#233: config data dictionaries + fresh<->link column ownership boundary (v0.44.3)

[NewGraphEnvironment]

Summary

• Renames configs/dimensions_columns.csv → dictionary_dimensions.csv and adds the missing sibling dictionary_parameters_fresh.csv (19 rows: type, group, owner, consumed_by, default, description). The nine cluster_* columns had no documentation anywhere outside fresh's source.
• Ends the re-derivation of the fresh↔link column-ownership boundary. It was settled by Replace observations parameter with barrier_overrides from link fresh#129 (fresh 0.12.7 removed observation_*: "fish passage interpretation belongs in link, not the network engine") but was findable only by archaeology through two repos' planning archives. Now encoded as the owner column — 14 fresh-owned engine params, 5 link-owned observation_* — and written up in RUNBOOK.md §7.
• Makes the dictionary load-bearing. audit_configs.R §3b reads owner instead of hardcoding grepl("^observation_", ...), so adding a link-owned column is a dictionary edit rather than a regex edit.
• Removes audit_configs.R's hardcoded setwd("/Users/airvine/...") — it now derives its repo root from its own script location and resolves paths via repo_path().

Two findings from verifying rather than assuming

Every consumed_by is a machine-verified file:line (24/24 resolve). That surfaced two things the issue body had wrong or missing:

• link never reads the cluster_* columns. It only passes the frame through at R/lnk_pipeline_connect.R:107 to fresh's .frs_run_connectivity(); the real readers are fresh/R/frs_habitat.R:1164-1230. The issue body's "→ lnk_pipeline_connect.R" was imprecise and is corrected in the dictionary.
• rear_gradient_min is read by no code in either package. Header-only in fresh's canonical CSV and every link bundle. Recorded as unused rather than dropped — fresh owns that schema, so removal is a fresh-side call.

Gotcha worth knowing

The bundles carry different column subsets — bcfishpass dimensions.csv has 30 columns to the three default* bundles' 32. Any coverage check must assert against the union across bundles; a per-bundle equality assertion would be wrong.

Related Issues

• Closes Rename dimensions_columns.csv → dictionary_dimensions.csv; add dictionary_parameters_fresh.csv #233
• Unblocks dimensions_columns.csv as source-of-truth: auto-gen README + lnk_rules_build validation #75 (auto-generated README + lnk_rules_build() validation) by giving both files a consistent shape
• SRED: NewGraphEnvironment/sred#24

Test plan

• devtools::test() — 1317 PASS (+23 new), 1 pre-existing failure
• Rscript data-raw/audit_configs.R — 0 findings, exit 0
• Audit runs clean from /tmp (outside the repo), and source()ing it leaves cwd untouched — the properties the hardcoded setwd was hiding
• Coverage guard negative-tested: dropping one dictionary row makes the audit exit 1 and fire both flags; file restored byte-identical
• lintr — no new lints (test-dictionaries.R 0; audit_configs.R 13, unchanged from baseline)
• Old-name sweep clean (only the historical NEWS.md entry and the archived findings retain it, both intentional)

Notes

One test fails and it is not from this branch. test-lnk_wsg_resolve.R:138 needs DB table public.wsg_outlet, whose builder is open follow-up #227. It is red on main too — a CSV rename cannot create a Postgres table.

Guarding needed two layers because data-raw/ is .Rbuildignored and the audit never runs for an installed package: tests/testthat/test-dictionaries.R is the CI-side guard; the audit is the pre-trifecta gate.

Generated with Claude Code