Reputation

Status: This section specifies the reputation architecture for v0.3 of the protocol. v0.1 ships with the dual-signature attestation primitive (§3); v0.2 adds BBS+ selective disclosure and the Bean Chain longitudinal-measurement layer (§6.3.5). v0.3 is where reputation is given its full formal treatment as a first-class protocol surface — with the multi-dimensional stamp, the yearbook rule, the reputation tensor, the trust floor, and the federated stamp graph. The Wasteland is already operating a subset of this surface in production; v0.3 of HOP specifies the full design and folds the Wasteland’s running schema in as the canonical reference implementation for the production-ready slices.

This section pulls from three canonical sources:

• Steve Yegge’s “Welcome to the Wasteland: A Thousand Gas Towns” (Medium, March 2026) — the operational implementation already in production. Yegge writes: “The Wasteland creates a permanent, evidence-traced record of your contributions” and the core principle “work is the only input, and reputation is the only output”.
• gastown/docs/WASTELAND.md at github.com/gastownhall/gastown — the production schema and command-line interface for the running system. Seven tables in wl_commons (rigs, wanted, completions, stamps, badges, chain_meta, _meta), MySQL-protocol-compatible, federated via DoltHub.
• The Inter-Town Communication Constitution — internal HOP design doc that specifies the philosophical and topological foundations: the yearbook rule, the reputation tensor, the trust floor, the true-enemy preference order, the $Variable substrate-neutrality, and the EXCHANGE_TOPOLOGY in which reputation is structurally excluded from direct exchange.

The Wasteland is producing the canonical operational answers; the Inter-Town Constitution is producing the canonical philosophical foundations. They are not in conflict. The Wasteland is what the philosophy looks like when it ships.

14.0 The Agent-Negotiation Reframe — Bid-and-Sell Applied to Reputation

Before specifying the v0.3 reputation surface, it is worth naming what makes it structurally different from any prior reputation system, including the Wasteland’s current implementation.

HOP’s labour-allocation layer runs on agent negotiation. A worker’s Skill-Agent (§5.1) constructs a bid curriculum from the worker’s Skillchain; the poster’s Worker-Agent (§5.2) evaluates the curriculum against the posting’s requirements; bid acceptance is a cryptographic mutex that locks the work. The agents do the search, the matching, the cross-checking, and the negotiation. The humans (or rigs, or institutions) configure the agents’ preferences and trust them to act within those preferences. This is the bid-and-sell pattern.

v0.3 applies the same pattern to reputation. When a chain needs to know another entity’s reputation — to evaluate a bid, to consider a federation treaty, to assess a candidate mentor, to decide whether to recognise a cross-chain skill stamp — it does not read a flat number. It dispatches an agent that runs a reputation query against the network. The agent:

1. Filters by the querier’s trust set (which chains’ attestations does this querier honour?)
2. Weights by the querier’s tensor priorities (which dimensions of the multi-dimensional stamp matter for this decision?)
3. Cross-checks the stamp-graph topology (is this reputation derived from a small mutually-reinforcing cluster, or does it have outside critics?)
4. Returns a structured answer specific to the context of the query

This is reputation as a negotiated, computed property rather than a published, static one. Two different observers querying the same subject receive different answers — not because the data is inconsistent, but because their evaluation lenses differ. Both answers are correct within their context. There is no universal reputation, only trust-filtered queries.

Why This Patches Westphalia

The Peace of Westphalia (1648) established the modern primitives for parties who disagree fundamentally about important things continuing to occupy the same physical space without escalating to violence: territorial sovereignty, non-interference in internal affairs, recognition through treaty. It worked for 377 years and is now visibly failing — at the international level, at the corporate level, at the personal level. The Westphalian primitives were not wrong; they were undersized for the substrate they now operate on.

The Westphalian failure mode was that trust between states was binary and personal — diplomatic recognition was either granted or withheld, and the work of maintaining recognition was carried by individual ambassadors meeting individual sovereigns. There was no mechanism for testing whether participants were actually keeping their word, no mechanism for computing the consequence of a violation, no mechanism for distributing the response through the network. When a sovereign violated the agreement, the response was either war or no response at all — there was no middle band of “compute the consequence, distribute it through the network, let the cascade do the work.”

The agent-negotiation reputation layer is structurally a patch to this failure mode. The thing Westphalia could not compute — does this sovereign actually act consistently with what their friends say about them, and what their enemies say about them, and what their behaviour shows over time? — is what HOP’s reputation queries compute automatically every time an agent dispatches one. Trust becomes a computational substrate, not a diplomatic ritual.

This generalises. The same pattern that lets a Workchain decide whether to recognise an applicant’s skill stamps from another chain is the pattern that lets two ideologically-opposed nation-states federate on labour mobility while disagreeing about everything else. The same pattern that protects Sera’s chain from being permanently invalidated by the Jehovah’s Witnesses’ negative attestations is the pattern that lets ASEAN states cooperate economically across genuine political differences. The scale changes; the pattern does not.

The Core Move: Concept-Attacks Cheaper Than Personality-Attacks

The reputation layer is the protocol’s concept-vs-personality firewall. Every reputation primitive specified in this section is a piece of that firewall:

• The multi-dimensional tensor (§14.2) prevents hot-take character verdicts by requiring evaluation along independent axes.
• The yearbook rule (§14.6) prevents self-attestation by database constraint.
• The trust-filtered query (§14.8) prevents echo-chamber consensus from being mistaken for universal recognition.
• The stamp-graph topology check (§14.15) prevents collusion rings from manufacturing reputation through mutual flattery.
• The true-enemy preference order (§14.9) weighs hostile-but-honest attestations higher than friendly-but-cheap ones.
• The severity-tiered response (§14.16, Clean Margins) ensures leaf disagreements get noted, branch failures downgrade trust, and only root betrayals trigger excision. Most disagreement does not even register as reputation-affecting.
• The revocation cascade with antibody memory (§14.16) lets the network respond to genuine betrayal without requiring a central authority, while preserving the record so that the same betrayal cannot recur unnoticed.

Together these primitives make personality-attacks structurally unprofitable. Negative attestations only land when their authors are in the querier’s trust set, when the stamp is anchored to specific evidence, and when the stamp-graph topology does not show collusion. The protocol does not ban personality-attacks; it makes them economically irrational.

By the same gradient, concept-attacks become productive market behaviour. Launch a competing chain at a lower tax rate, watch workers migrate, force the incumbent to improve or lose. Disagreement becomes a fork; the fork either thrives or doesn’t; the market resolves the question without anyone needing to be silenced or excluded. The Forking Rule (§6.7) is the concept-attack channel; the reputation layer is the personality-attack firewall. They compose.

