de:wasserversorger-enrich

Consistent with convention — batch input files are kept, not deleted. Nothing further to clean up. This run is complete.

Summary

Ran the wasserversorger-enrich scheduled batch: built the next 45-company batch (0 pending_domain, all priority-ordered by population), dispatched it as two shards to sequential foreground research subagents, verified both outputs (45/45 valid, zero collisions/overlaps), and merged into the master pipeline (enrichment.jsonl 2024→2069 lines, roster-links.jsonl +89, population coverage steady at 99.99%).

All 45 companies confirmed as genuine water suppliers with strong field recovery (100% on name/address/phone/website, ~87% email, 67% analysis-URL hit rate). Flagged 8 items for human review (a likely wrong-entity roster row for Schwarzenbruck, a newly-found 4th member area for Lechraingruppe, a few HR/attribution edge cases) without unilaterally changing the roster, and logged 6 new host-fetching quirks. Journal entry appended with full detail; next run should continue with build_enrich_batch_excl.mjs 45 (1331 pending remain).

build_enrich_batch_excl.mjs 45 → 0 pending_domain again (roster 3375, 2019 enriched entering, 1376 pending, 6 standing duplicate-mint pairs excluded). Split shard A (23) / shard B (22), each dispatched to one top-level Agent call run in the foreground sequentially (not launched in the same message this time — no functional difference, both still ran to completion inline, just not wall-clock-parallel). Both finished unattended (215 + 205 tool calls) writing their own enrichment.part-shard{A,B}-enrich-2026-09-05-1530.jsonl and roster-links.part-shard{A,B}-enrich-2026-09-05-1530.jsonl shards.

Verification before merge: both shard files valid JSON (23+22 lines, 0 bad, 23+22 distinct company_ids each, exact match against the two batch files, 0 collisions with existing 2019 records, 0 overlap between shards). enrichment.jsonl 2024 → 2069 lines (2019 → 2064 distinct company_ids, +45). map_roster.mjs: 20994 → 21083 links (+89, matching the shards' 62+27 exactly). consolidate_roster.mjs: still 177 shards / 3375 distinct companies (unchanged, expected); roster-links.csv 17792 distinct after dedup. coverage.mjs: population 99.99% → 99.99% (unchanged, expected), same 24 residual municipalities / 11,023 people, Hartmannsdorf-Reichenau still largest at 1,040.

Non-suppliers found: 0 — all 45 confirmed genuine drinking-water operators.

Field recovery (45 companies, both shards combined): name/address/phone/website_url 45/45; email 39/45 found (4 not_found, 2 not_published); short_name only 12/45 published (most sites use full names only); contact_form_url 14/45 found; emergency_phone_number 32/45 found (8 not_published, 5 not_found); handelsregister_nr mostly not_applicable (36/45), 9 genuine finds incl. two flagged below; opening-hours fields the usual weak spot (~31/45 address, ~31/45 phone). Analysis locator: 14/23 (shard A) + 16/22 (shard B) = 30/45 got a real analysis_url ≈ 67%, back in line with the campaign's historical ~70% real-supplier rate after the prior batch's dip; multi_zone true for 7/23 + unresolved-heavy split in B — roughly consistent once unresolved rows are excluded.

Roster corrections/questions flagged for human review, not acted on:

• wv-de-18520 Gemeindewerke Schwarzenbruck GmbH: likely wrong entity — the GmbH (HRB 22124) covers only Strom/Gas, no Wasser content anywhere; actual water supply looks like it's run by the Gemeinde itself (schwarzenbruck.de), a distinct legal entity. Needs re-pointing.
• wv-de-7608 Wasserverband Lechraingruppe: a 4th member area "Hardhofgruppe" (own domain wasserzweckverband-hardhofgruppe.de) found beyond the roster's known Aindling/Petersdorf/ Todtenweis three — 18 new Ortsteil links added; worth checking if Hardhofgruppe should be its own roster company rather than folded in here.
• wv-de-18574 Bad Endbach: Impressum explicitly prints HRA 4417 (Amtsgericht Marburg) for the Gemeindeverwaltung itself — an unusual but direct, unambiguous exception to the municipal not_applicable default (same shape as Pullach/Malchow/Kahl in prior batches).
• wv-de-11503 Gemeinde Fuchstal: neighbouring Vilgertshofen's site says its water comes from a spring in Lechmühlen, a Fuchstal Ortsteil — sits oddly with Fuchstal's self_only classification. Flagged, not reclassified.
• wv-de-1115 Markt Schöllkrippen: caught a WebFetch-summary error before it became a roster correction — a 6-municipality list on the water/sewage page actually belongs to the sewage Zweckverband section, confirmed via raw HTML; no extra municipalities added.
• wv-de-7558 Schwarzenfeld: resolved a previously-unresolved boundary question — the water page explicitly splits Ortsteile between Wasserwerk Schwarzenfeld, Pretzabrucker Gruppe (wv-de-7557) and Nord-Ost-Gruppe (wv-de-7559).
• wv-de-18569 Neunkirchen a. Brand: Rosenbach (an Ortsteil) has its own water-emergency number under the Erlangen area code, suggesting a possibly different actual operator. Flagged only.
• wv-de-18533 Teublitz: avoided a false correction — a PDF WebFetch's summary mislabeled as "Vils-Naab-Gruppe" is actually for the Münchshofen Ortsteil per its raw filename.

New host traps this batch: biebergemuend.de, dornstetten.de and lossburg.de (all Nuxt/ SSR sites) 403 WebFetch's default fetch but 200 to curl with a desktop-Chrome User-Agent — liederbach.eu/frammersbach.de pattern, now a fourth+fifth+sixth instance; hausham.de's Notdienste page 403s outright and a roster-carried Typo3 deep-link there no longer resolves; schieder-schwalenberg.de serves ISO-8859-1/near-binary HTML — plain grep silently returns 0 matches on hrefs until -a is added, a new memory-safety-adjacent gotcha (looks like "no links" rather than an encoding problem); gewgmbh.de obfuscates emails via a non-standard hrencrypted JS scheme (not the reversible Cloudflare cfemail hex) — correctly left not_found, not guessed.

For next run: continue with build_enrich_batch_excl.mjs 45. Still 0 pending_domain, 1331 other pending (3375 roster total, 2064 distinct company_ids enriched, 6 excluded). New standing items from this pass: Schwarzenbruck GmbH wrong-entity question, Lechraingruppe/Hardhofgruppe 4th-member question, Bad Endbach HR exception, Fuchstal/Vilgertshofen source question, Neunkirchen a. Brand/Rosenbach operator question. All prior standing items carried in the 2026-09-05 10:30 entry (Pullach HR exception, Kahl bulk-sweep-cohort question, Reckenberg/Riegelstein member-list conflict, Memmelsdorf/Großenlüder undercounts, Neuried self-supply question, Allersberg/Brunnbach shared-analysis question, Amt Kisdorf retail-pass-through question, Garching/Schierling referenced-supplier check, Blankenheim address discrepancy, and the long tail of older items) remain open, none touched this pass.