# PR-26 Walker Cache Input and Comparison Policy

Domain Owner: `docs/packer`

## Briefing

Once the durable cache model exists, the next narrow step is to let walkers consume prior cached file facts without coupling them to cache storage.

This PR defines the cache-aware walker contract and the ordered invalidation policy that decides when prior metadata can be reused and when a fresh probe is mandatory.

## Objective

Deliver walker-side prior-cache input and a deterministic comparison policy ordered by `size`, then newer `lastModified`, then fingerprint/hash.

## Dependencies

- [`./PR-24-asset-file-cache-hydration-and-walker-reuse.md`](./PR-24-asset-file-cache-hydration-and-walker-reuse.md)
- [`./PR-25-asset-cache-model-and-repository.md`](./PR-25-asset-cache-model-and-repository.md)

## Scope

- extend walker inputs so they receive prior asset cache state
- keep walkers independent from cache repository I/O
- apply walker-side internal file analysis only to already-registered assets
- define ordered invalidation checks:
  - changed `size` invalidates immediately
  - newer `lastModified` invalidates immediately
  - fingerprint/hash is evaluated last when stronger confirmation is still needed
- allow walkers to reuse prior metadata when cached facts remain valid
- preserve file-scoped diagnostics in walk output

## Non-Goals

- no runtime snapshot attachment yet
- no cache persistence integration in the loader yet
- no point write-path refresh yet
- no Studio UI contract changes

## Execution Method

1. Update walker contracts to accept prior asset cache input together with asset root and declaration.
2. Keep cache lookup outside concrete walkers; they should receive an already-resolved asset cache view.
3. Implement the ordered invalidation policy in the shared walker path.
4. Reuse prior metadata only when the ordered checks keep the cached entry valid.
5. Preserve file-scoped diagnostics in the resulting walk output for later runtime and UI consumption.

## Acceptance Criteria

- walkers receive prior asset cache entries as input
- the walker layer does not perform cache repository I/O directly
- unregistered assets are excluded from internal file analysis and cache-aware walker reuse
- changed `size` invalidates prior reuse immediately
- newer `lastModified` invalidates prior reuse immediately
- fingerprint/hash remains the final expensive comparison step
- stable files may reuse prior metadata without changing query-visible semantics
- file-scoped diagnostics remain segregated in walk output

## Validation

- walker tests proving changed `size` invalidates reuse
- walker tests proving newer `lastModified` invalidates reuse
- walker tests proving fingerprint/hash is only evaluated after cheaper checks do not already invalidate reuse
- walker tests proving stable files can reuse prior metadata
- tests proving file-scoped diagnostics remain attached to per-file walk output

## Affected Artifacts

- `docs/packer/pull-requests/**`
- `prometeu-packer/prometeu-packer-v1/src/main/java/p/packer/repositories/**`
- `prometeu-packer/prometeu-packer-v1/src/main/java/p/packer/models/**`
- `prometeu-packer/prometeu-packer-v1/src/test/java/p/packer/repositories/**`