de:wasserversorger-enrich

Everything checks out. The run is complete.

Summary: Enriched a 45-company batch of German water suppliers (population-priority order, 0 pending_domain in queue). Two parallel research agents each completed their shard fully (23 + 22 companies, ~45 min unattended, 889 tool calls total), with strong field coverage (contact fields 91%+, analysis URLs 45/45, 0 suspected non-suppliers) and several new host/CMS traps documented. Merged cleanly into enrichment.jsonl (1754→1799) and roster-links.jsonl (19662→20061, +399 links); population coverage holds steady at 99.99% (24 residual municipalities, ~11k people), as expected since this batch enriched already-claimed companies rather than adding new roster rows. Journal entry appended with full details, host traps, and scope-correction notes for the next run.

build_enrich_batch_excl.mjs 45 → 0 pending_domain (queue empty since 2026-09-03 20:30), batch drawn purely by population priority. Split into shard A (23) / shard B (22), each dispatched to one top-level Agent call, explicitly told not to spawn nested sub-agents and to append after each company via append-from-file.mjs. Both ran to completion unattended (~45 min each, 472+417 tool calls) with no OOM or nesting issues.

Verification before merge: both shard files validated as clean JSON (23 + 22 lines, 0 bad), all 45 batch company_ids present, zero collisions, zero extras. enrichment.jsonl 1754 → 1799 (+45). roster-links.jsonl 19662 → 20061 (+399, avg ~8.9 links/company — high this batch, several Zweckverbände with long Mitgliedsgemeinden lists plus one bulk/Fernwasser supplier).

Merge: map_roster.mjs: 20061 links total, preresolved 11869 / exact_norm 4611 / land_scoped 449 / ambiguous 100 / unmatched 2974 (unmatched count is cumulative across the whole file, not just this batch — in line with the standing baseline, nothing new flagged as alarming). consolidate_roster.mjs: still 177 shards / 3375 distinct companies (unchanged, expected — batch enriches already-claimed companies); roster-links.csv 16598 → 16964 distinct after dedup (+366 net). coverage.mjs: population 99.99% → 99.99% (unchanged, expected), same 24 residual municipalities / 11,023 people as prior runs (Hartmannsdorf-Reichenau still largest at 1,040).

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

Field recovery (45 companies): name/address/phone/website_url/analysis_url 45/45; email 41/45 found (4 not_found, incl. Cloudflare-obfuscated Ilmtalgruppe); emergency_phone_number 42/45 found (3 not_found, each earned by actually checking for a dedicated Notdienst page before concluding absent); short_name 29/45; address_opening_hours 29/45, phone_opening_hours 28/45 (still the weakest fields, as every pilot has found); contact_form_url 23/45 real forms; handelsregister_nr 9/45 found (all GmbH/GmbH & Co. KG), 36/45 not_applicable (Zweckverbände/Gemeinden/AöR/Eigenbetriebe, several with explicit "Körperschaft des öffentlichen Rechts" Impressum wording, others inferred more weakly from legal form alone and labelled as such); supply_area_municipalities 40/45; multi_zone true for 25/45 (56%, above the running baseline).

Notable scope findings (not non-suppliers, but narrower/different than roster label suggested):

• wv-de-10021 NOW (Zweckverband Wasserversorgung Nordostwürttemberg): a wholesale/Fernwasser bulk supplier (50 Gemeinden + 5 Stadtwerke + 16 Zweckverbände as members) — its 71 roster-links recorded with role: "bulk" rather than retail, per prior campaign precedent for this pattern.
• wv-de-4338 Markt Mömbris: only directly supplies Ortsteil Niedersteinbach; the rest of the Markt is served by two separate external Zweckverbände.
• wv-de-4281 Ochsenfurt: actual operator is a legally distinct Kommunalunternehmen (KSO, kso-ochsenfurt.de) with its own site, not the Stadt directly; explicitly does not publish a full analysis online, only on request — recorded as an honest low-confidence gap, not guessed.
• wv-de-941 Amt Bargteheide-Land: the Amt is itself the Wasserversorger for 6 of 8 member Gemeinden; 2 others (Delingsdorf, Todendorf, plus Ortsteile Timmerhorn/Kl. Hansdorf) are served by unrelated third parties (Holsteiner Wasser, WBV Oldesloe-Land, Hamburg Wasser) — not conflated into this company's links.
• wv-de-224 Gemeindewerke Hohenwestedt: Aukrug is run by a separately registered sibling entity, not this company_id.
• wv-de-7020 Wasserwerk Bopfingen: genuine multi-zone case (9.3-18.5 °dH across 4 zones) with unusually transparent own-vs-bulk-bought breakdown (own Wasserwerk Aufhausen vs. 3 neighbouring Zweckverbände).

