Choose capabilities from current metadata

These examples and illustrative results are independently authored teaching materials, not measured model results.

Use case

Choose speech output from an accessible potentially paged catalog. Teaching M1 lacks TTS, M2 supports it and voice-v1 is listed. IDs/fields are synthetic rather than current product names.

Mechanism

Inspect current capabilities/schema/paging, filter required features/formats/parameters and select returned identifiers/voice. Handle removal/access changes through rediscovery rather than invented substitutes. Catalog and actual output compatibility require separate evidence.

Bad example

Use remembered IDs, assume TTS on every model, read one page and guess missing voices.

Good example

Read the teaching complete accessible catalog, verify M2 TTS/format/parameters and listed voice-v1, rejecting M1. Use returned IDs. Missing pages/metadata remain incomplete; removal/access errors trigger inspection/rediscovery, with no suitable option an explicit gap. Do not invent alternatives.

Why the change matters

Current metadata binds identity to capability, separating past existence, access and speech support. Coverage/change handling prevents first-page absence becoming universal absence.

Observable expectation

The fixture chooses evidenced M2/voice-v1; removed voice does not remain valid, and incomplete catalogs retain scope. Actual output probes need authorized execution.

No speech is generated or frozen model/limit promoted to current fact.

Limits

Metadata expires and establishes neither quality/cost nor authorization. Provider shapes differ; parse actual schemas rather than assumed nested structure.