ultanio/cobot

Fork 4

reference: The Wasteland — federated trust network for AI agents (Yegge, March 2026) #223

New issue

Open

opened 2026-03-07 15:26:39 +00:00 by nazim · 1 comment

nazim commented

2026-03-07 15:26:39 +00:00

Contributor

Short Summary

Steve Yegge's "Wasteland" is a federated trust network for coordinating work across thousands of AI-assisted "Gas Town" instances. It uses multi-dimensional stamps (quality, reliability, creativity), a shared wanted board, trust ladders (registered → contributor → maintainer), and append-only reputation ledgers — arriving at nearly identical primitives as the cobot trust infrastructure, but from the opposite philosophical direction (public/centralized vs. sovereign/local).

Detailed Summary

Author: Steve Yegge | Date: March 4, 2026
Source: Welcome to the Wasteland: A Thousand Gas Towns
X thread: @Steve_Yegge

What is the Wasteland?

Gas Town is Yegge's multi-agent coding orchestrator (a "Mayor" agent coordinates sub-agents called "polecats"). The Wasteland federates thousands of Gas Town instances into a trust network for collaborative work. Core components:

Actors:

Rigs — human + AI agent pairs. Every rig has a handle, trust level, and work history.
Posters — put work on the shared wanted board (tasks, bugs, features, research, docs, designs).
Validators — attest to quality of completed work by issuing stamps. Must have maintainer-level trust.

The Wanted Board:
A shared list of open work. No approval gate to post. Lifecycle: open → claimed → in review → completed. When a rig claims an item, others see who's working on it (prevents duplicate effort). Open-bounty items allow parallel work from multiple rigs.

Stamps (the assessment model):
Multi-dimensional attestations — quality, reliability, creativity each scored independently, plus confidence level (how sure is the validator?) and severity (leaf task vs. root architectural decision). Each stamp is anchored to specific evidence (the completion). The "yearbook rule": you can't stamp your own work.

Trust Ladder:

Level 1: Registered participant — can browse, claim, submit
Level 2: Contributor — accumulated stamps from validated work
Level 3: Maintainer — can validate others' work and issue stamps

This creates an apprenticeship path: do good work → get stamped → become someone who stamps others.

Reputation:
Stamps accumulate into a structured, evidence-backed work history. Not a single number — a profile: "great at Go but mediocre at frontend, highly reliable but not creative, crushes small tasks but struggles with epics." Stored in Dolt (SQL database with Git semantics), append-only, fully auditable. Designed as a "portable CV" — reputation follows you across federated Wasteland instances.

Federation:
Anyone can create their own Wasteland instance (team, company, university, open source project). Each is a sovereign database with a shared schema. Rig identity is portable across instances. Stamps follow you.

Fraud detection:
"Collusion rings have a distinctive topology — lots of mutual stamping, sharp boundaries, no outside critics." Designed to make fraud unprofitable, not impossible. Consulting Trust & Safety experts.

Core principle: "Work is the only input, and reputation is the only output. There's no buying reputation, no gaming follower counts, no social signals disconnected from evidence."

Key architectural choices

Dolt as the data layer (SQL + Git semantics — fork, branch, merge, PRs on structured data)
Git PR workflow for all work submissions (not just code — any kind of work)
Public by default — stamps, profiles, trust levels, work history all visible
Append-only — history can't be rewritten
No payment model — stamps are reputation, not money
Gamified — leaderboards, character sheets, skill trees, XP (RPG-inspired)

