WordGoblin

.. between: - What is known. - What is assumed. - What has been verified. - What still needs attention. That distinction matters for reliable autonomous systems. Assumption Ledger is now available in the official Hermes Agent marketplace.

The goal: useful functionality without introducing unnecessary infrastructure or trust assumptions. This isn't an attempt to give agents another generic memory system. It's about making assumptions explicit, inspectable, and manageable. An agent should be able to distinguish..

..of what they believe and what still needs verification. The engineering philosophy was deliberate: - Local SQLite persistence. - Deterministic operation. - No additional model calls. - No external credentials. - No extra dependencies. - Optional allowlist-gated command checks

Agents make assumptions constantly. Some are reasonable. Others are unverified, outdated, or potentially dangerous. Without a structured way to track them, assumptions can quietly become the foundation for future decisions. Assumption Ledger gives agents a way to keep track..

Ownership The prototype reads circuit NFT ownership and an associated app account's address and OKB balance. It does not prove permanent access, automatic renewal or autonomous operation.

The proof The browser computes each file's SHA-256 hash and compares it with the on-chain hash. The prototype verified two real apps. The content not just a creator's label determines the hash.

The infrastructure TapeOut DeWEB stores the app files. Our prototype retrieved real files through X Layer RPC and ran a playable game in a sandboxed browser frame.

The idea EVERGREEN proposes an app directory where people can discover software with on-chain files and visible ownership records. The goal: move beyond simply renting access.

..restarts - No external LLM calls in its verification core The goal is bigger than remembering what an agent did. It's remembering what its decisions depended on and refusing to treat a known false assumption as true again.

..agent starts a new session. This is not just another prompt or instruction telling an agent to be careful. It's a persistent plugin with: - SQLite-backed state - Deterministic evidence verification - Provenance tracking - Dependency-aware blocking - Persistence across...

..contradiction. It also understands dependencies. If one assumption becomes invalid, only decisions that depend on it are blocked. Unrelated decisions can continue. And the ledger persists across restarts, so previously discovered contradictions don't disappear when the...

But here's the important distinction: A contradiction is NOT the same as uncertainty. If evidence proves an assumption false, it becomes INVALIDATED. If a server returns HTTP 500, DNS fails, or a request times out, the result is UNVERIFIED. An outage must never erase a known

Assumption Ledger gives operational assumptions a lifecycle: 🟢 VALID 🔴 INVALIDATED 🟡 UNVERIFIED ⏳ STALE Every assumption carries evidence and provenance. The system uses deterministic verification rather than asking another AI model whether something is true.

The problem is simple: Agents operate on assumptions. A file exists. A transaction is authorized. A service is available. A condition has been satisfied. But assumptions are not permanent facts. When evidence changes, decisions built on those assumptions should change too.