> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ameo.agiwithai.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Verifiable cognition

> What verifiable cognition means and how AMEO implements it.

**Verifiable cognition** means an autonomous agent's reasoning and authorization steps produce independently checkable artifacts — not just a narrative after the fact.

## Beyond explainable AI

Explainable AI (XAI) often means post-hoc summaries: "the model chose X because Y." That prose is not evidence. Verifiable cognition requires **structured, immutable commitments** at each stage:

1. **Observation snapshot** — balances, gas, market inputs at decision time.
2. **Reasoning trace** — hashed plan + optional 0G blob.
3. **Policy proof** — predicate results evaluated outside the model.
4. **Settlement proof** — tx hash, block, and `[DecisionLogged](/glossary#decisionlogged-event)` event.

An auditor replays the cycle without trusting the operator's UI.

## The four pillars in AMEO

| Pillar      | Artifact                                 | Where to verify            |
| ----------- | ---------------------------------------- | -------------------------- |
| Observation | JSON snapshot in JSONL + replay API      | `/app/replay` node 1–3     |
| Reasoning   | `rationaleHash`, planner output          | Replay node 4 + 0G indexer |
| Policy      | `guardrail_evaluated` violations or pass | Replay node 5              |
| Settlement  | Mantlescan tx + `DecisionLogged`         | Replay nodes 6–8           |

Empty states are honest — if 0G is not configured, the UI says so instead of showing a placeholder hash.

## Why it matters for production

Capital allocators need **replay**, not **trust**. Verifiable cognition is the minimum bar for autonomous treasuries, agentic wallets, and policy-bound DEX execution on public chains.

Next: [Agent identity](/concepts/agent-identity) · [Proof trail](/concepts/proof-trail)