Early recovery phase that type-checks the PRIOR CORE STATE before any later phase trusts it.

On a cold boot the record is genuinely durable — written by a previous epoch, possibly by a previous version of this software, and read back off object storage. (On a warm relaunch it is projected in memory from this coordinator's own last layout and never round-trips storage, so the check is cheap there and meaningful here.) Everything after this point locks and copies from the logs it names, so a type mismatch (integer tag ranges arriving as Version.t() binaries, say) would otherwise surface as an MVCC lookup failure far from its cause. Validating at the boundary makes the durable record the thing that fails, with diagnostics naming it.

What it checks is the logs field: each entry's tag ranges must be integers, not binaries. The validator also has a resolvers check, which is vacuous against this record — the prior core state carries no resolvers, and the previous epoch's resolver pids would be worthless if it did.

Error Handling

On validation failure, this phase stalls recovery with {:corrupted_tsl, details} to allow operators to investigate and fix the underlying data corruption rather than failing silently or propagating errors further into the recovery process.

Integration Point

Should run early in the recovery pipeline after TSL data is loaded but before any processing that depends on type-correct TSL fields. This provides a clear failure point with detailed diagnostics.

Transitions to the next appropriate recovery phase on successful validation.