Behaviour common to every tool
- Read-only by declaration and by construction. Every tool is annotated
readOnlyHint: true,destructiveHint: false,idempotentHint: trueandopenWorldHint: false. The server has no network, filesystem or key-handling code to misuse. - Closed inputs, with three open maps. Tool arguments and every bundle record (sources,
observations, assertions, derivations, conflicts, supersessions) are closed: an unknown field is
rejected, not ignored. A claim’s
subjectandvalueand an observation’spayloadare open maps of up to 16 entries; keys the claim’s contract does not use are accepted and ignored by the evaluation. Map values are short strings, booleans ornull, and a JSON number anywhere in evidence is rejected. Sizes are bounded: up to 64 sources, 128 observations, 128 assertions, 64 derivations, 32 conflicts and 64 supersessions per bundle, at most 32 IDs in a list, and output-safe text (letters, digits, spaces and_ . : / + -, at most 128 characters, or 64 forretrievedViaand version fields). - Explicit time. The engine never reads a clock. Tools that judge freshness take an
evaluatedAtUTC timestamp such as2026-10-23T12:00:30Z, and the same input with the sameevaluatedAtalways produces the same result. - Two renderings of each result. A successful call returns the result as
structuredContentand as the same JSON in a text block. - Weak results are results. Missing, stale, conflicting or unsupported evidence produces a weak state with reasons, never an error. Errors are reserved for input the tools cannot accept.
Errors
An error response hasisError: true and one text block holding a fixed-shape object:
Errors never echo what you sent.
issues names only fields from WalletWall’s own vocabulary; any other
path segment is shown as *, and array positions as #.
resolve_asset
Purpose. Look up a stablecoin deployment in WalletWall’s bundled registry: five deployments (USDC on
Ethereum, Base and Arbitrum One; USDT and DAI on Ethereum), verified once against issuer or protocol
publications and finalized chain state on 2026-10-24.
When to use it. To find the canonical contract for a chain, to check whether an address is a
registered deployment, or to see which deployments share a symbol before you gather evidence.
Inputs. Provide chainId together with address, or provide symbol.
Output.
Refusals. An
address without a chainId is INVALID_INPUT.
In the evidence model. A registry lookup never establishes identity, because no issuer attestation
was presented with the query. A symbol is never identity: a symbol lookup returns candidates, and an
unknown symbol comes back NO_SYMBOL_MATCH with the note “A symbol is not an identity. Identify assets
by chain ID and contract address.” Absence from the registry is not evidence of fraud; it means
WalletWall has no identity for that contract.
normalize_observation
Purpose. Normalize a balance-at-block reading into a canonical WWEP observation of claim type
TOKEN_BALANCE_AT_BLOCK, ready to place in a bundle.
When to use it. Whenever you bring your own chain reading into WalletWall, so that addresses,
quantities and digests are canonical before evaluation.
Inputs.
Output.
observation (addresses and hashes lowercased), canonicalJson (its RFC 8785 form),
evidenceDigest (SHA-256 of the whole observation), contentDigest (SHA-256 of result, subject and
payload only, so two readings of the same thing share it whatever their IDs, sources and times) and
balanceDecimal (exact, unrounded, or null when there is no value).
Refusals. A VALUE observation without balanceAtomic and decimals, an out-of-range quantity or
a non-canonical number (a leading zero, a JSON number) is refused with INVALID_QUANTITY or
INVALID_INPUT.
In the evidence model. The tool does not fetch chain data; the reading is yours. Normalizing it
does not make it more authoritative: a NO_RESULT reading stays without a value (payload: {},
balanceDecimal: null) and can never become zero.
inspect_evidence
Purpose. Evaluate every assertion in a WWEP 0.1 bundle against the bundled authority policies at
the evaluation time you supply.
When to use it. This is the core tool: use it whenever you need to know how well a bundle supports
its claims.
Inputs. bundle (a WWEP 0.1 bundle, see walletwall://schemas/evidence) and evaluatedAt.