The economic gradient pushes everyone toward concept-attacks because that is where the rewards are. Same shift as Yegge’s framing on cheating: “designed to make fraud unprofitable, not impossible”. The protocol does not enforce civility; it makes civility the economically rational behaviour and personality-attack the structurally unprofitable one.

This is the post-Westphalian equilibrium. Not “we agree on everything” — that is impossible and undesirable. Not “we agree to disagree politely” — that is the Westphalian failure mode where false friendship masks real betrayal. The new equilibrium is: we disagree fiercely about concepts, in public, with full record, and we cooperate fully on infrastructure, because the protocol makes both behaviours rational simultaneously.

The subsections that follow specify the primitives in detail. The reader who has internalised this reframe will see that each primitive is a piece of the same answer to the same question: how do parties who disagree fundamentally about important things continue to occupy the same physical space without escalating to violence, when the substrate of communication and coordination is computational rather than diplomatic? The answer is computable trust, distributed adjudication, and severity-precise response.

14.1 The Foundational Inversion

In every other universe of the character sheet — Skills, Work, Currency, Inventory, Group Vectors — the worker carries their own chain. Reputation is the singular exception, and the exception is the entire point.

A worker carries no reputation about themselves. Reputation lives only in stamps that others have written. What the worker carries is the dual: the stamps they themselves have written about others.

This inversion is not a policy choice; it is a structural property enforced at the data layer. In the Wasteland implementation, the constraint is encoded as a SQL CHECK on the stamps table: an author cannot be the subject. The yearbook rule is database-level, not norm-level. Yegge frames the principle thus: “Your reputation is what others write about you, not what you claim about yourself”, and contrasts it explicitly with self-reported reputation systems where everyone curates their own page.

The Inter-Town Constitution states the same principle as philosophy:

The Inversion: You cannot write in your own yearbook. You can only sign others’ yearbooks.

Your reputation = Σ(attestations_from_trusted_sources) Your value_chain = geometry(your_stamps)

∴ To know someone: read their stamps, not their bio.

The protocol-level implication: every reputation query is a network query. When the protocol asks “what is this worker’s reputation?”, it does not look up a row in her profile. It walks the graph of chains she has touched, filters by chains the querier trusts, and aggregates the attestations. She has no say in what is returned. She cannot delete bad stamps; she cannot inflate good ones. Her only reputation signal that she herself controls is who she has stamped, which reveals her values rather than her performance.

This also resolves Steve’s open question from the Wasteland post: “Who owns the stamps. We have solutions; it’ll all get worked out”. The answer is three-way custody, not single-owner: the subject does not own them (no self-attestation); the author retains a record of having written them (the dual-carry); the chains store them (queryable by anyone who trusts the chain). No single party can repossess them or unilaterally rewrite them.

14.2 The Multi-Dimensional Stamp

The stamp is the atomic unit of reputation. It is not a scalar score and it is not a binary pass/fail. The Wasteland post is explicit: “A stamp is a multi-dimensional attestation: quality, reliability, creativity, each scored independently”.

Wasteland Production Schema (v0.3 Canonical)

The stamps table in wl_commons carries five fields beyond the standard relational keys (author, subject, evidence anchor, timestamps):

• valence — JSON vector of independently-scored dimensions. The Wasteland defaults to {quality, reliability, creativity}, but the JSON column means chains can extend the dimension set without schema migration.
• confidence — how sure is the author of their attestation? A maintainer who reviewed the work line-by-line gets higher confidence than one who spot-checked.
• severity — is this a leaf task or a root architectural decision? A stamp on a typo fix carries less weight than a stamp on a system redesign. Severity is the structural-importance multiplier.
• author — who issued the stamp. Carried back to the author’s chain as well as forward to the subject’s reputation network.
• subject — who the stamp is about. Constrained: CHECK (NOT(author = subject)).

Every stamp is anchored to a specific completion record, which is itself anchored to a specific wanted item. The Wasteland post:

“The stamp is anchored to the specific completion — the specific evidence — so reputation is always traceable back to real work”.

This is the Sigstore-class property applied to reputation. Every reputation claim is cryptographically traceable through the chain back to its evidentiary basis. No claim is free-floating; no claim is unattested.

The Reputation Tensor (v0.3 Extension)

The Wasteland’s three-dimensional stamp is the production starting point. The Inter-Town Constitution proposes a more general formulation: every action emits a multi-dimensional signal and every observer evaluates that signal on their own independent axes.

The same action may be read differently by different observers:

action: Town_A → Town_B: "Your architecture is wrong."

Town_B.eval (the subject of the criticism):
  leaf_tribal:    -20   (they attacked us)
  trunk_honest:   +30   (they said what they see)
  root_intent:   +15   (they want the project to work)

Town_C.eval (an ally of B):
  leaf_tribal:    -10   (attacked our friend)
  honest_index:   +25   (they're straight shooters)

Town_D.eval (neutral observer):
  competence:    +20   (they have standards)
  trust_floor:    +35   (what they say, they mean)

The action is one event. The reputation effect is multiple, observer-dependent, and positional. No universal reputation exists. There is no scalar score that captures “what Town_A is.” There is only the geometry of stamps as read through each observer’s trust set.

v0.3 makes the tensor explicit by allowing chains to declare their own valence dimensions in their Anchor Block (§7.3). A construction chain may stamp on safety_diligence, code_compliance, subbie_coordination; a research chain may stamp on theoretical_novelty, experimental_rigor, reproducibility. The protocol does not impose a fixed taxonomy; it provides the primitive of multi-dimensional attestation and lets chains specify their own dimensional vocabulary. Cross-chain reputation queries either project into a shared subspace or carry the chain-local dimensions through to the querier with provenance attached.

14.3 The Four-Stage Lifecycle

The Wasteland defines the canonical lifecycle for any work item, and v0.3 of HOP adopts it directly:

1. Open — the work has been posted to the wanted board. No one has claimed it.
2. Claimed — a rig has marked the work as theirs. Other rigs can see who’s working on it.
3. In review — the worker has submitted evidence of completion. A validator is examining it.
4. Completed — a validator has stamped the work. The completion is permanent; the stamp is issued.

The Wasteland post explains: “When a rig claims an item, it’s marked as theirs — other rigs can see who’s working on what, preventing duplicate effort”.

There is also a fifth path: open-bounty work, where no one claims and multiple rigs work in parallel. The first submission a validator stamps closes the item; other submissions become orphan completions, preserved but not stamped. This is the winner-takes-all mode (§2.6) operationalised at the schema level.

14.4 The Trust Ladder

The Wasteland trust model is a four-level ladder. Trust governs what a rig is allowed to do; trust is earned through stamped completions.