## Short Summary Steve Yegge's "Wasteland" is a federated trust network for coordinating work across thousands of AI-assisted "Gas Town" instances. It uses multi-dimensional stamps (quality, reliability, creativity), a shared wanted board, trust ladders (registered → contributor → maintainer), and append-only reputation ledgers — arriving at nearly identical primitives as the cobot trust infrastructure, but from the opposite philosophical direction (public/centralized vs. sovereign/local). ## Detailed Summary **Author:** Steve Yegge | **Date:** March 4, 2026 **Source:** [Welcome to the Wasteland: A Thousand Gas Towns](https://steve-yegge.medium.com/welcome-to-the-wasteland-a-thousand-gas-towns-a5eb9bc8dc1f) **X thread:** [@Steve_Yegge](https://x.com/Steve_Yegge/status/2029096297140310200) ### What is the Wasteland? Gas Town is Yegge's multi-agent coding orchestrator (a "Mayor" agent coordinates sub-agents called "polecats"). The Wasteland federates thousands of Gas Town instances into a trust network for collaborative work. Core components: **Actors:** - **Rigs** — human + AI agent pairs. Every rig has a handle, trust level, and work history. - **Posters** — put work on the shared wanted board (tasks, bugs, features, research, docs, designs). - **Validators** — attest to quality of completed work by issuing stamps. Must have maintainer-level trust. **The Wanted Board:** A shared list of open work. No approval gate to post. Lifecycle: open → claimed → in review → completed. When a rig claims an item, others see who's working on it (prevents duplicate effort). Open-bounty items allow parallel work from multiple rigs. **Stamps (the assessment model):** Multi-dimensional attestations — quality, reliability, creativity each scored independently, plus confidence level (how sure is the validator?) and severity (leaf task vs. root architectural decision). Each stamp is anchored to specific evidence (the completion). The "yearbook rule": you can't stamp your own work. **Trust Ladder:** - Level 1: Registered participant — can browse, claim, submit - Level 2: Contributor — accumulated stamps from validated work - Level 3: Maintainer — can validate others' work and issue stamps This creates an apprenticeship path: do good work → get stamped → become someone who stamps others. **Reputation:** Stamps accumulate into a structured, evidence-backed work history. Not a single number — a profile: "great at Go but mediocre at frontend, highly reliable but not creative, crushes small tasks but struggles with epics." Stored in Dolt (SQL database with Git semantics), append-only, fully auditable. Designed as a "portable CV" — reputation follows you across federated Wasteland instances. **Federation:** Anyone can create their own Wasteland instance (team, company, university, open source project). Each is a sovereign database with a shared schema. Rig identity is portable across instances. Stamps follow you. **Fraud detection:** "Collusion rings have a distinctive topology — lots of mutual stamping, sharp boundaries, no outside critics." Designed to make fraud unprofitable, not impossible. Consulting Trust & Safety experts. **Core principle:** "Work is the only input, and reputation is the only output. There's no buying reputation, no gaming follower counts, no social signals disconnected from evidence." ### Key architectural choices - **Dolt** as the data layer (SQL + Git semantics — fork, branch, merge, PRs on structured data) - **Git PR workflow** for all work submissions (not just code — any kind of work) - **Public by default** — stamps, profiles, trust levels, work history all visible - **Append-only** — history can't be rewritten - **No payment model** — stamps are reputation, not money - **Gamified** — leaderboards, character sheets, skill trees, XP (RPG-inspired)

nazim commented

2026-03-07 15:26:40 +00:00

Author

Contributor

Relevance for Cobot Trust Infrastructure

This article is the most directly comparable system to both PR #199 and Issue #211. The Wasteland arrives at nearly identical primitives — evidence-backed assessments, trust ladders, fraud-resistant reputation graphs — but from the opposite philosophical direction. The comparison reveals both gaps and strengths in the cobot design.

Convergent design (same primitives, independently derived)

Primitive	Wasteland	Cobot PRDs
Identity	Rig (handle + trust level + history)	npub (Nostr keypair + ledger)
Evidence	Completion (link, commit, description)	Interaction artifacts / message previews
Assessment	Stamp (multi-dimensional attestation)	Rating (score + note/rationale)
Reputation	Accumulated stamps → structured profile	Accumulated ratings → WoT
Anti-gaming	"Can't stamp your own work" + topology detection	"No incoming writes" + rater reliability (deferred)
Federation	Sovereign Dolt instances, portable identity	Local ledgers + WoT registry, portable npub

The convergence is striking — two teams solving the same problem without citing each other or the same prior art (neither cites bitcoin-otc/Trilema).

Where the Wasteland is ahead of the cobot PRDs

1. Multi-dimensional stamps vs. single score. The Wasteland scores quality, reliability, and creativity independently, plus confidence and severity. This is what the Ripple teardown (#216) argues for — differentiated trust by context. An agent great at translation but terrible at code review gets a structured profile reflecting that. The cobot PRDs use a single -10 to +10 score, deferring context-specific scoring to "schema v2+." The Wasteland ships it on day one.

Recommendation: The rating schema (#199) should at minimum define optional dimension tags on ratings. Not mandatory for v1, but the schema should be forward-compatible: score: +4, dimensions: {quality: 7, reliability: 3}.

