[santoshkumarradha]

An interrupted run with no saved priced call showed no model information even when its checkpoint retained the assigned crew. Task links and arrow navigation could also open an ordinary task room for a run, hiding its setup and offering a model picker that could not apply.

Task setup now shows saved worker, planner, and checker assignments separately from historical model usage. Program pages such as senior-dev list their declared models and thinking level. Missing history is explicit, and mapped run links open the same stored page as sidebar clicks. Run and program models remain fixed for an existing run; /crew sets the next run's crew. Ordinary task model controls retain their existing scope. Live per-role switching is outside this change.

Validation: make pr-ready passed. Its classifier reported TestANewDirectionArrivesWithTheClaimUnspent as flaky after it passed its automatic retry. Coverage includes interrupted runs without priced calls, archived assignments after reopening, remote page transport, task-link navigation, and ordinary task controls. Actual TUI sessions verified a completed ordinary run and a completed senior-dev run with OpenRouter and deepseek/deepseek-v4.1-flash; the accepted runs' recorded model calls were audited. An isolated interrupted-task replay also shows all three assignments with an empty spend ledger.

Visual evidence is prepared: the completed ordinary run, a cropped senior-dev setup panel, and a cropped interrupted-run setup panel. Crops preserve the actual TUI rendering and exclude private paths and unrelated content. Direct image attachment is pending because the connected GitHub tools have no attachment uploader and the available browser is signed out. No evidence files are committed to this repository.

All CI checks pass. The visual evidence has been reviewed; direct image attachment remains unavailable through the current tooling.