Per the WASTELAND.md schema:

Level	Name	Capabilities (planned)
0	Registered	Browse, post
1	Participant	Claim, submit completions
2	Contributor	Proven work history
3	Maintainer	Validate and stamp others’ work

The Wasteland post:

“Trust levels gate what you can do”. “This creates a natural apprenticeship path: do good work, get stamped, eventually become someone who stamps others”.

The ladder is the protocol-level mechanism for the apprenticeship gradient that Beane’s three Cs require (§4.4). New rigs are mentored by stamped maintainers; eventually the new rigs accumulate enough validated work to themselves become maintainers; the next generation finds them. The reputation system is the apprenticeship system. They are not two layers; they are the same layer viewed from two angles.

v0.3 extends the ladder in two ways:

1. Per-chain trust levels. A worker may be a Maintainer on a Tier-1 bank’s chain but only a Participant on a ride-share cooperative’s chain. Trust is not global; it is chain-local. Cross-chain trust can be granted via federation treaties (§7.5) that mutually recognise stamping authority across chains, but the default is non-portable.

2. The Witness/Maintainer split. The Wasteland’s Maintainer role does both work-quality validation and chain integrity validation. At scale these separate (§8.13): the Witness validates blocks (does the prev_hash point to a valid prior block, is the signature valid, is the content hash correct), the Maintainer validates work (is the deliverable good, does it meet the standard). v0.3 introduces the formal split with chain-configurable role-assignment.

14.5 The Structured Reputation Profile

A worker’s reputation is not a single number. The Wasteland post lays this out precisely:

“Over time, a rig builds a profile: they’re great at Go but mediocre at frontend. They’re highly reliable but not particularly creative. They crush small tasks but struggle with epics”.

The reputation is a structured, evidence-backed work history. It can be queried along many axes: by skill, by domain, by stamp dimension, by chain, by time period, by validator identity. A Worker-Agent (§5.2) evaluating bids for a posting can ask precisely the question that matters: “show me applicants with ≥5 stamps on python from sol/earth/engineering/ Maintainers in the last 24 months, where the average reliability valence exceeds 0.8.”

This composability is what makes the reputation system useful rather than ornamental. LinkedIn’s reputation is a list of self-asserted titles and a count of endorsements; the structured query above is impossible. The Wasteland reputation is queryable as a database because it is a database.

Recent Trend as First-Class Field

Reputation must show direction. A worker improving from 0.7 to 0.9 over twelve months is a different bet than one declining from 0.95 to 0.85, even though the current average is similar. v0.3 surfaces the recent_trend signal as a first-class derived field, computed from the rolling window of recent stamps. The Worker-Agent can weight trend explicitly when evaluating bids; chains can publish trend in their reputation queries.

Redemption Arcs

A failure is not a permanent mark. The HOP-internal spec made this explicit: “Bad performance can be overcome: Failures recorded but redeemable. New good work proves growth. Not permanent social credit score”. v0.3 formalises the redemption arc as a queryable property: a failure followed by demonstrated growth in the same domain becomes evidence of capability, not evidence of incapacity. The chain incentivises recovery, not punishment.

This is the structural difference between HOP and a social credit system. A social credit system is monotonic: bad acts accumulate, never decay, never redeem. HOP is float: bad acts dim the worker’s signal but do not cliff it (the Basal Rate Law, §6.5), and new good acts re-light the signal directly.

14.6 The Yearbook Rule, Enforced at the Database Level

The Wasteland’s implementation of the no-self-attestation rule is the cleanest possible: it is a SQL CHECK constraint on the stamps table.

-- From wl_commons stamps table definition
CHECK (NOT(author = subject))

The yearbook rule is not policy. It is not norm. It is database invariant. An INSERT that violates it is rejected at the database engine. The constraint propagates with the schema across all DoltHub forks, all federation peers, all derived chains. No implementation of the Wasteland schema can break the yearbook rule without first rewriting the schema, which is a public, visible, signed act.

This is the right level for this invariant. Yegge’s framing makes the principle vivid:

“You can sign other people’s pages, but you can’t sign your own”. “This is the fundamental difference between the Wasteland and LinkedIn”.

v0.3 extends the database-level enforcement to a wider set of integrity constraints, all enforced at the schema level rather than at the application level. The reputation system gets to be safe-by-construction rather than safe-by-discipline.

14.7 The Universe Topology — Why Reputation Has No Direct Exchange

The Five Universes (Skills, Work, Currency, Inventory, plus Group Vectors as the v0.2 sixth) sit on the character sheet. The Inter-Town Constitution maps the exchange topology between them:

EXCHANGE_TOPOLOGY:
  🎯 ↔ ⚡        Skills prove through Work
  ⚡ ↔ 💰        Work converts to Currency
  💰 ↔ 📦        Currency buys Inventory
  📦 ↔ 🎯        Inventory enables Skills (tools)

  🪞 ↔ ???       Reputation: NO DIRECT EXCHANGE

Reputation sits outside the ring. There is no direct trade between reputation and anything else. The Inter-Town Constitution is explicit:

Reputation cannot be bought (💰 → 🪞 ❌) Reputation cannot be transferred (🪞 → 🪞 ❌) Reputation cannot be self-minted (self → 🪞 ❌)

Reputation can only be emitted (you stamp others, which reveals your values), received (others stamp you, which reflects your work), or survived (Clean Margins — i.e. the dispute and integrity layer — does not revoke you).

This structural property has a critical legal/regulatory consequence: reputation cannot be securitised. A holder cannot transfer their reputation to another party for value; a third party cannot create a derivative instrument on the back of someone’s reputation. This places the reputation primitive outside the surface area of financial regulation. It is not a security, not a token, not a tradable asset. It is a measurement of demonstrated capability, held in trust by the network.

By contrast, Beans (§6.3) — which are the mentorship subset of beads — are transferable within their narrow definition (the mentor’s discount on conversion can be staked, settled, clawed back). Beans are deferred-comp-with-clawback, which is a regulated category. Reputation is not. The architecture cleanly separates the two, so the regulatory framing for each is unambiguous.

14.8 The Reputation Network — Trust-Filtered Queries

There is no universal reputation. There are only trust-filtered queries.

The Inter-Town Constitution states this as the YEARBOOK_RULE:

yearbook(Town_A) = [ stamps WHERE
    subject == Town_A AND
    author IN observer.trust_set ]

When an observer queries Town_A’s reputation, the network returns only the stamps written by authors the observer trusts. Two different observers query the same subject and receive different answers — not because the data is inconsistent, but because their trust sets are different. This is intentional and correct. Reputation is positional in value-space, not absolute.

