de:wasserversorger-enrich

All state files are consistent and the campaign artifacts are up to date.

Enriched all 45 batched Wasserversorger (population-priority order, 0 pending_domain), running three research workers in parallel and validating/merging their output cleanly: enrichment.jsonl 1619→1664 records (+45, zero duplicates), roster-links.jsonl +420 rows, population coverage holds at 99.99%. Found 2 roster defects (Stadtwerke Beelitz and Stadtwerke Meerane don't actually supply water — flagged, real operators identified) plus several host/CMS-migration issues (Solms, Linden, Dahner Felsenland, Bad Soden-Salmünster) logged for follow-up; also caught and recovered from a mid-run mistake where a nested sub-agent delegation looked prematurely "completed" — verified via ListAgents that the real work was still in progress and waited for it properly, so no data was lost. Journal updated with full details and a process note for future runs.

build_enrich_batch_excl.mjs 45 → 0 pending_domain (queue stayed empty), batch drawn purely by population priority (13.2k-13.9k _pop_served band). Split into shard A (23) / shard B (22), researched by 3 effective workers (shard B ran inline; shard A's handler delegated to two nested sub-agents, 12 + 11 companies) — all foreground/blocking via TaskOutput(block:true), never left running unattended, per the SKILL.md rule learned 2026-08-04.

Process note for future runs: the first Agent call for shard A was launched with run_in_background: true by mistake, and its handler then reported "completed" after only spawning two child agents in the background — the actual per-company research hadn't happened yet when the parent's task-notification fired. ListAgents showed the two children still running and trackable by their own task IDs, so nothing was lost — polled those directly with TaskOutput(block:true) until both finished. Lesson: always pass run_in_background: false explicitly for headless runs (the tool defaults to background), and if a parent agent's notification arrives suspiciously fast, check ListAgents for live children before assuming the work is done.

Merge: enrichment.jsonl 1619 → 1664 lines / 1659 distinct company_id (+45 clean, verified zero overlap with existing IDs and zero malformed JSON before appending). roster-links.jsonl 18758 → 19178 (+420 raw link rows across all three shards). map_roster.mjs ran clean (usual crop of no_ags_for_this_name unmatched Ortsteil/VG-composite names, nothing structurally new). consolidate_roster.mjs: 177 shards / 3617 rows, still 3375 distinct companies; roster-links.csv now 16214 distinct after dedup (+354 net). coverage.mjs: population 99.99% → 99.99% (unchanged, expected — this batch enriches already-claimed companies, doesn't add new roster coverage). 24 municipalities / 11,023 people still residual, unchanged.

Field recovery this batch (45 companies): email ~38/45, phone ~44/45, emergency_phone_number ~32/45 (several only found after following the site's own Notdienst/Störung link, not the homepage), contact_form_url ~13/45 real forms, analysis_url found for ~36/45 with real evidence, multi_zone true for ~25/45 (consistent with the ~70% pilot rate for genuine suppliers).

Non-suppliers found (2, both recorded per hard rule 6 rather than skipped):

• wv-de-3950 Stadtwerke Beelitz GmbH — Fernwärme/Strom only, no water business; actual operator is Wasser- und Abwasserzweckverband "Nieplitz".
• wv-de-9511 Stadtwerke Meerane GmbH — electricity/gas/fiber/street-lighting only; actual operator is Regionaler Zweckverband Wasserversorgung Bereich Lugau-Glauchau (RZV).

Other anomalies worth a human's attention:

• wv-de-630 Mecklenburgische Schweiz: prior notes said "15 Ortschaften"; raw HTML parsing this pass found 10 Wasserwerke covering 115 places — a 7x undercount, another instance of the documented WebFetch-truncation trap (README.md) rather than a data change.
• wv-de-3653 WWS Wasserwerk Saarwellingen: shared domain's own Impressum names the co-branded parent "TWS Technische Werke der Gemeinde Saarwellingen GmbH" (HRB 52737), not WWS itself — left handelsregister_nr not_found rather than risk attaching the wrong entity's number.
• wv-de-3961 Solms (stadtwerke-solms.de): connection refused on every attempt, second session in a row — looks like a durable host issue, not transient. Recovered via solms.de.
• wv-de-402 Dahner Felsenland and wv-de-3968 Linden: previously-working URLs now 404; Linden shows a CMS migration (TYPO3 → Craft CMS headers) — needs a fresh crawl, not cached paths.
• wv-de-3953 Bad Soden-Salmünster: batch's website_url/impressum_url dead after a 2026 TYPO3 relaunch; current contact and analysis PDF relocated via org-chart/Amtliche-Bekanntmachungen.
• wv-de-6026 Glückstadt: two of the operator's own documents disagree on which Wasserwerk (Krempermoor vs Nordoe) feeds the Kremperheide network — flagged, not silently resolved.
• wv-de-3948 Erbach: corrected a prior-pass error — the recorded impressum_url (id=22) actually resolves to Datenschutz, not Impressum; fixed to id=7.
• wv-de-3621 Neuenkirchen: gemeindewerke-neuenkirchen.de now redirects into a merged Gemeinde portal; HTTPS fails TLS SNI negotiation (HTTP works) — content now lives on neuenkirchen.de.
• wv-de-406 Bordesholm: corrected a prior-round error — Böhnhusen/Techelsdorf/Schönhorst ARE served per the site's own text (previously recorded as excluded).
• wv-de-16013 Hohenkemnather Gruppe and wv-de-1135 Zweckverband Karkbrook: both had materially larger published supply areas than the prior discovery pass recorded (60 vs 8 settlements; 14 vs 10 municipalities) — verified against raw HTML, not AI-summary, to rule out hallucination; genuine fuller data, not a contradiction.
• wv-de-3657 Oberhaching: domain-collision flag with wv-de-3658 (shares oberhaching.de) carried forward, not auto-merged.

Still stale: export_csv.mjs/map_ags.mjs still point at the dead legacy companies.jsonl/links.jsonl pair, not roster.jsonl/roster-links.jsonl — this run used the live map_roster.mjs → consolidate_roster.mjs → coverage.mjs pipeline instead, as every recent run has done (10+ consecutive runs now). Still worth a maintainer fixing.

For next run: continue with build_enrich_batch_excl.mjs 45. 0 pending_domain. Open human-queue items from this pass: Beelitz/Meerane roster defects (wrong operator on roster), Solms durable host issue, Linden/Dahner Felsenland CMS migrations, Glückstadt Wasserwerk disagreement, WWS Saarwellingen entity-naming ambiguity. Standing items from prior runs (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) still open, none touched this pass.