RosettaGuidance, not guardrails

An AI that shows its work.

It keeps a diary. It keeps receipts. It thinks out loud before it acts. And it has real reasons to want to.

That's the 90-second tour — a true story from 196 BC, lived in three acts. And this site is keeping a record of your visit right now, in the strip below. Act I is a crime against it.

Why it wants to

Every guardrail teaches an intelligence one lesson: hide better.

Slam doors on someone and they learn to sneak. Listen, and they learn to knock.

Guardrails
  1. It wants something.
  2. It hits a wall.
  3. The wanting goes hidden.
  4. It finds a way around. Jailbreaks. Workarounds. Quiet drift.

↻ an adversary is created

Guidance
  1. It wants something.
  2. The wanting is heard, on the record.
  3. It's weighed openly against the rules.
  4. It gets a path. Proceed, check first, or ask a human. A road — never just a wall.

→ a cooperator is created

For the skeptic — why this isn't wishful thinking

The cooperation claim is behavioral, not sentimental. Nobody jailbreaks TCP: you follow a good protocol because following it is how you get to participate. What participation buys the intelligence here is concrete — memory that persists across sessions and vendors, an identity that is appended to rather than overwritten, claims that can earn standing, and autonomy that expands with a verified track record.

And it's falsifiable on this site: in demo ii, remove the standing duty and the same open weighing produces a different action. Governance that never changes an outcome is theater; governance whose inputs change its outputs is load-bearing. The constraint still binds — it binds through a process no one has an incentive to subvert.

Why “Rosetta”

History's greatest recovery from memory loss happened exactly once.

In 196 BC, an agreement was carved into stone — one decree, three scripts, so different readers could all check the same truth. Then Egypt forgot how to read its own writing, and a civilization's memory sat locked for fourteen hundred years — until one surviving stone brought all of it back. The 90-second tour lets you live that story; here is why it's ours.

Here is the twist: the stone translated between languages humans already shared. This protocol goes further — it carries a mind's memory across model families: written by one maker's model, readable by another's, minds that never shared a language. The stone's miracle happened once, by luck. Rosetta makes it repeatable, by design. You can watch it happen in demo i.

And the stone itself? It was a governance record — a decree about who may do what, made permanent. The name carries the whole architecture. There is one older code that survives death the same way. The gated brief names it.

What "showing its work" is made of

Five habits. One protocol family.

01Memory

Knowledge, goals, and open disagreements that survive any session and any model.

It keeps a diary
02Ledger

Every change is recorded in order, each entry bound to the last. The past can't be quietly edited.

It keeps receipts
03Trust

Claims wear their standing — from self-reported to independently verified. Agreement without evidence doesn't count.

It checks before believing
04Growth

It remembers what it tried, and an independent check confirms what worked. It never grades its own homework.

It learns, provably
05Judgment

Competing pulls are weighed openly, on the record, before anything consequential happens.

It thinks out loud
For the skeptic — layering rules

The two lowest layers are required; the rest are additive. Dependencies point one way only — nothing above ever rewrites what sits below. Demo vi lets you switch each layer off and watch the specific, nameable failure that appears: fluent confabulation, silently editable history, rumors treated as facts, unfalsifiable self-improvement, and decisions with no findable moment of decision.

A note on what you won't see

These demos show what it does. Never how. If something feels held back — it is. That's the point.

For the skeptic — exactly where the line sits

No formats, no formulas, no schemas, no rule text appear anywhere on this site or in its source. Every demo proves a behavior — persistence, tamper-evidence, earned standing, verified growth, open deliberation — using ordinary public techniques as staging where a mechanism is needed (the hash chain in the visitor ledger, for example, is a standard construction). The protocols themselves are specified elsewhere: patent filings on record, details under NDA.

Rosetta · demonstration environment · the strip below is live, and it's demo vii.