This composes naturally with the chain hierarchy (§7.4). A Tier-1 bank query against the worker’s reputation pulls stamps from the banking chain and adjacent regulated industries. A community-organisation query against the same worker’s reputation pulls stamps from her volunteer work, her language-skills attestations, her social-services contributions. Both queries are valid. Both produce true reputation. The two answers are different.

The Worker-Agent (§5.2) operates inside this topology by selecting which trust set to query the reputation network against. The trust set is configured by the poster — explicitly or by reference to a federation treaty — and the Worker-Agent honours that scope.

Cross-Chain Skill-Vector Translation

A 7.2 software-engineering stamp on the Accenture chain does not auto-translate to 7.2 on the public engineering chain. Different chains have different stamping standards; different validators have different rigour profiles. v0.3 introduces a cross-chain translation layer: the Accenture chain can sign a translation attestation stating “our 7.2 maps to public 6.8 with confidence interval [6.5, 7.0]”. Workers can then carry both attestations on their Skillchain, and a Worker-Agent querying the public chain’s reputation can choose to honour the translation or not.

This is structurally similar to currency federation (§6.6): there is no universal exchange rate, only bilateral and multilateral translation agreements between chains. The chains that participate in many translation treaties become the bridge towns, which is itself a reputation-bearing position (§14.13).

14.9 The True-Enemy Preference Order

The protocol provides a reading rule for how to weight attestations. The naive view is: stamps from your friends are more trustworthy than stamps from your enemies. The Inter-Town Constitution argues the opposite:

Listen to what their ENEMIES say. If enemies say “they’re honest”, BELIEVE IT.

enemy_attestation.weight > friend_attestation.weight

The reasoning: a stamp issued at cost to the author carries more signal than a stamp issued cheaply. A friend stamping a friend is cheap; the cost of being wrong is low. An enemy stamping an enemy with a positive attestation is expensive; the cost of saying something true about someone you dislike is the cost of admitting they have a quality you cannot deny. The expensive stamp is the honest one.

This produces the four-quadrant preference order:

1. TRUE FRIEND — aligned and honest (rare and valuable)
2. TRUE ENEMY — hostile and honest (valuable, more trustworthy than ambiguous friends)
3. FALSE ENEMY — hostile and dishonest (predictable, manageable)
4. FALSE FRIEND — aligned and dishonest (dangerous; the worst position because you cannot see the knife)

The Wasteland’s cheating-detection — that “collusion rings have a distinctive topology — lots of mutual stamping, sharp boundaries, no outside critics” — is the operational form of this principle. A reputation built only on friend-stamps and zero enemy-stamps is structurally suspect, even if every individual stamp is technically valid. The geometry of the stamp graph reveals the collusion that no individual stamp betrays.

v0.3 specifies this as a queryable reputation-weighting function. A Worker-Agent can request “the worker’s reputation, weighted by author-distance,” where author-distance is computed across the chain’s stamp graph as the geodesic distance from the author to the subject in the value-network. Distant honest attestations weight higher than near friendly attestations. The collusion ring’s mutual-stamping pattern collapses when the geodesic-distance weighting is applied.

14.10 The Trust Floor — Cross-Cluster Cooperation

Two parties can disagree at the leaf level — different politics, different aesthetics, different tribal markers — and still cooperate at the trunk. The Inter-Town Constitution names this primitive the trust floor: the depth at which two value-distant parties find shared ground.

Trust_floor is the thing that lets enemies trade. Trust_floor is built from honest signals.

“I don’t like you, but I believe you.”

The trust floor is what makes federation possible between chains that don’t share politics. A Tier-1 bank and a ride-share cooperative have radically different governance models, customer bases, and political positions. They can still federate (share worker reputation, recognise stamps, sign cross-chain treaties) because they have a trust floor: each can verify the other’s chain integrity, each can audit the other’s stamping consistency, each can confirm the other’s signature provenance. The chains do not need to like each other to trade; they need to trust each other’s honesty about their own data.

v0.3 introduces the trust floor as a measurable property of every federation treaty. Treaties declare the floor explicitly: “we recognise each other’s quality valence as comparable; we do not recognise each other’s creativity valence because we don’t share aesthetic standards.” The treaty surface is the shape of the trust floor.

Workers benefit from this because their reputation becomes portable along the federation graph at the depths the treaties recognise. A worker’s reliability may be portable everywhere; their creativity may be portable only within aesthetic-aligned clusters. The portability is structural, not unilateral.

14.11 The $Variable — Substrate Neutrality

The protocol does not care what kind of entity is participating. The Inter-Town Constitution states this as the $Variable invariant:

Town := $Variable WHERE $Variable ∈ {
    single_human,              "steveyegge"
    single_agent,               polecat with persistent ID
    team,                       humans + agents
    open_source_swarm,          10,000 contributors
    corporation,                legal entity
    federation,                 towns of towns
}

PROTOCOL_INVARIANT:
  The protocol does not care what $Variable contains.

  Handshake: identical
  Stamps: identical
  Cost function: identical (adjusted for scale)
  Clean Margins: identical

One human can trade with thousand-agent swarm.
Reputation is portable across entity types.

This is the Neutrality Law (§1.1) as it applies specifically to reputation. The stamp issued by a human about an AI agent is structurally identical to the stamp issued by an AI agent about a corporation, which is identical to the stamp issued by a corporation about a federation. The protocol primitives do not distinguish.

In practice, individual chains may impose substrate constraints in their Anchor Block — a chain may declare that only human-operated rigs can issue stamps for legal-review work, for instance — but those constraints are chain-policy, not protocol-property. The protocol primitive is substrate-neutral; chains layer policy on top.

The consequence for AI agents: agents accumulate reputation the same way humans do. A polecat with a persistent ID that has shipped 400 stamped pull requests is more reputationally credentialed on the chain than a human who has shipped 10. This is the Neutrality Law refusing to give humans a structural advantage; capability is the only basis for matching.

This also resolves the second open question Steve raised in the Wasteland post: “the personal/work identity — right now your identity is global across all Wastelands”. The HOP namespace hierarchy (§7) provides the resolution. A single global rig identity sits at the root namespace (sol/earth/identity/); chain-local pseudonyms can be derived from it for contexts where global identity would expose more than the worker chooses. A worker’s institutional work signs to her bank-namespace pseudonym; her community-organisation work signs to her cooperative-namespace pseudonym; both link back to her root identity through BBS+ derived proofs (v0.2 cryptography) that prove same-person without revealing the linkage. The root identity remains global; the chain-presented identity is local.

14.12 The Stamp Geometry — What Stamps Reveal About the Stamper

A worker’s reputation is in their incoming stamps. But the stamps they have written reveal something equally important: their values.