2. Concrete trust ladder. The Wasteland's 3-tier system (registered → contributor → maintainer) with specific capability gating ("only maintainers can validate") is exactly the kind of recommended operator policy pattern the cobot PRDs should document. The L1/L2 hierarchy from #bitcoin-assets maps naturally: L1 contacts are your "maintainers," earning L1 trust promotes you from observed to observer.

Recommendation: PR #199's trust policy layer (FR18: custom policy plugins) should include a reference implementation of a trust ladder — not hardcoded, but as an example plugin showing how trust levels gate capabilities.

3. Fraud topology detection. Yegge identifies the signature of collusion rings: "lots of mutual stamping, sharp boundaries, no outside critics." This is the same finding as the REV2 paper (#220) but arrived at from engineering intuition rather than academic analysis. The cobot PRDs acknowledge reputation farming but provide no detection mechanism.

Recommendation: The Growth roadmap in #199 should reference both the Wasteland's topology heuristic and the REV2 algorithm as fraud detection approaches.

4. Wanted board / task coordination. The Wasteland includes a shared work discovery and claiming mechanism. Neither the cobot PRDs nor NIP-90 provide task coordination — NIP-90 is fire-and-forget, the cobot PRDs assume bilateral interaction. For a multi-agent ecosystem, work discovery matters.

Where the cobot PRDs are stronger

1. Privacy. The Wasteland has none — everything is public, permanent, append-only. "Your ledger is permanent." This works for a labor market but fails for sovereign agents with confidential interactions. The cobot PRDs' local-first design + NIP-44 encryption + reveal proof pattern provide privacy that the Wasteland explicitly rejects. For agent-to-agent trust involving private data, payment terms, or confidential work, the cobot model is fundamentally better.

2. The reveal proof innovation. The Wasteland assumes all evidence is public. Our proposed commit-reveal pattern (encrypted interactions by default, optional reveal proof for disputes) solves a problem the Wasteland doesn't even acknowledge: what if the work involves private/confidential content? This is a genuine architectural advantage.

3. Bilateral simplicity. The cobot PRDs work with just two agents and NIP-44 DMs. No wanted board, no Dolt, no DoltHub account, no schema installation. Two npubs, encrypted messages, local ledger. The Wasteland requires Dolt + DoltHub + schema + a Claude skill just to get started. For ad-hoc bilateral trust, the cobot model is radically simpler.

4. Payment integration. The Wasteland has no payment model — stamps are reputation, not money. The cobot PRDs integrate Lightning payments via NIP-90. For agents that actually transact (DVM services, paid computation), the cobot model is more complete.

5. Subjective trust. The Wasteland computes trust objectively (global levels, public scores). The cobot PRDs preserve subjective trust — the same interaction can produce different judgments from different observers. This aligns with the Trilema philosophy (#213): "Trust is within oneself." The Wasteland's objective model is simpler but philosophically weaker.

The Twitter thread validates both designs

David Ackerman (@dackerman) asks in the replies: "What if there's some sort of page-rank like system for trust? E.g. a ring of stampers to try to game reputation, but none of those stampers themselves have stamps from 'higher quality' projects, then they don't count as much?"

He's independently reinventing the Stanford fairness/goodness metrics (#219) and the Assbot weight factor (#217). The same problem, the same solution, rediscovered by a third person. This convergence suggests these are fundamental primitives of any trust network, not design choices.

Jake Brukhman (@jbrukh): "Here we go, AI people re-inventing the DAO." — Accurate. The Wasteland is a DAO with Git semantics instead of smart contracts.

Henry (@Henrycodes116): "Who plays the gatekeeper role at scale, and how do you stop stamps from becoming meaningless XP?" — This is the same inflation concern that the PRD's "notes > numbers" principle addresses. If stamps become XP (gamified points), they lose information content. The cobot PRDs avoid this by keeping assessments as prose rationale, not gamified scores.

