Legacy cleanup and work-plan review

Reviewed 2026-09-06 against the long-term development plan, now stored at docs/work/long-term-development-plan.md. The plan content is unchanged by its move.

Plan assessment

The plan replaces HY-8 replication with an independent, literature-led solver. Separating geometry, numerical methods, hydraulic relationships, regime selection, and crossing aggregation supports that change. Analytical tests should accompany implementation, with external comparisons later as the plan specifies.

Resolve these points during Phase 0 before implementation:

• Define tailwater as absolute elevation, group-specific depth, or a rating relationship, including treatment of differing outlet inverts.
• Define support or rejection rules for adverse slopes, reverse flow, dry conditions, and inconsistent length/slope/invert inputs. Choose authoritative geometric inputs.
• Define free-surface versus pressurised geometry at the crown: top width, hydraulic depth, and cases where Froude number is undefined.
• Document applicability of identical-barrel scaling and shared-headwater crossings, including excluded approach-flow interactions.
• Specify quantity-specific tolerances, convergence failures and result contracts. Require monotonicity only over physically supported solution branches.

This is a scope/architecture review, not completion of the literature review or engineering acceptance process. The next work is the plan's Phase 0 deliverables.

Removed

• culvertflow/: legacy models, approximate hydraulics, defaults, CLI and HY-8 adapter.
• Old tests and golden outputs: these encode the retired API and implementation, rather than independently validated targets for the new solver.
• Batch comparison scripts and wheel updater: tied to the retired package and bundled runner. Later verification should use external tooling per the plan.
• Installers, packaging wrappers, pyproject.toml, pytest configuration and test wrapper: these built or tested a package that no longer exists.
• Conflicting instructions.txt, outdated README content and temp_view.txt.

Ignored bytecode caches, their directories and .coverage remain: automatic approval review rejected their cleanup commands as blocked by policy. They contain no active source and are not part of the new implementation.

The user's existing dist/ and third_party/ deletions were preserved. No new implementation or placeholder package was introduced ahead of Phase 0.

Discarded prototype source

The previous culvertflow implementation and its obsolete development history were intentionally discarded. They are not compatibility references or reuse candidates. The maintained src/culvert_solver implementation, its primary literature, and its current tests are the only software baseline. Any future method must be derived and validated from those sources rather than reconstructed from the retired prototype.

Retained evidence

File here	Original location / purpose
some-culverts.csv	Root input table; old parser conventions include millimetres for diameter.
sample_cases.csv	tests/data/sample_cases.csv; metre-based legacy API examples with obsolete engine names.
test.hy8	Root historical HY-8 project.
comparison_terms.csv	scripts/diagnostics_terms.csv; saved Python/HY-8 differences.
batch_attempts.csv	scripts/hy8_attempts.csv; batch trace.
root_attempts.csv	Root hy8_attempts.csv; separate trace, retained without merging runs.

These are candidate scenarios and historical evidence, not accepted golden results. Saved comparisons contain substantial flow, velocity and headwater differences; input/default mismatches may also contribute. Independently check inputs before reuse.

The 776 generated files from scripts/hy8_runs/ are retained locally under archive/legacy-hy8-runs/, with contents hash-verified after relocation. This folder remains Git-ignored and needs a separate backup before discarding the checkout. CSV report/workspace paths are unchanged historical provenance; map their hy8_runs/<run>/... suffix to the archive when locating reports.

The local hif12026.pdf remains at the root, Git-ignored. The small CSV traces here are outside the old ignore patterns so they can be reviewed and versioned.

Verification

The earlier pre-cleanup check passed 21 tests and skipped the HY-8 parity module. That established regression behaviour, not engineering validation. No external hydraulic application was rerun for this cleanup.

There is now no executable package or active suite. Introduce new analytical and internal tests alongside the components specified in the plan.