thirdanswer 0.1.0

The Third Answer — evaluate AI uncertainty with T,I,F (Truth, Indeterminacy, Falsity)

thirdanswer

The Third Answer — evaluate AI uncertainty with T,I,F (Truth, Indeterminacy, Falsity).

Based on the book "The Third Answer: Why AI Doesn't Know What It Doesn't Know — And How Ancient Logic Can Fix It" by Leyva-Vazquez & Smarandache (2026).

Install

pip install thirdanswer

For LLM analysis (free via Groq):

pip install thirdanswer[groq]

Quick Start

Without LLM (pure logic)

from thirdanswer import Compass

# Create a reading
c = Compass(T=0.7, I=0.4, F=0.5)
print(c.zone)              # "contradiction"
print(c.zone_emoji)        # "🟠"
print(c.confidence)        # 0.0
print(c.is_paraconsistent) # True (T+F=1.2 > 1)
print(c.should_abstain)    # False
print(c.zone_action)       # "Investigate both sides..."

Analyze any text (with Groq, free)

from thirdanswer import analyze

r = analyze(
    "Coffee is definitively good for your health",
    provider="groq",
    api_key="gsk_..."  # Free at console.groq.com
)
print(r)           # Analysis(T=0.55, I=0.30, F=0.45) → 🟠 Contradiction
print(r.zone)      # "contradiction"
print(r.errors)    # ["confident_ignorance"] if detected
print(r.honest)    # Rewritten version with uncertainty

# Claim-by-claim
for claim in r.claims:
    print(f"{claim.zone_emoji} {claim.text[:50]}  T={claim.T}")

Ask honest questions

from thirdanswer import ask

r = ask("Is intermittent fasting healthy?", provider="groq", api_key="gsk_...")
print(r.answer)
print(r.what_i_dont_know)  # "Long-term effects beyond 2 years..."
print(r.zone)              # "contradiction"

Compare two AI responses

from thirdanswer import compare

diff = compare(chatgpt_response, claude_response, provider="groq", api_key="gsk_...")
print(f"Agreement: {diff.agreement:.0%}")
print(f"Conflicts: {diff.conflicts}")
print(f"More honest: {diff.more_honest}")

Decompose text into claims

from thirdanswer import decompose

claims = decompose("Long AI response...", provider="groq", api_key="gsk_...")
for c in claims:
    print(f"{c.zone_emoji} T={c.T} I={c.I} F={c.F} | {c.text}")

The Four Zones

Zone	Condition	Action
🟢 Consensus	T high, I low, F low	Trust
🟡 Ambiguity	I high	Investigate
🟠 Contradiction	T high AND F high	Explore both sides
⚫ Ignorance	All low or I overwhelming	Stop

Providers

Provider	Cost	API Key	Install
Groq	Free (~30 req/min)	console.groq.com	pip install thirdanswer[groq]
Ollama	Free (local)	None	ollama.com

Compass works without any provider — pure logic, no API calls.

Links

• The Third Answer App
• Prompt Templates
• Book (coming soon)
• ORCID: M. Leyva-Vazquez

License

MIT