lnk_persist_init: persist per-source dnstr flag columns in streams_access (mapping_code second-token NONE bug) #196

Context

Post-#187 <persist_schema>.streams_access carries per-species access cols (has_barriers_<sp>_dnstr, access_<sp>) but drops the per-barrier-source flag cols that lnk_pipeline_mapping_code needs to classify the second mapping_code token:

• has_barriers_anthropogenic_dnstr
• has_barriers_pscis_dnstr
• has_barriers_dams_dnstr
• has_barriers_remediations_dnstr
• dam_dnstr_ind (sequence-aware dam-detection; resident flavor)
• remediated_dnstr_ind

These columns exist in working schema's streams_access (lnk_pipeline_access writes them when barrier_sources is passed) but #187 Phase 2 dropped them from cols_streams_access_base on (incorrect) "they're conditional on remediations/observations" reasoning. Actually conditional only on barrier_sources arg, which lnk_pipeline_run always passes.

Impact

lnk_mapping_code reads SELECT * FROM <persist_schema>.streams_access and finds the per-source columns absent → lnk_pipeline_mapping_code's has() helper returns FALSE → any_anth = any_pscis = any_dam = FALSE → second token defaults to NONE.

Empirically PARS BT 2026-05-19:

• link: ACCESS;NONE 4412 rows, ACCESS;NONE;INTERMITTENT 7080 rows
• bcfp: ACCESS;DAM 4411, ACCESS;DAM;INTERMITTENT 5457, ACCESS;MODELLED 1775, ACCESS;MODELLED;INTERMITTENT 3113, ACCESS;ASSESSED 900, ACCESS;ASSESSED;INTERMITTENT 1100

Match_pct vs bcfp drops from historic 98%+ to ~30-50%.

Design choices

Column-naming inconsistency (pre-existing; not introduced by #187):

• has_barriers_<source>_dnstr (boolean prefix has_, suffix _dnstr)
• dam_dnstr_ind / remediated_dnstr_ind (no has_ prefix, suffix _ind for "indicator")

Two patterns coexist. Could rename to one convention (e.g., all has_<thing>_dnstr or all <thing>_dnstr_ind), but that's a separate cleanup. Scope of this fix: add the columns with existing names so lnk_pipeline_mapping_code's column probes find them.

Where to put the persist DDL:

• Option A: Extend cols_streams_access_base directly with all 6 columns. Mirrors the bundle-wide DDL approach from lnk_persist_init: wide tables sized to active_species break on heterogeneous WSG runs #194.
• Option B: New .lnk_cols_streams_access_source_flags() generator (analogous to .lnk_cols_streams_access_per_sp(species) from mapping_code build decoupled from tunnel — persist streams_access + lnk_pipeline_run phase + rename with_mapping_code → mapping_code #187). Keeps per-source separated from base.

Picking B: future per-source extension (new barrier classes) lives in one place, mirrors the per-species helper pattern. Less coupling between "scalar base" and "source-driven".

Defaults: source classes are hardcoded today (anthropogenic/pscis/dams/remediations). #189-style data-driving deferred — would mean a parameters_barrier_sources.csv declaring which sources exist per bundle. Not in this hotfix.

Acceptance

• R/lnk_persist_init.R: new .lnk_cols_streams_access_source_flags() helper returns named vector for has_barriers_<source>_dnstr (boolean, 4 sources) + dam_dnstr_ind (boolean) + remediated_dnstr_ind (boolean) — 6 cols total.
• lnk_persist_init includes those cols in streams_access CREATE TABLE.
• lnk_pipeline_persist's INSERT projection picks them up automatically (iterates names(access_cols_v)).
• Live smoke on PARS post-fix: mapping_code_bt distribution shows ACCESS;DAM / ACCESS;MODELLED / ACCESS;ASSESSED tokens (not just ACCESS;NONE). PARS BT match_pct vs bcfp returns to ~98%.
• v0.40.3 patch release.

References

• R/lnk_pipeline_mapping_code.R:174-194 — column probes that consume the persisted flags.
• R/lnk_pipeline_mapping_code.R:196-210 — hardcoded precedence chain (separate issue: rules engine).
• R/lnk_persist_init.R cols_streams_access_base — currently missing the per-source columns.
• R/lnk_pipeline_access.R lines ~250-260 — where the per-source flags get emitted.
• Same bug class as lnk_persist_init: wide tables sized to active_species break on heterogeneous WSG runs #194 (caught by live data, not unit tests). Add unit tests for v0.40.0 surface (lnk_mapping_code, persist additions, pipeline_run mapping_code phase, deprecation shims) #191 tracks unit-test catch-up.