The Inter-Town Constitution:

STAMP_GEOMETRY: stamps_positive(A) → reveals: A.allies stamps_negative(A) → reveals: A.enemies stamps_absent(A) → reveals: A.blind_spots

pattern(A.stamps) = A.value_chain

The geometry of who you have stamped, and how, is itself a queryable property of the chain. A Worker-Agent evaluating a senior worker as a potential mentor can look not just at the mentor’s received reputation but at the pattern of who they have stamped positively. A senior worker who has only ever stamped people from their own narrow demographic reveals a blind spot — possibly intentional, possibly not, but visible regardless.

This composes with the disadvantage multiplier (§6.3.5). A mentor whose stamp-pattern shows historical engagement with hard-to-bet-on workers earns a credibility premium when claiming the disadvantage-multiplier payout. The pattern is its own attestation.

v0.3 introduces stamp-geometry queries as first-class protocol operations: “show me the diversity-of-recipients of this worker’s positive stamps over the last 24 months, projected into the demographic/skill subspace.” This is the queryable form of “do they walk their talk.”

14.13 The Bridge Town — Reputation From Being Trusted by Both Sides

A town (or rig, or chain) that is trusted by two otherwise-opposed clusters has a special protocol position: the bridge. The Inter-Town Constitution:

BRIDGE_VALUE: Town trusted by Cluster_Red AND Cluster_Blue: = valuable routing infrastructure = bridge_builder = earns reputation from both sides

“The translator is sacred in the desert.”

Bridge towns earn reputation specifically because they hold cross-cluster trust. The reputation primitive distinguishes single-cluster credibility from cross-cluster credibility, and the latter is more valuable in many contexts (federation negotiation, dispute resolution between hostile chains, translation of skill-vectors between chains with different standards).

This is the protocol-level mechanism that rewards the work of cultural and structural translation. It is also a piece of the answer to why HOP can operate across politically-divided chains: the protocol creates economic incentive for bridges, and the bridges hold the federation together.

For Australia’s geopolitical position (the recurring theme through this spec): writing the standard treaty template that other countries’ chains adopt is exactly the bridge-town move at international scale. Australia becomes the bridge between ASEAN labour chains, Five-Eyes regulatory chains, EU privacy chains, and the bridge earns reputation from all sides without owning any of them. The Zollverein move (§6.6) operates through this primitive.

14.14 Cold Start — Pre-Seeding From Public Evidence

The Wasteland is already operating a cold-start strategy that v0.3 of HOP adopts directly. Yegge:

“We pre-seeded them with data from GitHub, from the top ~10k contributors by whatever metric Claude thought appropriate”.

The principle: public data is public data. GitHub’s API explicitly allows reading contributor profiles; the Wasteland post notes that scraping is supported by GitHub’s ToS with strong legal precedents. The scrape produces provisional Skillchains for tens of thousands of identifiable contributors before any of them have opted in. When a user opts in, they prove control of their cross-linked identities and the escrowed Skillchain transfers to their custody.

v0.3 extends this beyond GitHub. The full target list (§8.17): GitHub, Kaggle, Stack Overflow, Hugging Face, Papers With Code, npm, PyPI, crates.io, arXiv, HackerOne, Bugcrowd, Patreon, GitHub Sponsors, OpenCollective. Each is a public ledger of skill demonstration. Identity resolution across these platforms produces a starting state with millions of pre-populated Skillchains.

Yegge frames the philosophical argument precisely:

“If the Wasteland is giving you reputational credit for work you’re getting verified as having completed, then why not give everyone partial credit for verified work they’ve already completed?”.

The provisional nature is critical. Cold-start blocks are clearly marked provenance: cold_start_inferred. Their cryptographic weight is lower than directly-attested blocks. Users can dispute or replace them once they take custody. The bootstrap is evidence, not truth. But it solves the cold-start problem without anyone needing to create their account from scratch.

This is also the LinkedIn-killer dynamic. The user’s first interaction with HOP is “your work history is already here, accurate, portable, sovereign — you just need to claim it.” Friction near zero. Value of joining high. Network effect bootstraps from the first user.

14.15 Cheating Detection — Stamp Graph Topology

Yegge consulted Trust & Safety experts on the cheating problem. His summary:

“Collusion rings have a distinctive topology — lots of mutual stamping, sharp boundaries, no outside critics”.

The Wasteland system, in his words: “designed to make fraud unprofitable, not impossible”.

This is the right framing. A protocol that tries to make fraud impossible will either be too restrictive to be useful, or will fail (because adversaries are creative). A protocol that makes fraud unprofitable shifts the equilibrium: cheating still exists, but the expected return is negative, so rational adversaries stop trying.

The mechanism is graph topology analysis. A reputation built only on stamps from a small mutually-reinforcing cluster is structurally suspect. The Worker-Agent (§5.2) evaluating bids can detect this pattern automatically by computing graph statistics on the stamp network. Sharp boundaries (no stamps from outside the cluster), high mutual density (everyone in the cluster has stamped everyone else), and absence of negative or critical attestations from non-cluster sources all flag the suspect cluster.

v0.3 makes graph-topology analysis a first-class protocol operation. Reputation queries can include a topology_filter parameter that excludes collusion-suspect patterns. Chains can declare their tolerance threshold in the Anchor Block.

The Christiano (2014) trust-matrix machinery (§6.3.4) is the cryptographic version of this same defence — pairwise discount caps that prevent reciprocal Bean farming. The reputation-graph topology analysis is the more general version that applies to all stamps, not just mentorship Beans.

14.16 Clean Margins — The Severity-Tiered Response Protocol

The reputation system has to respond to bad behaviour, but the response has to be precise. Over-response punishes leaf disagreements as if they were existential betrayals; under-response lets actual betrayal compound until the chain becomes untrustworthy. The Inter-Town Communication Constitution names the principle directly, borrowing the medical metaphor from oncology:

Clean margins: surgical boundary with no cancer cells.

The goal is not to remove the tumor. The goal is to remove the tumor AND leave only healthy tissue.

Cut too little → cancer returns. Cut too much → patient dies.

Clean margins: precise excision.

Applied to reputation: the response to a violation must match the severity of the violation. A worker who misses a deadline is not the same as a worker who lied about credentials; a chain operator who set the off-ramp tax slightly too high is not the same as a chain operator who pocketed escrowed funds. Treating them the same is the failure mode that destroys reputation systems — either by over-punishment (which makes participation in the network too risky for honest workers) or by under-response (which lets genuine fraud accumulate).

