[Pigbibi]

Summary

Document the declared-install gap in runtime monitoring dependencies and propose
a bounded packaging correction. This is a documentation-only change; the ADR
remains proposed.

Why

Crypto's normal entrypoint imports QPK's performance monitor. At the consumed
QPK revision, that path imports Pandas and NumPy while the normal dependency
metadata does not declare them. Existing CI installs both libraries separately,
and one entrypoint check can skip when Pandas is absent. Those checks do not
establish that a normal wheel installation is sufficient.

Changes

• Add a proposed ADR with fixed source references, alternatives, compatibility
  requirements, a clean-install validation plan and rollback.
• Recommend verifying a normal dependency declaration first, without selecting
  version constraints or changing the QPK pin.
• Explain why a QPK monitoring extra needs separate adoption and why delaying
  an import inside a best-effort handler can change failure behavior.
• Add the proposal to both existing README documentation indexes.

Testing

• The existing documentation-checker suite passes all 36 tests.
• The documentation check reports zero blocking findings and the same existing
  README outline advisory as the baseline.
• Fixed source links and line ranges, unchanged-file hashes and the three-file
  diff were checked; git diff --check passes for the candidate diff.

A clean wheel import experiment has not been run. Static source evidence is
not presented as a reproduced production failure. Runtime, package metadata,
dependencies, locks and workflows are unchanged.

Risk and rollback

The proposal requires a later decision between a normal dependency declaration,
a shared-package monitoring extra, or separating monitoring from strategy
imports first. Dependency constraints and consumer adoption need their own
compatibility review. Revert the documentation commit to withdraw the proposal.

Breaking changes

None.

Independent review

A separate reviewer checked the exact three-file candidate, all seven fixed source references, the dependency ownership alternatives and the validation limits. No blocking findings remained. All three checks on the reviewed PR head have completed successfully.

CI results

Checks associated with head 5fb9cffaa7e706d664498307c9417efebcc07f6d are green:

• Test job: 257 tests and 168 subtests passed; dependency consistency, Ruff, whitespace and QPK pin/lock checks passed; source and wheel distributions built.
• Actionlint: passed.
• Documentation check: 36 tests passed; zero blocking findings and the same existing README advisory.

GitHub tested the synthetic PR merge with current main; the intervening main change only adds repository-file guidance in AGENTS.md. The proposed documentation files match the reviewed blobs. The PR remains a draft and is unmerged.

CI installs NumPy and Pandas explicitly. These green results do not establish a complete normal-install dependency closure or approve the proposed dependency correction.