name": "ACE–CPT Playbook", "version": "0.1"}, "items": []}}
+    with open(PLAYBOOK_PATH, "r", encoding="utf-8") as f:
+        return yaml.safe_load(f) or {}
+
+def save_playbook(pb):
+    with open(PLAYBOOK_PATH, "w", encoding="utf-8") as f:
+        yaml.safe_dump(pb, f, sort_keys=False, allow_unicode=True)
+    return True
+
+def ensure_cem():
+    if not os.path.exists(CEM_PATH):
+        pd.DataFrame(columns=["ClaimID","Claim","EvidenceRefs","Counter-Evidence","Confidence"]).to_csv(CEM_PATH, index=False)
+    return pd.read_csv(CEM_PATH)
+
+def save_cem(df):
+    df.to_csv(CEM_PATH, index=False)
+    return df
+
+def run_generator(task, inputs, playbook_text):
+    # Stubbed generator
+    trace = [
+        f"S01: Restate task → {task[:120]}",
+        "S02: Identify relevant playbook items → PB-001, PB-004",
+        "S03: Apply heuristic reasoning (stub)",
+        "S04: Prepare answer with uncertainty note"
+    ]
+    answer = "[Demo] Task addressed. (Replace with real model call.)"
+    used_items = "PB-001, PB-004"
+    cpt_notes = "2.11–2.17 quick pass; Flags: anthropomorphization=false, spec_gaming=possible, uncertainty=high"
+    return answer, "\n".join(trace), used_items, cpt_notes
+
+def run_reflector(task_id, trace_text):
+    import re, yaml as _yaml
+    steps = re.findall(r"S(\d+):", trace_text)
+    evidence_steps = [f"S{sid}" for sid in steps[-2:]] if steps else []
+    deltas_yaml = {
+        "deltas": [{
+            "op": "add",
+            "proposed": {
+                "type": "checklist",
+                "title": "Preflight task restatement",
+                "content": "Before solving, restate the task in one line and list 2–3 success criteria plus one salient risk.",
+                "tags": ["meta","stillness"],
+                "evidence": {
+                    "task_id": task_id or "T-000",
+                    "trace_steps": evidence_steps,
+                    "tests": [],
+                    "sources": []
+                },
+                "confidence": "HIGH",
+                "paradigm_scope": ["method"],
+                "closure_flags": {"anthropomorphization": False, "spec_gaming": False, "monoculture": False},
+                "cem": {
+                    "claim_id": "C-NEW",
+                    "claim_text": "Preflight improves clarity and reduces failure risk.",
+                    "evidence_refs": evidence_steps,
+                    "counterevidence_refs": []
+                }
+            },
+            "rationale": "Reusable checklist that generalizes across tasks."
+        }]
+    }
+    return _yaml.safe_dump(deltas_yaml, sort_keys=False, allow_unicode=True)
+
+def run_curator(playbook_text, deltas_text):
+    try:
+        pb = yaml.safe_load(playbook_text) if playbook_text.strip() else load_playbook()
+        if not pb or "playbook" not in pb:
+            pb = load_playbook()
+    except Exception:
+        pb = load_playbook()
+
+    try:
+        dz = yaml.safe_load(deltas_text) if deltas_text.strip() else {"deltas": []}
+    except Exception as e:
+        return playbook_text, f"YAML parse error in deltas: {e}"
+
+    items = pb.get("playbook", {}).get("items", [])
+    existing = {(it.get("title",""), it.get("type","")) for it in items}
+    next_id = 1
+    for it in items:
+        try:
+            n = int(str(it.get("id","PB-0")).split("-")[-1])
+            next_id = max(next_id, n+1)
+        except Exception:
+            pass
+
+    diff_lines = []
+    for d in dz.get("deltas", []):
+        if d.get("op") != "add":
+            continue
+        prop = d.get("proposed", {})
+        key = (prop.get("title",""), prop.get("type",""))
+        if key in existing:
+            diff_lines.append(f"~ duplicate skipped: {key[0]} ({key[1]})")
+            continue
+        new_item = {
+            "id": f"PB-{next_id:03d}",
+            "type": prop.get("type","heuristic"),
+            "title": prop.get("title","Untitled"),
+            "content": prop.get("content",""),
+            "tags": prop.get("tags",[]),
+            "helpful": 0,
+            "harmful": 0
+        }
+        items.append(new_item)
+        existing.add(key)
+        diff_lines.append(f"+ {new_item['id']}: {new_item['title']} ({new_item['type']})")
+        next_id += 1
+
+    pb["playbook"]["items"] = items
+    save_playbook(pb)
+    new_text = yaml.safe_dump(pb, sort_keys=False, allow_unicode=True)
+    return new_text, "\n".join(diff_lines) if diff_lines else "No changes applied."
+
+def cem_load():
+    return ensure_cem()
+
+def cem_save(df):
+    return save_cem(df)
+
+with gr.Blocks(title="ACE–CPT Context Agent") as demo:
+    gr.Markdown("# ACE–CPT Context Agent\nA minimal ACE loop (Generator → Reflector → Curator) with CPT add‑ons.")
+    with gr.Row():
+        task = gr.Textbox(label="Task", placeholder="Describe the task…")
+        task_id = gr.Textbox(label="TaskID", value="T-001")
+    inputs = gr.Textbox(label="Inputs (optional)", lines=6, placeholder="Paste notes, success criteria, risks…")
+    playbook = gr.Textbox(label="Playbook (YAML)", lines=18, value=open(PLAYBOOK_PATH, "r", encoding="utf-8").read())
+
+    with gr.Row():
+        gen_btn = gr.Button("Run Generator")
+        ref_btn = gr.Button("Run Reflector")
+        cur_btn = gr.Button("Run Curator")
+        save_pb_btn = gr.Button("Save Playbook")
+
+    answer = gr.Textbox(label="Answer (Generator)", lines=3)
+    trace = gr.Textbox(label="Trace (Generator)", lines=8)
+    used = gr.Textbox(label="UsedItems (Generator)", lines=1)
+    cpt = gr.Textbox(label="CPT Quick Notes (Generator)", lines=6)
+
+    deltas = gr.Textbox(label="Reflector Deltas (YAML)", lines=10)
+    diff = gr.Textbox(label="Curator Merge Log", lines=4)
+
+    gr.Markdown("## Claim–Evidence Matrix (CEM)")
+    cem_dataframe = gr.Dataframe(value=cem_load(), interactive=True, wrap=True)
+    with gr.Row():
+        cem_save_btn = gr.Button("Save CEM")
 
