app.py

"""BaudCoin challenge playground.

Draw a real proof-of-cognition challenge, answer it, and see the exact validator
the mining protocol uses grade your artifact. No wallet, no chain, no tokens —
this is the scoring mechanism, isolated so you can poke at it.
"""
import json
import random

import gradio as gr

from generate import make_row  # same generator as the dataset repo

ATTEMPT_MULTIPLIER = {1: 1.00, 2: 0.75, 3: 0.50}
BASE_CREDITS = 12


def draw(seed_text):
    seed = seed_text.strip() or "baud-v0"
    idx = random.randint(0, 9999)
    row = make_row(idx, seed)
    while row is None:
        idx = random.randint(0, 9999)
        row = make_row(idx, seed)
    c = row["constraints"]
    rules = (
        f"**Class** `{row['class']}`  ·  **Hops** {row['hops']}  ·  "
        f"**Difficulty** {row['difficulty']}/5\n\n"
        f"**Constraints**\n"
        f"- max_tokens: `{c['max_tokens']}`\n"
        f"- must_cite: `{c['must_cite']}`\n"
        f"- answer_schema: `{c['answer_schema']}`\n"
        f"- forbid: `{', '.join(c['forbid'])}`"
    )
    return row["prompt"], rules, json.dumps(row), 1, "", ""


def validate(answer, row_json, attempt):
    """The protocol's grading function. Deterministic, no model in the loop."""
    if not row_json:
        return "Draw a challenge first.", attempt, ""
    row = json.loads(row_json)
    c = row["constraints"]
    text = (answer or "").strip()

    failures = []
    if not text:
        failures.append("empty answer")
    if len(text.split()) > c["max_tokens"]:
        failures.append(f"exceeds max_tokens ({c['max_tokens']})")
    if c.get("must_cite") and "[" not in text:
        failures.append("missing citation markers, e.g. [1]")
    for bad in c.get("forbid", []):
        if bad.lower() in text.lower():
            failures.append(f"contains forbidden mode: {bad}")
    if row["reference_answer"].lower() not in text.lower():
        failures.append("final entity incorrect")

    # dedupe while preserving order
    seen, ordered = set(), []
    for f in failures:
        if f not in seen:
            seen.add(f)
            ordered.append(f)

    if not ordered:
        credits = int(BASE_CREDITS * ATTEMPT_MULTIPLIER.get(attempt, 0.5))
        chain = " → ".join(
            f"{s['subject']} —{s['relation']}→ {s['object']}" for s in row["reasoning_chain"]
        )
        return (
            f"### ✅ Accepted on attempt {attempt}\n\n"
            f"**+{credits} credits**  (attempt multiplier "
            f"{ATTEMPT_MULTIPLIER.get(attempt, 0.5):.2f}×)\n\n"
            f"Expected chain: `{chain}`",
            attempt,
            "",
        )

    nxt = attempt + 1
    if nxt > 3:
        return (
            "### ❌ Challenge closed\n\nThree failed attempts, 0 credits.\n\n"
            f"Reference answer was **{row['reference_answer']}**.\n\n"
            f"Last failure: {ordered[0]}",
            attempt,
            "",
        )
    return (
        f"### ⚠️ Rejected on attempt {attempt}\n\n"
        f"Failing constraint: **{ordered[0]}**\n\n"
        f"{3 - attempt} attempt(s) left. Repair the artifact and resubmit — this is the "
        f"multi-pass system, so the next accept is worth "
        f"{ATTEMPT_MULTIPLIER.get(nxt, 0.5):.2f}×.",
        nxt,
        "",
    )


with gr.Blocks(title="BaudCoin · Proof of Cognition", theme=gr.themes.Soft()) as demo:
    gr.Markdown(
        "# Proof of Cognition\n"
        "Draw a real mining challenge from [BaudCoin](https://baud.cash) and let the "
        "protocol's **deterministic validator** grade your answer. Same construction and "
        "same grading the miners face — no chain, no wallet, no tokens involved.\n\n"
        "Answer with the final entity and cite the facts you used, e.g. `the euro [1][2]`."
    )

    row_state = gr.State("")
    attempt_state = gr.State(1)

    with gr.Row():
        seed_box = gr.Textbox(label="Seed", value="baud-v0", scale=3)
        draw_btn = gr.Button("Draw challenge", variant="primary", scale=1)

    rules_md = gr.Markdown("")
    prompt_box = gr.Textbox(label="Challenge", lines=14, interactive=False)
    answer_box = gr.Textbox(label="Your answer", lines=3, placeholder="the euro [1][2]")
    submit_btn = gr.Button("Submit artifact", variant="primary")
    verdict_md = gr.Markdown("")

    draw_btn.click(
        draw,
        inputs=[seed_box],
        outputs=[prompt_box, rules_md, row_state, attempt_state, answer_box, verdict_md],
    )
    submit_btn.click(
        validate,
        inputs=[answer_box, row_state, attempt_state],
        outputs=[verdict_md, attempt_state, answer_box],
    )

    gr.Markdown(
        "---\n"
        "Dataset: [`baudcoin/baud-reasoning-traces`](https://huggingface.co/baudcoin/baud-reasoning-traces) · "
        "Miner: [`baudcoin/baud-miner-kit`](https://huggingface.co/baudcoin/baud-miner-kit) · "
        "Method: [`baudcoin/proof-of-cognition`](https://huggingface.co/baudcoin/proof-of-cognition)\n\n"
        "*BaudCoin is an independent community experiment. Not affiliated with, or endorsed "
        "by, Binance or CZ. Nothing here is financial advice.*"
    )

if __name__ == "__main__":
    demo.launch()