Define reasoning and cached-token translation for cloud harness adapters

Overview

Anthropic translator maps prompt/completion usage but not reasoning_content or cache_read_input_tokens; OpenCode config provides input/output prices only. Custom-provider usage-meter UI was explicitly deferred.

Priority: P2 / post-mvp. Impact: roadmap improvement. Initial planning estimate: Medium; refine after inventory. This issue does not add a release blocker.

Current evidence

Anthropic translator maps prompt/completion usage but not reasoning_content or cache_read_input_tokens; OpenCode config provides input/output prices only. Custom-provider usage-meter UI was explicitly deferred.

• Source: src/anthropic/translate.cpp#L369
• Source: src/anthropic/translate.cpp#L422

Design and implementation plan

Define supported reasoning and cache semantics by upstream/downstream protocols; preserve billing truth separately from optional display.

1. Reconcile the current implementation and any already completed work with the evidence above.
2. Record the owner, selected scope, dependencies and unresolved decisions before implementation.
3. Implement or execute the selected scope using existing components and artifacts.
4. Attach reproducible validation or outcome evidence and record any deferred work.

Acceptance criteria

• Test streaming/nonstreaming reasoning and cache fixtures including absent fields, fragmented deltas and zero counts.
• Do not double count cached tokens or expose hidden reasoning contrary to model protocol.
• Document intentionally unsupported fields and keep backend billing independent.

Related work and deduplication

No existing issue has this acceptance scope; existing release defects remain separate.

• https://github.com/RunanywhereAI/InferenceInfra/issues/492

Requirement provenance

Reconciled September 13, 2026 against WALLY_CLOUD_PRODUCT_SOURCES.md (SHA-256 e601b7b154b750965fa8d25d4533d3542621bd00b3bb30038c1fae553a549e7b) and the consolidated requirements report. Historical proposals are not evidence of current delivery, customer agreement or approved pricing.

Start with src/anthropic/translate.cpp at the referenced lines and inventory the existing streaming and nonstreaming usage translation. Define the supported reasoning and cache semantics with the relevant upstream and downstream protocols before implementation. Done means the listed fixtures and edge cases are validated, cached tokens are not double-counted, hidden reasoning is handled appropriately, and unsupported fields are documented separately from billing.