Output. The wwepVersion, evaluatedAt, a bundleDigest, the versions in force (WWEP,
registry, policies, rules), an invariantViolationCount, and one evaluation per assertion:
For example, the synthetic
balance-established example evaluates as:
INVALID_INPUT. A bundle
that passes the schema but whose subject or value does not match the claim’s contract is evaluated:
the assertion is NOT_ESTABLISHED with INVALID_SUBJECT, SUBJECT_MISMATCH, INVALID_VALUE or
VALUE_MISMATCH, and diagnostics says which keys are wrong without echoing any value:
check_invariants
Purpose. Check a bundle against the thirteen mandatory WWEP evidence invariants (INV-01 to
INV-13), plus INV-00, the engine-rule check: a declared state that exceeds the computed one where
no more specific invariant applies, a dangling reference, a duplicate source, observation or derivation
ID, and a derived value that does not recompute.
When to use it. Before you rely on a bundle someone else assembled, and whenever a bundle carries
declared states.
Inputs. bundle and evaluatedAt.
Output. pass, violationCount, violations and the bundleDigest. Each violation names the
invariant, a code, the assertion and the rule text:
ESTABLISHED on an identity the engine computes as PARTIAL. Other codes
include NO_SOURCE_PRESENTED_AS_FACT and NO_RESULT_AS_VALUE (both INV-01): an assertion that
carries a value with no source behind it, or a value its only observation never returned. A zero or
false value in that position is reported as INV-04 (NO_RESULT_AS_ZERO) or INV-05
(UNKNOWN_AS_FALSE) instead.
In the evidence model. Violations are reported, never repaired: the engine does not mutate the
evaluation to make a bundle pass. inspect_evidence reports the same count as
invariantViolationCount.
trace_provenance
Purpose. Return the lineage of one assertion and all of its ancestors, as a graph shaped like W3C
PROV-JSON.
When to use it. To explain why WalletWall says something: which observations and sources an
assertion rests on, and through which derivations.
Inputs. bundle and assertionId.
Output. target, a flat lineage (each node’s id, kind and depth), a prov graph and
issues. In the graph, observations and assertions are entities, derivations are activities, and
sources and the WalletWall rule engine are agents; relations are wasDerivedFrom, wasAttributedTo,
wasGeneratedBy, used and wasRevisionOf (supersession). Identifiers use the prefix
ww = urn:walletwall:wwep:0.1:, a name that is never dereferenced.
assertionId that is not in the bundle is not an error: the result has an empty
lineage and the issue TARGET_NOT_FOUND.
In the evidence model. Transport is recorded as the source’s ww:retrievedVia attribute and is
never treated as authority.
assess_vault_readiness
Purpose. Evaluate WalletWall’s published VAULT_TECHNICAL_READINESS rules
(wwvr.vault-technical-readiness, 0.1.0-draft) over one asset-identity assertion and one
balance-at-block assertion.
When to use it. When the question is whether a specific balance of a specific stablecoin meets the
technical evidence criteria for a vault workflow.
Inputs. bundle, evaluatedAt, identityAssertionId (an ASSET_IDENTITY assertion) and
balanceAssertionId (a TOKEN_BALANCE_AT_BLOCK assertion).
Criteria. Each is MET, NOT_MET or INDETERMINATE, with a reason code:
Classification. Any
NOT_MET criterion gives CRITERIA_NOT_MET. Otherwise any INDETERMINATE
criterion gives NOT_ESTABLISHED. Only when all six are met is the result CRITERIA_MET. There is no
generic “ready” value.
Output. classification, criteria, the inputs with their support states, supportOfInputs
(the weakest), limitations, doesNotEstablish, a suggestedDerivation and suggestedAssertion (the
VAULT_TECHNICAL_READINESS records a bundle could carry for this result) and a resultDigest.
Refusals. Wrong or swapped assertion IDs are not an error: every criterion is INDETERMINATE
with REQUIRED_INPUT_ASSERTION_MISSING_OR_WRONG_CLAIM_TYPE, and the classification is
NOT_ESTABLISHED.
In the evidence model. The classification is WalletWall-derived (its suggested assertion has origin
WALLETWALL_DERIVED) and always carries WALLETWALL_DERIVED_NOT_UPSTREAM_AUTHORITY. It does not establish investment merit, solvency,
creditworthiness, legal status, regulatory compliance, absence of fraud or future performance, and
every result lists those non-implications in doesNotEstablish. It also configures, authorizes and
executes nothing: the vault research prototype’s own execution policy is described in
Policy-Gated Execution.
Where to go next
Resources
The eleven static resources, including the schemas these tools validate against.
Evidence Model (WWEP)
Every state and limitation the tools can return, and the order in which they are decided.
Quickstart
Run the published synthetic examples through the real server.
Troubleshooting
Rejected input, weak results and adapter problems, by symptom.

