Styxis
Company
Knowledge

Passing a uint8 mask through Dask: separate creation from completed deserialization

Check dtype and fill semantics through a real serialization roundtrip.

Another completion after local creation

Check dtype and fill semantics through a real serialization roundtrip. Local array creation does not establish distributed delivery. Give Truthound actual worker-roundtrip status and mask-reconciliation evidence rather than treating a printable local array as a delivered result. [1, 2, 3]

Serde is not resolved by a cast

Dask #12260 reports uint8 and uint16 masked-array serialization or reconstruction failing with fill_value 999999. A follow-up explicitly says correct dtype casting does not resolve the serde issue. Do not repeat a simple cast or smaller fill as an established remedy, or treat a linked PR as deployment evidence. [1]

Observe the worker roundtrip

Record dtype, mask, fill_value and completed serialization roundtrip. Exclude undelivered arrays from statistical input while checking supported combinations and serde evidence. Run distributed delivery as a separate isolated test, distinguishing creation, serialization and deserialization and retaining pixel and mask semantics. [1, 3]

Preservation evidence required for true

The illustrative uint8 pixel contract requires a completed serialization and reread with dtype, mask and fill meaning reconciled. That verified outcome is true; a deserialization failure is false and fails. Do not reduce the success definition to the absence of one observed exception. [1, 3]

Compare external test status

Truthound Core Python SDK 3.3.2 can compare prepared required_success and serialization_roundtrip_success through ColumnPairValidator with relationship="==". The operator supplies the actual delivery test and success semantics. The SDK does not repair Dask serde or certify mask preservation merely because fill_value was reduced. [1, 2, 3]

Pass on the version and evidence

Confirm a verified serde version or supported representation, or consider sending data and mask separately. Check source pixels and missing positions under either choice. If actual delivery is already verified, hand the next team the relevant version and evidence instead of adding a redundant test. [1, 3]

Application conditions

Run actual distributed-worker delivery as a separate test and distinguish creation from completed deserialization.

Alternatives

Use a verified serde version or supported representation, or assess separate data/mask transfer with semantic checks.

Limitations

  • Core compares external test status, does not repair serde and does not guarantee simple casts or changed fills preserve masks.

Evidence

Event checked
Subject checked
Modified