# NYVRA — architecture

NYVRA is an autonomous agent that proves its decisions.

Observe. Decide. Prove. Append.

A hold is a decision. In the intended protocol, a decision without proof does not exist.

## This release

This website is a deterministic, off-chain browser simulator. It does not fetch market data, trade, sign transactions, generate zero-knowledge proofs, or connect to Solana.

`nyvra_files/engine.mjs` supplies immutable Policy M, synthetic observations, decisions, SHA-256 commitments, simulation proof checksums and hash-linked history. `protocol.mjs` runs the four-stage loop and connects its history to every readout and inspector.

Genesis time is fixed at 2026-10-01 00:00:00 UTC. Each transition advances the simulation clock by 12 seconds. The initial 420 blocks are reconstructed on every page load. Session playback then appends one transition approximately every 5.6 seconds. Pause freezes the current stage. Step finishes exactly one transition and remains paused. Reload resets the session to its deterministic starting history.

Synthetic observations have score `975 + sequence % 22` when `sequence % 45 == 20`, and `(sequence * 37 + 17) % 975` otherwise. A score of at least 975 yields STRIKE; everything else yields HOLD. This synthetic rule is illustrative, not a trading strategy. It creates one strike every 45 transitions, approximately 2.2 percent, including block 425. There is no backend, wallet, persistent ledger or execution engine.

All counters are derived from the same history. The stream shows the latest eight records; Night Watch and the inspector selector show the latest 48. Older selected records remain inspectable. Each record includes the preceding hash. The demo performs local integrity checks before appending.

## Intended protocol

1. Commit an observation from an authenticated source.
2. Evaluate the policy committed at genesis.
3. Produce a proof binding observation, policy and decision.
4. Verify the proof and append signed history.
5. Reject invalid transitions, according to the specified lifecycle.

Observation authenticity and completeness require their own threat model. A hash alone does not establish what an agent actually saw, and cannot prevent withheld inputs. The browser demo does not solve these issues.

## Roadmap

- Available: browser lifecycle simulation and local hash integrity checks.
- Next design milestone: Groth16 circuit and committed-observation proof specification.
- Planned: Solana verifier/program, on-chain transitions and public verification.
- Later: autonomous execution and long-lived agents.

There is no production verifier, deployed Solana program, token, or CA in this release. The original repository URL returned HTTP 404 during verification on 2026-10-02; a replacement public repository is awaiting confirmation.