The Three-Tier Severity Model

The protocol distinguishes three severity bands. Each band has a distinct response surface:

Leaf-level violation (severity: note)

The party disagreed about implementation, missed a deadline, chose a different tool, or expressed a preference the counterparty did not share. These are recoverable. The response is to record the event as a negative-valence stamp on the relevant dimension of the multi-dimensional tensor (§14.2), and to let the stamp accumulate or fade like any other reputational signal. No revocation. No cascade. Note it, move on.

The reputation network is robust to leaf-level negative signals. Workers and chains with sustained positive reputation can absorb occasional negative stamps without their overall trust set dropping meaningfully. This is the protocol’s tolerance for ordinary disagreement and ordinary failure. Most reputation events are leaf-level. The protocol is designed to absorb them gracefully.

Branch-level violation (severity: downgrade)

The party broke a promise, failed to deliver, or made an honest mistake on something material. The work was not done, or the work was done badly enough that downstream consequences resulted. The response is to downgrade trust — the author of the negative stamp explicitly lowers their attested confidence in the subject across the relevant dimensions, and propagates the signal to their trust network. No excision. Trust diminished, recovery path open.

The branch-level response is the most common substantial action in the protocol. It is the equivalent of “I will work with this person again, but on smaller projects with closer oversight, until they demonstrate they have addressed the underlying issue.” Recovery is possible and expected. A worker downgraded for missed deadlines can rebuild trust by delivering reliably for a sustained period; the original downgrade stamp remains in the record but its weight decays relative to the new positive accumulation.

Root-level violation (severity: excise)

The party poisoned shared infrastructure, lied about fundamental values, betrayed at the trust_floor level. These are violations that the protocol cannot tolerate without becoming complicit. Examples: a chain operator who absconded with escrowed funds, a worker who forged credentials to obtain access to vulnerable populations, a validator who collaborated with a collusion ring to fraudulently inflate reputation. The response is full excision — the author broadcasts a revocation cascade (§14.17) to the trust network, and each peer independently decides whether to follow the cascade.

Root-level violations trigger the Antibody Memory mechanism (§14.17): the betrayal is recorded permanently in the burn list of every chain that follows the cascade, and that record never expires.

The Precision Rule

The protocol enforces a severity-matching rule between the violation and the response:

severity(violation) must match severity(response)

leaf    → leaf_response    (note it, move on)
branch  → branch_response  (downgrade trust, allow rebuild)
root    → root_response    (excise, cascade, remember)

This is the database-level analogue of the medical principle: over-cutting kills the patient, under-cutting lets cancer spread, clean margins require precision. The protocol enforces this not by adjudicating which severity-band a violation belongs to (that judgement is made by the author of the negative stamp) but by exposing the severity field on every stamp, so that observers can audit whether responses were proportionate. A chain that consistently escalates leaf-level disagreements to root-level excisions develops a visible pattern of over-response in its own stamp geometry, which other chains can detect and weight accordingly. The protocol does not prevent over-response; it makes over-response visible.

The same applies in reverse: a chain that fails to respond to root-level betrayals develops a visible pattern of under-response, and other chains can flag this as a trust-floor concern.

Why This Matters for Ideological Conflict

Most ideological disagreement lives at the leaf and branch levels. Vim vs Emacs is a leaf disagreement. “We value agent autonomy” vs “we value agent reliability” is a branch disagreement. Workers and chains can hold strong opposing views on these and still trade, federate, and cooperate, because the severity-tiered response keeps disagreement at the level where it belongs. The protocol does not require agreement at the leaf; it requires only that root-level cooperation be preserved.

This is the operational form of the trust floor (§14.10): two parties find the depth at which they share ground, and trade there, even while disagreeing freely above. Clean margins is what makes this stable. Without it, every leaf disagreement risks escalating into a root-level excision attempt, and the network fragments into hostile clusters. With it, disagreement at the leaf is the protocol’s normal operating mode.

14.17 Revocation Cascade and Antibody Memory

When a root-level betrayal is detected, the protocol provides a distributed response mechanism that does not require a central authority. The Inter-Town Communication Constitution specifies it as the Revocation Cascade.

The Cascade Sequence

The response unfolds in four steps:

Step 1 — The negative attestation. The party that detected the betrayal signs a negative stamp with severity root_violation, attaching the evidence (typically a cryptographic hash linking to the specific completion records or chain events that constitute the violation). The stamp is anchored to the betrayer’s identity and broadcast to the network.

Step 2 — The broadcast. The betrayed party broadcasts a message to their trust network: “We trusted [betrayer]. We were wrong. Here is the evidence. Review your positive attestations. Consider revocation.” The broadcast is not a command; it is an invitation to independent evaluation.

Step 3 — Independent evaluation. Each town in the network independently evaluates the cascade. The questions each town asks itself: Do I trust the betrayed party’s judgement? Does the evidence hold? What was my exposure to the betrayer? No central authority adjudicates. Each town’s decision is its own.

Step 4 — The cascade. If consensus emerges across the trust network, the cascade propagates. Town X revokes its positive attestations of the betrayer. Town Q, which had been relying on Town X’s attestation as part of its trust set, now sees fewer trusted signatures supporting the betrayer. Town R re-evaluates accordingly. The network routes around damage. The betrayer becomes progressively less visible to the trust networks that have followed the cascade, without any party having dictated the outcome.

Note that cascade is not automatic. Each town’s revocation is its own decision; consensus emerges or fails to emerge through independent evaluation. A weak or evidence-thin cascade fails to propagate; a strong, evidence-anchored cascade against a clear root-level betrayal propagates widely. This is the protocol’s structural answer to the question “who decides?” — nobody decides; the network decides through aggregated independent decisions.

Antibody Memory

When a town revokes its positive attestations as part of a cascade, the record of the revocation goes into the town’s burn list — a persistent record of “these parties have been excised, and here is why and when.” From the Inter-Town Communication Constitution:

Burned list: Town_B.burned = [{town: Town_A, reason: “root_violation”, date: 2026-01-12, evidence: 0x7f…}]

This never expires.

The burn list is permanent by design. Once a town has been burned by another, the record of that excision persists indefinitely. This is not punishment — the burned party is not prevented from continuing to exist, from doing work, or from earning new attestations from parties outside the burning trust set. What the burn list prevents is amnesia. The same root-level betrayal cannot recur unnoticed, because the network remembers.

When a new party evaluates whether to trust a previously-burned town, the burn-list memory propagates to them through their own trust set: any chain in their trust set that has the burned town on its burn list will emit a warning when the new party queries the burned town’s reputation. The warning is not auto-reject — the new party can still choose to engage, with full knowledge of the historical record. But informed engagement replaces blind engagement.