New host/methodology traps, all measured this pass:

• neuenrade.de is JS-rendered (ColdFusion, content-length: 0 on every raw curl) — needed WebFetch/WebSearch entirely.
• pirmasens-land.de returns HTTP 403 to bare curl but works with a full browser User-Agent + Accept-Language header (simple WAF, not a captcha).
• wasserportal.info (BDEW third-party consumer portal) is the designated channel for at least two companies (Kissing, Bad Schwalbach) instead of an own-domain PDF.
• zv-pettenhofenergruppe.de: broken/self-signed TLS on www. plus a 302 that drops the request path — fetch the bare domain directly with curl -k.
• raubling.de: HTTP 403 to curl's default User-Agent; needs a full browser-style UA + Accept-Language header.
• extertal.de (ColdFusion): ordinary .html pages return HTTP 200 / content-length 0 to curl even with a cookie jar; only /city_info/display/dokument/show.cfm?... document endpoints work — WebFetch needed for all normal pages.
• bargteheide-land.de serves ISO-8859-15 — same class as the known neunkirchen-siegerland.de trap; pipe through iconv -f ISO-8859-15 -t UTF-8.
• rheinau.de: Nuxt.js SSR app — WebFetch returned 403, but raw curl HTML contains a large embedded JSON state payload with real content, readable by direct byte-offset inspection.
• A CMS family (hw_/ionas4 classes, seen on Nußloch, Marktheidenfeld, Ochsenfurt/KSO, Grünwald, Extertal) renders news/detail pages mostly empty via curl — WebFetch was the reliable fallback.
• Third-party aggregator sites (wasseranalyse-trinkwasseranalyse.de, wasserqualitaet.net) surfaced in searches for Dautphetal and Mömbris but were correctly rejected as generic template content, not operator-published data.
• No PDF-rendering tool (pdftotext/poppler/strings) is available in this sandbox — PDF existence/liveness was confirmed via HTTP but bodies were not text-extracted where WebFetch couldn't summarize them either; flagged explicitly in the affected records' notes rather than guessed at.

For next run: continue with build_enrich_batch_excl.mjs 45. 0 pending_domain. No new human-queue items from this pass beyond the scope notes above (all resolved inline, not left ambiguous). Standing items from prior runs (Beelitz/Meerane roster defects, Solms durable host issue, Linden/Dahner Felsenland CMS migrations, Glückstadt Wasserwerk disagreement, WWS Saarwellingen entity-naming ambiguity, Gauting/Würmtal re-key, Gilching duplicate, Heilbad Heiligenstadt non-supplier suspicion, wv-de-16016 Ainring / wv-de-3724 Prenzlau self_only contradictions, Oberasbach RangauWerke freshness gap, Sinzig/Rhein-Ahr merger, Kronach HRB conflict, Hille/Hartum mis-key, Woringen Ungerhausen contradiction, Halver/Kierspe naming coincidence, Römerberg-Dudenhofen domain collision, Grefrath 403 trap, Raunheim CMS migration, WVS Siegen-Wittgenstein own-enrichment gap, Krumbach/Stauden double-claim, Neutraubling self-supply unconfirmed, vgwerke-weilerbach.de trap, OeWA Oebisfelde _pop_served scope mismatch, Sulz-am-Neckar bulk-supply Stadtteile, Kerken Kamp-Lintfort claim unconfirmed, Meitingen/Nordendorf Ortsteil boundary, WLV Ost-Obereichsfeld member-list revision) still open, none touched this pass.