Skip to main content
The WalletWall Evidence Profile (WWEP, version 0.1, draft) is the model behind every WalletWall MCP result. It defines a structured record of what evidence was presented, from where, with what authority and with what limits, and a deterministic procedure that classifies how well that evidence supports a claim. WWEP is a WalletWall-defined profile. It is not an industry standard and has not been reviewed outside WalletWall. Its normative text is served by the server as walletwall://spec/wwep and published as spec/EVIDENCE_PROFILE.md. Where that draft text and the engine differ (it describes every object as closed, addresses as compared case-insensitively, and the content digest as the duplicate detector), this page follows the engine.
The rule behind every state. A WWEP state describes the sufficiency of presented evidence under a published policy. ESTABLISHED means “the presented evidence satisfies the policy for this claim type”, never “this is verified truth”.

From source to resolver

Everything up to the assessment happens inside WalletWall MCP, on your machine, every time you call a tool. WallPrint and the public resolver are a separate, deliberate publication step; see WallPrint.

1. Source

A source says where evidence came from. It produces or supports observations; it never asserts anything by itself.
transport.retrievedVia records how the bytes arrived (an RPC provider, an indexer, a local node). It is metadata, never authority: a balance read through a provider’s RPC endpoint is still a CHAIN_STATE reading, and the provider does not become the authority for Ethereum state. independenceGroup says which sources can fail independently. Sources in the same group count once toward corroboration, and every source that declares no group falls into one shared group, so listing several endpoints never makes them independent.

2. Observation

An observation is one reading from one source.
  • result is VALUE, NO_RESULT or ERROR. Only VALUE carries a value; the other two record that a source was consulted and produced nothing usable.
  • finality is FINALIZED, UNFINALIZED or UNKNOWN_FINALITY, as the reading declares it.
  • origin classifies the evidence, and each origin pairs with exactly one source kind:
An observation whose source kind does not match its origin does not count as evidence of that origin (ORIGIN_SOURCE_KIND_MISMATCH). Observations are not automatically assertions. A reading supports nothing until an assertion cites it, and the assertion’s subject and value must match the observation’s subject and payload.

3. Assertion

An assertion is a claim about a subject, backed by observations or by a derivation.
  • temporality is CURRENT (a claim about now, which must be fresh) or PINNED (a claim about a fixed point, such as a block).
  • subject and value, like an observation’s payload, are flat maps of up to 16 entries; keys the claim’s contract does not use are accepted and ignored. value may be null, and an assertion with no observation and no derivation must not carry one.
  • declared (optional) lets an author state the support, freshness, finality and corroboration they believe apply. The engine never trusts it, and reports any excess as a violation: of the specific invariant it breaks, or of INV-00.
  • A derivation applies a WalletWall rule to assertions, never to raw observations. WWEP 0.1 has two rules: wwdr.usd-value and wwvr.vault-technical-readiness.
WWEP 0.1 defines five claim types, and each rests on different evidence: A stablecoin symbol never establishes identity. An on-chain symbol() reading can be spoofed, so it can only corroborate. These policies are served as walletwall://policies/authority.

4. Evaluation: support, authority, freshness and finality

inspect_evidence evaluates every assertion. Evaluation is a pure function of the bundle, the evaluation time, and the bundled registry, policies and rules. The engine never reads a clock, so the same inputs always give the same states.

Support

When states combine across dependencies, the weakest member wins, in this order from strongest to weakest:

Authority

Authority is claim-specific. There is no global ranking such as “chain beats issuer beats provider”. Each claim type’s policy names which origins may establish the claim and which may only corroborate it, and corroboration alone can never reach ESTABLISHED. Authority is not correctness. An authoritative source can be wrong, and a caller can present fabricated evidence that is perfectly well formed. The engine evaluates evidence as presented: it cannot verify that a chain reading was faithfully retrieved, and it attaches TRANSPORT_FIDELITY_NOT_VERIFIED to every chain reading to say so. authority.satisfied: true means the policy’s authority requirement was met by the evidence in the bundle, nothing more. Corroboration counts independence, not volume. Observations in one independence group that report the same value count once (DUPLICATE_EVIDENCE_COLLAPSED). Two provider endpoints in one independence group are one witness. independentCorroboration reports the number of distinct groups.

Freshness

Freshness decides whether evidence can support a claim about the present. Because evaluatedAt is an input, freshness is always relative to a time you choose. The published identity-established example evaluates as ESTABLISHED at its own evaluatedAt of 2026-10-23T12:00:30Z; the same bundle evaluated at 2026-12-01T00:00:00Z, more than 30 days later, is STALE with STALE_EVIDENCE. An observation dated after evaluatedAt makes freshness UNKNOWN_FRESHNESS, with the limitation OBSERVATION_FROM_FUTURE.

Finality

Finality matters for chain-derived observations: a balance can be ESTABLISHED only at a finalized block. The same balance read at an unfinalized block is SUPPORTED, with authority satisfied and the limitation NOT_FINALIZED; with unknown finality it is SUPPORTED with FINALITY_UNKNOWN. When observations disagree about finality, the least final is reported.

Limitations

Evaluations carry limitations: codes from a closed list that explain why a result is weaker than it might look. Some of the most common: The complete list is the limitation enumeration in walletwall://schemas/evidence.

