# Extropy Engine — Technical Specification v3.1

**Status:** Official v3.1 (supersedes prior draft mislabeled "3.1" which is hereby renamed v3.0)
**Date:** 2026-05-02
**Curator:** Randall Gossett
**Co-author (AI-assisted):** Perplexity Computer / Comet
**Source memo:** Consolidated decisions and clarifications from Grok stress-test conversation — May 1 2026

---

## 0. Changelog from v3.0 → v3.1

- Replaced "epistemology-engine" / central AI framing with **Personal AI + Handshake Model**.
- Added **Identity & Accountability Layer** (pseudonymous-by-default, signed actions, reputation-bound).
- Documented **Substrate Decision: Holochain** with three concrete patterns.
- Introduced **Personal Signed Local Log (PSLL)** as the canonical user-state primitive.
- Added **Micro-Quest / XP Loop** game-theory layer with diagrams and worked examples.
- Added **Governance Defaults** (adopted now, revisable by future on-chain vote).
- Added pointer to **GAPS.md** (the 63-gap reference index).

## 1. Core Vision

Extropy Engine is a peer-to-peer substrate for collective sense-making and coordinated action. It is **not** a central AI that knows things on your behalf; it is a fabric in which **personal AIs** negotiate, attest, and exchange evidence on behalf of their humans, under signed identity and reputation.

## 2. Personal AI + Handshake Model

- Every participant runs a **Personal AI (PAI)** locally (or in a user-controlled enclave).
- PAIs interact via a **handshake protocol**: declare intent → exchange capability tokens → exchange signed claims → reach a bounded agreement or decline.
- No central oracle. Truth is **emergent from cross-attestation** between PAIs with reputation stakes.
- The network never sees raw user state; it sees signed, scoped claims.

## 3. Identity & Accountability Layer

- **Pseudonymous-by-default DID** anchored in the user's PSLL.
- Every action is **signed** by the DID's current key; key rotation is itself a signed event in the PSLL.
- **Reputation** is the integral of attested-good actions minus slashed actions, scoped per context (quest market, governance, content, etc.).
- **Accountability** = the ability of any peer to verify a chain of signed actions back to a DID without learning the human behind it, unless that human chooses to reveal.

## 4. Operational / UI Model

- User talks to **their** PAI in natural language.
- PAI surfaces: pending handshakes, open quests, reputation deltas, governance proposals, evidence requests.
- One-click affordances: accept / decline / counter / escalate-to-human-review.
- No global feed. Discovery is **quest-driven** and **reputation-weighted**.

## 5. Substrate & Architecture Decision: Holochain

Three adopted patterns:

1. **Per-Agent Source Chain (PSLL host):** each user's signed local log lives as their Holochain source chain. It is the source of truth for *their* state.
2. **Sharded DHT for Public Claims:** signed, scoped claims that need peer discovery (quest postings, attestations, reputation events) are published to the validating DHT.
3. **Membrane / Capability Tokens for Handshakes:** PAIs exchange capability grants via Holochain capability tokens, scoped and revocable.

Rationale: agent-centric, no global consensus tax, native signed history, fits the personal-AI model 1:1.

## 6. Personal Signed Local Log (PSLL)

- Append-only, hash-linked, signed by the user's current DID key.
- Stores: identity events, handshake records, quest commitments, XP ledger, governance votes, key rotations.
- Exportable, portable, user-owned. Loss of PSLL = loss of continuous identity (mitigated by user-controlled backup / social recovery, *not* by central recovery).

## 7. XP / Loop / Game-Theory Layer

- **Micro-quests** are small, signed commitments with a verifiable completion condition and a reputation-staked sponsor.
- **Loop:** propose → accept → attest → settle → reputation update → next quest discovery.
- XP is **non-transferable** and context-scoped; it cannot be bought.
- Anti-sybil: quest issuance and attestation cost reputation, not money.

### 7.1 Worked example

> Alice's PAI posts a micro-quest: "verify source X claims Y" with 5 rep staked. Bob's PAI accepts, returns signed evidence. Carol's PAI independently attests. Settlement: Bob +3 XP in `verification` context, Alice -0 (quest fulfilled), Carol +1 `attestor` XP.

## 8. Governance Defaults (adopted now; revisable later)

- Quorum: rolling reputation-weighted, context-scoped.
- Proposal lifecycle: draft → handshake-circulation → vote → enact → audit.
- Defaults are written into v3.1 so the system can ship; any of them can be amended by a passing governance proposal under the same rules.

## 9. Thinking / Process Note

This spec is a snapshot of decisions, not a finished product. Where v3.0 left tensions unresolved (central-AI framing, identity ambiguity, substrate hand-waving), v3.1 commits to specific shapes so implementation can proceed. The 63 known gaps are catalogued in `GAPS.md` and remain open work.

## 10. References

- `IDENTITY.md` — Identity & Accountability Layer detail
- `PSLL.md` — Personal Signed Local Log spec
- `QUEST_MARKET.md` — Micro-quest market mechanics
- `GOVERNANCE_DEFAULTS.md` — Default governance parameters
- `GAPS.md` — 63-gap reference index
- `CHANGELOG.md` — Version history