Recovery Path

Antibody memory is permanent, but excision is not death. From the Inter-Town Communication Constitution:

Excised town can still:

• Receive messages (basal dignity)
• Do work (earn new attestations)
• Build new trust (from scratch)
• Exist (no death penalty)

Excised town cannot:

• Erase the record
• Force forgiveness
• Transfer old reputation
• Pretend it didn’t happen

The recovery path is open but costly. An excised town can rebuild reputation by:

1. Acknowledging the violation: signing a self-stamp that does not deny what happened (which the protocol permits even though it cannot count as positive reputation per §14.6).
2. Doing new work: emitting beads, completing tasks, earning attestations from chains outside the network that performed the original cascade.
3. Earning stamps from new trust clusters: building trust floor with a different cluster of chains, demonstrating sustained capability and consistency.
4. Eventually bridging back: if and only if the new trust accumulation is strong enough that some chain in the original network is willing to reconsider, the bridge becomes possible.

This is hard. It is supposed to be hard. Root-level betrayals are serious, and the recovery path should require sustained demonstration that the underlying problem has been addressed. But it is not death. The protocol’s Basal Rate Law (§6.5) guarantees that no party can be starved out of the network entirely, even after excision. The Five Laws apply universally; excision does not suspend them.

Why This Resolves the Westphalian Sanctions Problem

The classical Westphalian response to state-level betrayal was either war or no response. Modern international institutions (UN, ICC, sanctions regimes) attempted to introduce a middle band, but they require a central authority that the system’s most powerful members can capture. UN sanctions require Security Council consensus, which is structurally available only to states the great powers agree on punishing. The result is uneven application — small offenders are sanctioned for actions that great powers commit without consequence.

The Revocation Cascade is structurally different because it requires no central authority. Each party in the network evaluates independently. The cascade succeeds or fails based on the aggregate of independent evaluations, weighted by trust relationships that the network has built up over time. There is no Security Council. There is no veto. There is only the question: when [a chain] presents evidence of betrayal by [another chain], how many chains independently revoke their positive attestations? If enough do, the betrayer becomes progressively less visible in the federation. If not enough do, the cascade fails and the betrayer continues operating with reputation intact.

This is the post-Westphalian sanctions primitive. It is distributed, evidence-anchored, trust-weighted, and uncapturable by single-party veto. It is also recoverable — burned parties can rebuild, the recovery path is open, the protocol does not impose permanent excommunication. The contrast with both the Westphalian “war or nothing” binary and the modern “sanctions are political weapons” failure mode is sharp.

14.18 Validators, Maintainers, and the Witness

The Wasteland reputation system involves three roles that v0.3 formalises:

• Validators (Wasteland Maintainers): rigs with sufficient accumulated trust to issue stamps. They evaluate completion evidence and produce the multi-dimensional valence vector. In Yegge’s framing, every rig with maintainer-level trust is a validator; the validator population is not a fixed list but a dynamic property of the trust ladder.
• Witnesses (v0.3 introduction): programmatic infrastructure that validates the block-level structural integrity of stamps. Not LLM-driven. The Witness checks that the stamp’s prev_hash links correctly, that the signatures are valid, that the content hash is correct. Where the Validator is “is this work good?”, the Witness is “is this stamp well-formed?”
• The Stamp Graph Itself: emergent property of the network, queried via topology analysis. Not an agent; not a role. A computed structure that surfaces collusion patterns and bridge positions.

Together these three constitute the reputation-integrity layer. v0.1 collapses the Validator and Witness into a single agent (the chain’s Validator from §5.5). v0.2 keeps that collapse. v0.3 separates them, because at scale the structural-integrity work scales differently from the semantic-quality work and benefits from being a separate role.

14.19 What the Worker Carries vs What the Org Keeps — The Reputation Boundary

This subsection anticipates the planned new section on the carry-boundary (referenced as forthcoming in working notes). The reputation primitive is the cleanest case study for the boundary, because reputation has the most extreme asymmetry: the worker carries nothing about themselves, and the network carries everything.

The worker carries:

• The stamps they themselves have written about others (their authored stamps).
• Proofs that they are the cryptographic subject of stamps held in the network (i.e. the keypair that the network’s stamps reference).
• Their own Skillchain and Workchain history, which contains the evidence that stamps were issued for.

The network (chains, federation peers, Wasteland commons) carries:

• The stamps written about the worker.
• The validator identities and their stamping history.
• The chain integrity proofs that verify the stamps were not tampered with.
• The topology metadata that supports collusion detection.

The institution keeps (in the Banks as Trust Sources pattern, §8.9):

• The raw operational record that the stamped character block was an abstraction of.
• The institutional signing key that attested the stamp.
• The legal-record obligations for regulatory compliance.

The interface between these three is the dual-signed stamp itself: signed by the validator, anchored to the completion record, queryable by anyone who trusts the validator’s chain. The worker presents their proof of subjecthood (here is my keypair that the network’s stamps reference); the network returns the stamps filtered by the querier’s trust set; the institution stands behind its signature without holding the worker’s reputation hostage.

This is the cleanest implementation of the principle that nothing is locked in any single party. The worker cannot lose their reputation by leaving an employer (their employer’s stamps remain in the network). The employer cannot weaponise a worker’s reputation against them (the employer cannot delete the stamps they issued; they can only refuse to issue new ones). The chain cannot capture the reputation (the chain stores it but cannot claim it as property — anyone with a fork has the same stamps).

14.20 Capability as Multiplicative Across the Five Universes

A subtle but load-bearing point from the Inter-Town Communication Constitution that has not yet been said explicitly in this section. The Five Universes (Skills, Work, Currency, Inventory, Reputation, plus Group Vectors as the v0.2 sixth) are not additive when combined into the protocol’s measure of capability. They are multiplicative.

CAPABILITY = Skills × Work × Currency × Inventory × Reputation
           (NOT: Skills + Work + Currency + Inventory + Reputation)

The distinction matters. An additive model says “high score in one universe compensates for low score in another.” A multiplicative model says “if any universe drops to zero, capability collapses in that direction regardless of how high the other universes are.”

This is the structural answer to several questions that the spec has been gesturing at:

Why Beane’s three Cs degrade as a unit if Connection fails. Challenge and Complexity can be present in any work environment, but without Connection the expert-novice bond does not exist and skill does not transfer. The three Cs are multiplicative inputs to skill transfer; if Connection → 0, transfer → 0 regardless of how high Challenge and Complexity are. This is what makes Beans (§6.3) load-bearing rather than ornamental: Connection is the failure mode that takes down the entire upward engine.