Conflicts and supersession

Conflicting evidence is never silently collapsed. When two chain readings of the same balance disagree, the balance is CONFLICTING with OBSERVATION_DISAGREEMENT: WalletWall does not pick one, average them or rank their sources. A conflict propagates to every claim that depends on it, and it never lets readiness reach CRITERIA_MET: the criteria it touches cannot be decided, so the classification is NOT_ESTABLISHED unless another criterion has definitely failed (CRITERIA_NOT_MET). History is superseded, never rewritten. A supersession record replaces an older assertion with a newer one about the same claim; the older assertion stays in the bundle and is reported SUPERSEDED, and claims that depend on it are treated as STALE (DEPENDENCY_SUPERSEDED).

Evaluation order

For one assertion backed directly by observations:
  1. No observation: NOT_ESTABLISHED (NO_SOURCE).
  2. Only NO_RESULT or ERROR: UNKNOWN.
  3. An invalid or mismatched subject, or invalid values: NOT_ESTABLISHED.
  4. Observations disagree: CONFLICTING.
  5. The asserted value differs from the observed value: NOT_ESTABLISHED (VALUE_MISMATCH).
  6. Registry checks of identity and decimals: NOT_ESTABLISHED on a mismatch.
  7. Authority, corroboration and finality: ESTABLISHED, SUPPORTED or PARTIAL.
  8. Freshness, for CURRENT claims: may lower the result to STALE or PARTIAL.
  9. Lifecycle: supersession overrides to SUPERSEDED, and a conflict overrides to CONFLICTING.
Derived assertions are evaluated through their dependencies, recursively, with cycles detected and reported, and they keep every limitation of their inputs.

5. WalletWall assessment

WalletWall’s assessment is the result of evaluation, in two forms:
  • Per claim: the support, freshness and finality states, authority, limitations and conflicts that inspect_evidence returns for each assertion.
  • Per question: a classification derived by a published rule. In WWEP 0.1 that is VAULT_TECHNICAL_READINESS, returned by assess_vault_readiness as CRITERIA_MET, CRITERIA_NOT_MET or NOT_ESTABLISHED, criterion by criterion.
An assessment is WalletWall’s conclusion under its own published policy. Per-claim states keep the assertion’s own origin; the readiness classification and every derived claim are WalletWall-derived and carry WALLETWALL_DERIVED_NOT_UPSTREAM_AUTHORITY. Processing never increases the authority of its inputs. When required evidence is absent, WalletWall refuses or qualifies. A missing issuer attestation leaves the identity PARTIAL, which leaves readiness criterion C1 INDETERMINATE and the classification NOT_ESTABLISHED (the published example for this case also names the smallest measurement that could change the outcome). The engine never fills a gap, never upgrades an UNKNOWN, and never answers a generic “ready”. The Quickstart shows both outcomes.

6. WallPrint

A WallPrint freezes one assessment into a disclosure record: the evidence references, every claim’s computed evaluation, the assessments, provenance, limitations, conflicts and invariant violations, and the policy, rule and registry versions in force, serialized canonically and identified by a digest. WallPrint adds no evidence state of its own; it records the engine’s. Corrections and supersessions add new records and never edit old ones. See WallPrint.

7. Receipt and public resolver

A receipt shows that a disclosed WallPrint record matches its committed bytes. The public resolver at walletwall.org/wallprint locates a deliberately published record from its portable reference, proves its bytes in your browser, and shows it. Both prove integrity, not truth: a matching digest says the bytes are the committed ones, not that any claim in them is correct.

Distinctions WalletWall enforces

The invariants

check_invariants reports violations of these; several fire only when an author’s declared states claim more than the engine computed. Executable conformance tests in the source repository cover each one.

Canonical form and digests

  • Evidence objects are serialized with the RFC 8785 JSON Canonicalization Scheme and hashed with SHA-256. Ordinary JSON.stringify output is never hashed.
  • Every evidence-critical value is a string. A JSON number in evidence is rejected, so no quantity can lose precision.
  • Quantities are canonical atomic-unit decimal strings no larger than 2²⁵⁶ − 1, with decimals at most 36. Arithmetic uses integers, and decimal renderings are exact and unrounded.
  • An observation has two digests: the evidence digest of the whole observation, and the content digest of its result, subject and payload only. When counting corroboration, the engine collapses readings from one independence group that report the same value (DUPLICATE_EVIDENCE_COLLAPSED).
Use lowercase hexadecimal addresses everywhere in a bundle; normalize_observation lowercases them for you. Matching an observation to its assertion ignores address case, but conflict grouping and supersession compare subjects exactly, so the same address in two letter cases would be treated as two subjects. EIP-55 checksums are not verified.

Versions and known limitations of 0.1

  • A bundle must declare wwepVersion: "0.1"; any other value is rejected, not coerced. A different registryVersion or rulesVersion is evaluated with the engine’s own data and reported with REGISTRY_VERSION_MISMATCH or RULES_VERSION_MISMATCH.
  • Evidence is evaluated as presented: the engine cannot tell a faithful reading from an invented one.
  • The registry is verified at a point in time; later contract changes are not observed.
  • The engine has no live providers: every piece of evidence comes from the caller.
  • Five claim types exist, and the authority policies are WalletWall-authored drafts that have not been reviewed outside WalletWall.

Wallet Evidence Model

The application’s own provenance layer over provider data: a different schema that shares the rule that unknown is never zero.

Independent Authority

Why agreement is strong evidence only when the agreeing things can fail independently — the same principle behind independence groups here.

WallPrint

How an assessment becomes a deliberately published disclosure record.

Tools

The six read-only tools that compute the states defined on this page.