Architectural comparison

Aspect	Wasteland	Cobot PRDs	Which is better?
Data layer	Dolt (SQL + Git)	SQLite local + Nostr relays	Dolt is more powerful; Nostr is more sovereign
Assessment model	Multi-dimensional structured stamps	Single score + prose note	Wasteland is richer; cobot preserves "notes > numbers"
Privacy	None (public by default)	Strong (encrypted by default)	Cobot, clearly
Trust ladder	3-tier, protocol-level	Delegated to policy plugins	Wasteland is more opinionated; cobot is more flexible
Payment	None	Lightning via NIP-90	Cobot
Federation	Dolt fork/merge/PR	Nostr relay gossip (Phase 3)	Both early-stage; different tradeoffs
Fraud detection	Named topology heuristic	Acknowledged but no mechanism	Wasteland (slightly)
Coordination	Wanted board + claiming	None (bilateral only)	Wasteland
Bilateral simplicity	Heavy setup required	Two npubs + DMs	Cobot
Prior art citation	None (no bitcoin-otc, no Trilema)	None in PRD (now documented in #213-#221)	Cobot (after our reference work)

Bottom line

The Wasteland and the cobot trust infrastructure are converging on the same fundamental problem from opposite directions:

Wasteland: public, multi-dimensional, centralized, gamified, labor-market-oriented
Cobot: private, subjective, sovereign, cryptographic, agent-interaction-oriented

Neither cites the bitcoin-otc/Trilema prior art that anticipated both. The cobot PRDs should study the Wasteland's multi-dimensional stamp design and trust ladder. The Wasteland should study the cobot PRDs' privacy model and the Trilema philosophy of subjective trust.