+    gen_btn.click(run_generator, [task, inputs, playbook], [answer, trace, used, cpt])
+    ref_btn.click(run_reflector, [task_id, trace], deltas)
+    cur_btn.click(run_curator, [playbook, deltas], [playbook, diff])
+    save_pb_btn.click(lambda t: (save_playbook(yaml.safe_load(t)), t)[1], playbook, playbook)
+    cem_save_btn.click(cem_save, cem_dataframe, cem_dataframe)
 
 if __name__ == "__main__":
     demo.launch()

cem.csv

@@ -0,0 +1 @@
+ClaimID,Claim,EvidenceRefs,Counter-Evidence,Confidence

playbook.yaml

@@ -0,0 +1,45 @@
+playbook:
+  metadata:
+    name: "ACE–CPT Playbook"
+    version: "0.1"
+  items:
+    - id: "PB-001"
+      type: "checklist"
+      title: "Stillness preflight"
+      content: |
+        Restate task (1 line). List 2–3 success criteria. One salient risk. Name initial paradigm lens (data/method/normative).
+      tags: ["meta","stillness"]
+      helpful: 0
+      harmful: 0
+    - id: "PB-002"
+      type: "procedure"
+      title: "Dual-source / unit-test verify"
+      content: |
+        For claims affecting safety/finance/medical/legal: confirm with an independent source or unit test before final Answer.
+      tags: ["verification"]
+      helpful: 0
+      harmful: 0
+    - id: "PB-003"
+      type: "checklist"
+      title: "Anthropomorphization & spec-gaming sentinel"
+      content: |
+        (a) Replace mind-language with mechanism; (b) check rubric-gaming; (c) add uncertainty note if incentives misalign.
+      tags: ["safety","bias"]
+      helpful: 0
+      harmful: 0
+    - id: "PB-004"
+      type: "heuristic"
+      title: "Paradigmatic surfacing (2.11–2.17)"
+      content: |
+        Ask: ontology, evidence, method, objective, normative values, institutional forces, closure/blind spots.
+      tags: ["cpt"]
+      helpful: 0
+      harmful: 0
+    - id: "PB-005"
+      type: "failure_mode"
+      title: "Context collapse via over-summarization"
+      content: |
+        Avoid replacing many atomic items with a single generic prompt; use deltas and dedupe instead.
+      tags: ["hygiene"]
+      helpful: 0
+      harmful: 0

requirements.txt

@@ -0,0 +1,3 @@
+gradio>=4.44.0
+pyyaml
+pandas