TL;DR Most systems treat “verified” as “allowed.” StegCore separates those concepts.

StegCore is a small, docs-first project that defines a decision boundary: given verified continuity (from an external system), it answers allow / deny / defer, with explicit constraints like quorum, guardian review, veto windows, or time-locks.

No policy engine yet. No AGI claims. Just the missing layer.

⸻

The problem

Modern automation — especially AI-driven automation — usually collapses three things into one: 1. Truth (is this authentic / verified?) 2. Authority (is this allowed?) 3. Execution (do the thing)

That works… until it doesn’t.

When something goes wrong, there’s no clean place to: • pause an action • require consent • escalate to a human • recover without shutting everything down

Verified truth alone doesn’t tell you what is permitted.

⸻

What StegCore does

StegCore defines a narrow interface:

Given verified continuity, can this actor perform this action right now — and under what constraints?

Inputs: • verified continuity evidence (opaque to StegCore; e.g. from StegID) • actor class (human / AI / system) • action intent • policy context (structure only)

Output: • allow, deny, or defer • a stable, machine-readable reason code • optional constraints (quorum, guardian, veto window, time-lock, escalation)

StegCore: • does not verify receipts • does not store identity • does not execute actions • does not claim autonomy or intelligence

It declares decisions. Other systems act (or don’t).

⸻

Why “defer” matters

Most systems only support allow or deny.

In real systems, the safest answer is often: • “not yet” • “with consent” • “after review” • “after a delay”

StegCore treats defer as a first-class outcome, not a workaround.

That’s the difference between: • brittle automation and • recoverable automation

⸻

What’s in the repo today • Clear decision model and policy shape docs (authoritative) • Explicit agent lifecycle (intent → continuity → decision → execution) • A minimal, deterministic decision interface with tests • Scaffolding for state/audit signals (not continuity truth)

There is no policy engine yet. That’s intentional.

The docs are the contract; code is subordinate.

⸻

What this is not • Not an AGI claim • Not an auth system • Not identity management • Not a rules engine • Not a replacement for existing security tooling

It’s a missing layer that can sit between verification and execution.

⸻

Why this exists

We kept seeing the same failure mode:

“The system was technically correct, but it shouldn’t have been allowed to do that.”

StegCore exists to make “allowed” explicit.

⸻

Positioning (locked)

We’re not building general intelligence.

We are enabling:

AI systems that are accountable, recoverable, and constrained by verifiable continuity.

⸻

Status • v0.1 • docs-first • minimal decision boundary implemented • open to feedback before any policy runtime is built

Repo: https://github.com/StegVerse-Labs/StegCore

⸻

Questions we’d love feedback on • Is the separation between truth and permission clear? • Are “defer” + constraints useful in your systems? • Where does this boundary already exist implicitly, but undocumented? • What would you want before trusting a decision runtime?

Thanks for reading — happy to answer questions and clarify boundaries.

Show HN: StegCore – a decision boundary for AI systems (truth ≠ permission)