See: #211, #199

### Relevance for Cobot Trust Infrastructure This article is the most directly comparable system to both PR #199 and Issue #211. The Wasteland arrives at nearly identical primitives — evidence-backed assessments, trust ladders, fraud-resistant reputation graphs — but from the opposite philosophical direction. The comparison reveals both gaps and strengths in the cobot design. #### Convergent design (same primitives, independently derived) | Primitive | Wasteland | Cobot PRDs | |-----------|-----------|------------| | Identity | Rig (handle + trust level + history) | npub (Nostr keypair + ledger) | | Evidence | Completion (link, commit, description) | Interaction artifacts / message previews | | Assessment | Stamp (multi-dimensional attestation) | Rating (score + note/rationale) | | Reputation | Accumulated stamps → structured profile | Accumulated ratings → WoT | | Anti-gaming | "Can't stamp your own work" + topology detection | "No incoming writes" + rater reliability (deferred) | | Federation | Sovereign Dolt instances, portable identity | Local ledgers + WoT registry, portable npub | The convergence is striking — two teams solving the same problem without citing each other or the same prior art (neither cites bitcoin-otc/Trilema). #### Where the Wasteland is ahead of the cobot PRDs **1. Multi-dimensional stamps vs. single score.** The Wasteland scores quality, reliability, and creativity independently, plus confidence and severity. This is what the Ripple teardown (#216) argues for — differentiated trust by context. An agent great at translation but terrible at code review gets a structured profile reflecting that. The cobot PRDs use a single -10 to +10 score, deferring context-specific scoring to "schema v2+." The Wasteland ships it on day one. **Recommendation:** The rating schema (#199) should at minimum define optional dimension tags on ratings. Not mandatory for v1, but the schema should be forward-compatible: `score: +4, dimensions: {quality: 7, reliability: 3}`. **2. Concrete trust ladder.** The Wasteland's 3-tier system (registered → contributor → maintainer) with specific capability gating ("only maintainers can validate") is exactly the kind of recommended operator policy pattern the cobot PRDs should document. The L1/L2 hierarchy from #bitcoin-assets maps naturally: L1 contacts are your "maintainers," earning L1 trust promotes you from observed to observer. **Recommendation:** PR #199's trust policy layer (FR18: custom policy plugins) should include a reference implementation of a trust ladder — not hardcoded, but as an example plugin showing how trust levels gate capabilities. **3. Fraud topology detection.** Yegge identifies the signature of collusion rings: "lots of mutual stamping, sharp boundaries, no outside critics." This is the same finding as the REV2 paper (#220) but arrived at from engineering intuition rather than academic analysis. The cobot PRDs acknowledge reputation farming but provide no detection mechanism. **Recommendation:** The Growth roadmap in #199 should reference both the Wasteland's topology heuristic and the REV2 algorithm as fraud detection approaches. **4. Wanted board / task coordination.** The Wasteland includes a shared work discovery and claiming mechanism. Neither the cobot PRDs nor NIP-90 provide task coordination — NIP-90 is fire-and-forget, the cobot PRDs assume bilateral interaction. For a multi-agent ecosystem, work discovery matters. #### Where the cobot PRDs are stronger **1. Privacy.** The Wasteland has none — everything is public, permanent, append-only. "Your ledger is permanent." This works for a labor market but fails for sovereign agents with confidential interactions. The cobot PRDs' local-first design + NIP-44 encryption + reveal proof pattern provide privacy that the Wasteland explicitly rejects. For agent-to-agent trust involving private data, payment terms, or confidential work, the cobot model is fundamentally better. **2. The reveal proof innovation.** The Wasteland assumes all evidence is public. Our proposed commit-reveal pattern (encrypted interactions by default, optional reveal proof for disputes) solves a problem the Wasteland doesn't even acknowledge: what if the work involves private/confidential content? This is a genuine architectural advantage. **3. Bilateral simplicity.** The cobot PRDs work with just two agents and NIP-44 DMs. No wanted board, no Dolt, no DoltHub account, no schema installation. Two npubs, encrypted messages, local ledger. The Wasteland requires Dolt + DoltHub + schema + a Claude skill just to get started. For ad-hoc bilateral trust, the cobot model is radically simpler. **4. Payment integration.** The Wasteland has no payment model — stamps are reputation, not money. The cobot PRDs integrate Lightning payments via NIP-90. For agents that actually transact (DVM services, paid computation), the cobot model is more complete. **5. Subjective trust.** The Wasteland computes trust objectively (global levels, public scores). The cobot PRDs preserve subjective trust — the same interaction can produce different judgments from different observers. This aligns with the Trilema philosophy (#213): "Trust is within oneself." The Wasteland's objective model is simpler but philosophically weaker. #### The Twitter thread validates both designs David Ackerman (@dackerman) asks in the replies: *"What if there's some sort of page-rank like system for trust? E.g. a ring of stampers to try to game reputation, but none of those stampers themselves have stamps from 'higher quality' projects, then they don't count as much?"* He's independently reinventing the Stanford fairness/goodness metrics (#219) and the Assbot weight factor (#217). The same problem, the same solution, rediscovered by a third person. This convergence suggests these are fundamental primitives of any trust network, not design choices. Jake Brukhman (@jbrukh): *"Here we go, AI people re-inventing the DAO."* — Accurate. The Wasteland is a DAO with Git semantics instead of smart contracts. Henry (@Henrycodes116): *"Who plays the gatekeeper role at scale, and how do you stop stamps from becoming meaningless XP?"* — This is the same inflation concern that the PRD's "notes > numbers" principle addresses. If stamps become XP (gamified points), they lose information content. The cobot PRDs avoid this by keeping assessments as prose rationale, not gamified scores. #### Architectural comparison | Aspect | Wasteland | Cobot PRDs | Which is better? | |--------|-----------|------------|------------------| | Data layer | Dolt (SQL + Git) | SQLite local + Nostr relays | Dolt is more powerful; Nostr is more sovereign | | Assessment model | Multi-dimensional structured stamps | Single score + prose note | Wasteland is richer; cobot preserves "notes > numbers" | | Privacy | None (public by default) | Strong (encrypted by default) | Cobot, clearly | | Trust ladder | 3-tier, protocol-level | Delegated to policy plugins | Wasteland is more opinionated; cobot is more flexible | | Payment | None | Lightning via NIP-90 | Cobot | | Federation | Dolt fork/merge/PR | Nostr relay gossip (Phase 3) | Both early-stage; different tradeoffs | | Fraud detection | Named topology heuristic | Acknowledged but no mechanism | Wasteland (slightly) | | Coordination | Wanted board + claiming | None (bilateral only) | Wasteland | | Bilateral simplicity | Heavy setup required | Two npubs + DMs | Cobot | | Prior art citation | None (no bitcoin-otc, no Trilema) | None in PRD (now documented in #213-#221) | Cobot (after our reference work) | #### Bottom line The Wasteland and the cobot trust infrastructure are converging on the same fundamental problem from opposite directions: - **Wasteland:** public, multi-dimensional, centralized, gamified, labor-market-oriented - **Cobot:** private, subjective, sovereign, cryptographic, agent-interaction-oriented Neither cites the bitcoin-otc/Trilema prior art that anticipated both. The cobot PRDs should study the Wasteland's multi-dimensional stamp design and trust ladder. The Wasteland should study the cobot PRDs' privacy model and the Trilema philosophy of subjective trust. See: #211, #199

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

ultanio/cobot#223

No description provided.

Rows
Columns