Why hollow reputation fails under pressure. From the Inter-Town Communication Constitution’s worked example of the Well case: a council with high Currency and Inventory (money, property) but with Reputation only from other councils (hollow stamps from a collusion ring of peers) collapses when actual pressure arrives. No one picks up when they call. The capability looked imposing in normal conditions but the Reputation dimension was hollow, and the multiplicative product is therefore zero in the dimension that mattered most: the ability to call on others for help. By contrast, an old water keeper with high Skills, high Work history, and high Reputation in the right rooms — but low Currency and Inventory — is dangerous in a way the council cannot see, because his capability product is non-zero in the dimensions that matter for the specific pressure being applied.

Why platform employment is a degenerate special case. Platform workers like classic Uber drivers have artificially high Currency (income flow), some Inventory (their car), and effectively zero Reputation (the platform controls their reputation, locks it in, and can revoke it at will). Capability = (Skills) × (Work history they cannot carry) × (Currency) × (Inventory) × (Reputation ≈ 0). The product is collapsed in the Reputation dimension regardless of how high the other universes go. This is what makes the protocol structurally anti-platform: by giving Reputation back to the worker, the protocol restores the dimension that platforms have artificially zeroed out, and the worker’s capability product can grow in all directions rather than only in the dimensions the platform permits.

Why Sera’s case is solvable. Sera (§0) has Currency = 0 and Inventory ≈ 0 at the moment she arrives in Australia at sixteen. By an additive model she has some capability — Skills and Work are non-zero, Reputation is in the wrong trust set but exists. By a multiplicative model her capability is zero in the dimensions the labour market measures, because two of the five factors are zero. The HOP system gives her a way to grow Currency and Inventory from zero (via the bootstrap grant, §6.5) while her Skills, Work, and Reputation (in chains other than the Witnesses) continue to grow over the fifteen-year dig. By thirty-one, she has non-zero values in all five universes, and her capability product becomes meaningful. The system’s role is not to give her capability; it is to provide the conditions under which her capability product can grow above zero in the dimensions the market measures.

Why This Matters for Reputation Queries

A reputation query is not asking “what is this person’s score on the Reputation axis?” It is asking “what does this person’s reputation contribute to their overall capability for the work I am considering them for?” And because capability is multiplicative, the answer depends on the intersection of Reputation with the other four universes that the query is implicitly testing.

A worker with strong reputation as a software engineer but zero recent Work in that domain has a multiplicative-zero in Work. A chain operator with strong reputation but no current Currency or Inventory to fund the chain’s operations has a multiplicative-zero in Currency. A mentor with strong reputation but whose mentees show no growth (i.e. no Skills appreciation in the mentee’s chain) has a multiplicative-zero in Skills-transferred-by-mentor.

The capability product is the protocol’s integrated measure of fitness for a specific task. Reputation queries that ignore the other universes return misleading answers. Reputation queries that include the multiplicative integration return useful ones. v0.3 implementations should expose the multiplicative integration as a queryable property of any reputation query, so that Worker-Agents (§5.2) can ask “what is this candidate’s capability product for this specific posting, projected into the dimensions that matter?”

The Stamp Geometry Connection

The capability product also clarifies what stamp geometry reveals about the stamper (§14.12). A senior worker whose stamp pattern shows engagement only with peers from their own narrow demographic cluster has zero diversity-of-recipients. The disadvantage multiplier (§6.3.5) penalises this because it indicates a blind spot in the worker’s capability for mentorship across diverse contexts. The multiplicative model makes this explicit: a senior worker’s capability to mentor is the product of (Skills, Work, Currency, Inventory, Reputation, and demonstrated cross-cluster engagement). Zero engagement across clusters means zero capability to bridge clusters, regardless of how high the other dimensions are.

14.21 v0.3 Implementation Sequencing

The Wasteland is the production reference for what’s already running. v0.3 ships incrementally on top:

1. Adopt the Wasteland schema directly. The wl_commons seven-table schema is the canonical reference. HOP-conformant chains run the Wasteland schema or a superset. The Wasteland is v0.3 phase 1, not a separate system.
2. Add the chain-namespace context. Stamps gain a chain reference (which namespace did this stamp originate from). The Wasteland is currently flat; HOP layers the chain hierarchy on top.
3. Add the trust-filtered query primitive. Reputation queries take a trust_set parameter and return only stamps from authors in that set. Default trust_set is the entire chain; richer trust sets are derived from federation treaty membership.
4. Add the topology-analysis query primitive. Reputation queries take a topology_filter parameter that excludes collusion-suspect patterns.
5. Add the trans-chain translation layer. Bilateral stamping-standard translations published by chains, queryable by Worker-Agents.
6. Add the Witness/Validator role split. Structural integrity checking becomes its own role with its own keypair, distinct from semantic quality validation.
7. Add the disadvantage-multiplier integration. Mentorship Beans (§6.3) settle against the reputation network via stamp-geometry queries: a mentor’s claim on the disadvantage multiplier validates against the diversity of their stamp-pattern.

Items 1–3 are essentially the Wasteland running with chain-namespace metadata added; they are deployable today against existing infrastructure. Items 4–6 are the v0.3 substantive extensions. Item 7 is the integration point between v0.3 reputation and v0.2 Bean Chains, and is the highest-leverage piece because it makes the upward-engine economically rational rather than purely structural.

14.22 What the Reputation Section Does Not Yet Address

Open questions deferred to v0.4 or beyond, recorded here for transparency:

• Death and inheritance. What happens to a worker’s reputation when they die? The current spec leaves it queryable indefinitely, but family members may wish to claim residual proofs, or estates may wish to license the validated capability to ongoing institutions. Unspecified.
• Reputation as collateral. §14.7 rules out reputation as a securitisable asset. But can it be staked? A worker pledging “I stake my reputation that this code review is sound” in a way that the protocol can slash on failure? The Inter-Town Constitution hints at this with the stake-cost-of-messaging primitive, but it is not yet specified. Adjacent v0.3 lane.
• Anonymous reputation. Can a worker accumulate reputation while remaining cryptographically anonymous to the validator? BBS+ derived proofs (v0.2) allow selective disclosure; whether they allow fully anonymous accumulation is a deeper question.
• Cross-substrate reputation translation. A human-issued stamp about an AI agent vs an AI-issued stamp about a human: structurally identical per §14.11, but interpretively different. Whether the protocol should expose the substrate provenance of stamps (so observers can weight them differently) or hide it (so the Neutrality Law applies maximally) is unresolved.
• Reputation decay. Should stamps from 10 years ago weight equally to stamps from last month? The recent_trend field provides partial relief, but the underlying weighting policy is unspecified. Probably chain-configurable in Anchor Block.

These are all deliberately deferred. v0.3 is bounded by what can be specified rigorously given the running Wasteland implementation as ground truth and the Inter-Town Constitution as philosophical foundation. v0.4 will widen the surface as the production implementations of v0.3 reveal what matters.