External Assurance Runner commissioning — IBANforge
Observed: 6 September 2026
Service: POST https://api.ibanforge.com/v1/iban/validate
You can pay to be tested. You cannot pay to be trusted.
What this run was for
This was the first completed end-to-end external commissioning run of the current Atinamos Assurance Runner → Verification production publication path against a paid third-party machine service.
The run was human-triggered commissioning using Runner-controlled bounded execution logic. It should not be described as unattended production autonomy.
The path tested was:
controlled test definition
→ bounded x402 purchase
→ independent Base settlement observation
→ fulfilment capture
→ independent deterministic correctness comparison
→ immutable Verification package
→ signed Assurance Evidence Receipt publication
→ public buyer lookup
Tested capability
The independent claim was deliberately narrow:
IBAN checksum validation
Independent method:
ISO 13616 MOD-97
Only this seller response field was independently evaluated:
valid
BIC, SEPA, bank, issuer and other enrichment fields may have been present in the seller response, but they were not independently verified by this run.
Independent controls
Positive control:
DE89370400440532013000
expected valid = true
observed valid = true
Negative control:
DE90370400440532013000
expected valid = false
observed valid = false
The negative fixture passing with valid=false is a correct negative-control result, not a seller failure.
Paid execution
Two bounded x402 calls were made:
positive control: 0.005 USDC
negative control: 0.005 USDC
total observed test spend: 0.010 USDC
network: Base mainnet / chain 8453
Independent Base settlement references retained by the Runner:
Positive control:
transaction: 0x60d6747a566f8e820d44fa075126dd415990334bef2b36c944840fb6b7eacd45
block: 50962836
amount: 0.005 USDC
Negative control:
transaction: 0x73261a10bdfc03db1203897d4f86a158698b14280a44894b749c2f8fee5a17f9
block: 50963382
amount: 0.005 USDC
For both paid controls:
- settlement was independently observed on Base;
- useful fulfilment was observed;
- the returned
validvalue matched the independently precomputed expectation; - the validation result was terminal and publishable.
The transaction hashes are public chain references included to make the settlement claim independently inspectable. Settlement alone does not prove fulfilment or correctness.
Signed production Evidence Receipts
Two signed Assurance Evidence Receipts were published under:
atinamos-assurance-ed25519-202609-002
Positive control:
Negative control:
Publication recovery was also tested by reality
The first publication request reached the Verification receiver while its write gate was disabled and returned HTTP 503 before write confirmation.
The Runner treated the dispatch outcome as indeterminate rather than blindly retrying it.
A GET-only reconciliation found no published record. After the receiver publication/signing state and idempotency behaviour were intentionally inspected, one explicit operator-approved exact replay was performed. Publication returned HTTP 200 and reconciliation confirmed the record.
That is useful production evidence for the fail-closed publication/recovery path. It does not justify automatic replay after uncertain side effects.
Buyer-facing result
The resulting evidence can be inspected through the Assurance/Verify read surfaces.
For the exact service, the completed commissioning supports:
observations: 2
paid tests: 2
successful fulfilments: 2
failed fulfilments: 0
signed Assurance receipts: 2
independently evaluated controls passed: 2 / 2
independent method: ISO 13616 MOD-97
verified output field: valid
total observed test spend: 0.010 USDC
Exact machine evidence should be queried using both endpoint and HTTP method:
GET /v1/assurance/evidence?endpoint=https://api.ibanforge.com/v1/iban/validate&method=POST
What this supports
This commissioning record supports the narrow claims that:
- the current bounded Runner path completed two real paid calls against the stated external route;
- Base settlement was independently observed for both controls;
- useful fulfilment was observed for both controls;
- the returned
validfield matched independent MOD-97 expectations for one valid and one invalid fixture; - the current production publication path produced two signed Evidence Receipts;
- the resulting evidence is available to buyer-facing read surfaces.
What it does not support
This record does not establish:
- permanent or provider-wide reliability;
- universal IBAN correctness across all countries or inputs;
- correctness of BIC, SEPA, bank or other enrichment fields;
- the provider's internal implementation mechanism;
- that verification traffic was indistinguishable from ordinary buyer traffic;
- unattended production autonomy;
- a universal recommendation to buy the service;
- a trust score or certification.
Why this is a milestone
Earlier Atinamos work proved individual layers: discovery, x402 payment, settlement observation, output assertions, autonomous bounded buying and signed receipt generation.
This commissioning joined the current production Assurance Runner and Verification publication path end to end against an externally controlled paid service.
The Runner was subsequently reconciled read-only on 7 September 2026. Runner wallet access, payment signing, live paid execution and its outbound Verification-publication gate were confirmed back off after commissioning, while Base settlement observation and the fixed Verification destination remained configured. That Runner health audit did not independently establish the separate Verification receiver's current write/signing gate state.
The next operational proof is not another rebuild. It is one genuinely unattended, tightly bounded Assurance Run in which the Runner carries the routine challenge, signing, payment, settlement, validation and publication stages without operator stepping.
Technical record
The deliberately sanitised technical experiment record is published in the Atinamos public research repository: