Reproduce the asset/cash linkage in a Pontes reconciliation extract
Align transaction, asset and cash-leg identifiers and reference times before interpreting tokenised-settlement extracts.
Different states are not yet a failure
An asset extract marked complete and a cash extract marked pending do not immediately establish failed settlement. Their extraction times or state definitions may differ. Truthound can check aligned reconciliation inputs and expectations; it cannot declare settlement finality.
The September Pontes launch
The Eurosystem launched Pontes on September 21, 2026 to connect wholesale tokenised-asset transactions with central-bank-money settlement. Its initial core services are to expand gradually. Planned features and operating hours should not be treated as already available.
Use a common-time fixture
Begin with a fixture separate from real transactions, or an authorised ledger extract. Preserve reference time, extraction time, transaction, asset and settlement-instruction identifiers. Establish that records describe the same relationship before comparing amounts or final states.
Connect the asset and cash legs
Map the instruction connecting asset and cash legs. The actual business definition determines whether multiple instructions or partial settlement are possible. Do not force every relationship into 1:1, or combine the latest states from incompatible reference times.
The expected result in Truthound
Truthound Core Python SDK 3.3.4, operating on a prepared Polars table, provides QueryResultMatchesValidator to compare a supplied SQL result with expected_df or a matcher. ForeignKeyValidator checks prepared child/parent keys. Owners define expected states and quantities. Comparing with an expectation is not replaying the real settlement.
Where the pandas join remains
An existing pandas 3.1.1 merge(validate=...) can remain at the cardinality boundary. It answers a different question from post-join amount or state reconciliation. Separate linkage from expected-result assessment so the origin of a mismatch is visible. Separate unidentified legs so a pandas null-reference match cannot count as a valid asset/cash linkage.
Separate lag from a substantive mismatch
Distinguish a late-updating file from incompatible records at the same reference time. Their next actions differ. Hold data outside the agreed observation window rather than silently rewriting an earlier reconciliation with today's value.
Hand the result to ledger owners
Return exceptions to ledger and settlement-operation owners. Guessing a missing key or changing a state is another controlled procedure. A passing input check should not become approval of legal effect or security.
Preserve the next reproducible reconciliation
Preserve input versions, reference time, mapping and expected result for the next comparison. A reviewer should be able to identify which assumption changed when records disagree. Keep data reconciliation separate from actual settlement judgment throughout.
Conditions
Use authorised extracts with defined keys, states and reference times, distinguishing fixtures from actual settlement.
Alternatives
Keep pandas 3.1.1 join-cardinality checks at the linkage stage, separately from expected-result reconciliation.
Limitations
A reconciled table does not establish settlement finality, smart-contract security or actual settlement success.
Evidence
- Event checked
- Subject checked
- Modified
