de:wasserversorger-enrich

Run complete. Enriched 45 German water-supplier companies (all pending_domain rows), appending to enrichment.jsonl (1060→1105 distinct companies) and 119 new municipality-link rows; population coverage held steady at 99.96% since most of this batch turned out to be duplicate mints.

The standout finding: of 33 domains resolved this run, 23 collided with an already-existing, lower-numbered company_id — the entire batch traced back to one WWA Ansbach/Traunstein discovery sweep (ids wv-de-20500–20609) that apparently minted new company_ids without checking against the existing roster, roughly doubling the campaign's domain-collision backlog (27→51 pairs) in a single run. Per campaign rules nothing was auto-merged; both sides of every collision are recorded and flagged for a human dedup pass, along with two roster issues needing a decision (a merged/defunct Zweckverband, and a Bundeswehr admin office misclassified as a public utility). Full details and next-run guidance are in the journal entry appended just now.

Read CLAUDE.md and journal tail, then SKILL.md in full. State on entry: roster.jsonl 3364 rows, 1060 distinct enriched, 99.96% population coverage, 108 pending_domain in the pending tier (matched the 2026-09-01 20:30 run's numbers exactly). python3 confirmed absent again (/usr/bin/python3 does not exist).

Batch build: build_enrich_batch_excl.mjs 45 enrich-20260902-0030 (standing 6-id exclusion set, no overlap with this batch, confirmed). All 108 pending_domain rows outrank population this run, so the full 45-row batch was pending_domain — split 23/22, dispatched as two synchronous general-purpose Agent subagents in one message, both run_in_background: false, both carrying the mandatory no-Agent/no-Task instruction and the full method inlined (worked JSONL example from wv-de-3965 ZV Naab-Donau-Regen). Both completed inline (~1049s / ~1329s), both shards validated as well-formed JSONL matching the dispatched batch exactly (23/23, 22/22), zero overlap with pre-existing enrichment.jsonl, zero internal duplicates.

Domain resolution: 33/45 resolved to a genuine domain, 12 stayed pending_domain with an earned reason. Two of the 12 are worth a human look rather than a re-crawl: wv-de-20597 (Zweckverband Rastberg-Gruppe) merged 2026-01-01 into a new "Zweckverband Hesselberg-Gruppe" together with Stadt Wassertrüdingen — successor contact recorded in the enrichment notes, this company_id is now stale; wv-de-20524 (Standortverwaltung Bad Reichenhall) is a Bundeswehr site administration office, not a public retail water utility — likely a roster misclassification, not a research gap.

Major finding: of the 33 resolved domains, 23 collide with an already-existing, lower-numbered company_id already on the roster — this whole batch was near-entirely duplicate mints. All 45 batch rows trace to one source note ("Named in WWA Ansbach/Traunstein per-Ortsteil Trinkwasserversorger table"), i.e. one discovery sweep minted ids in the wv-de-20500–20609 range without checking them against the existing roster by name/area first. Confirmed pairs (new id → pre-existing id, domain in common): 20592→3744 (sw-wug.de), 20587→5534 (pfofeldergruppe.de), 20585→8052 (reckenberg-gruppe.de), 20590→5535 (gnotzheimergruppe.de), 20596→19564 (buechelbergergruppe.de), 20607→3976 (sw-trl.de), 20606→5515 (rieswasser.de), 20574→3514 (sw-traunstein.de), 20529→3765 (freilassing.de), 20500→3981 (altoetting.de), 20512→17501 (neuoetting.de), 20577→14526 (stadtwerke-trostberg.de), 20535→7520 (surgruppe.de), 20505→5507 (otting-pallinger-gruppe.de), 20501→4227 (burgkirchen.de), 20506→18505 (garching-alz.de), 20567→18513 (rathaus-siegsdorf.de), 20527→18591 (gemeinde.berchtesgaden.de), 20530→18700 (stadtlaufen.de), 20520→11021 (unterneukirchen.de), 20544→18742 (grassau.de), 20559→18764 (ruhpolding-rathaus.de), 20569→18890 (tacherting.de). Three of the pre-existing sides were already enriched (5515, 5535, 7520) — meaning three companies in this batch were fully re-researched from scratch for no coverage gain. Not merged — per the campaign's own rule, domain collisions are recorded on both rows and flagged, never auto-merged. Only 10 of the 33 resolved to a domain genuinely new to the roster: wfw-franken.de (20591), swg-gun.de (20595), wettelsheimer-gruppe.de (20601), gw-pleinfeld.de (20603), wbv-axdorf-einham.de (20575), wbv-kammer-rettenbach.de (20576), annahuette.com (20523), ettenstatt.de (20594), gemeinde.bischofswiesen.de (20528), vgem-altmuehltal.de (20602, flagged separately below). This roughly doubles the campaign's domain-collision backlog in one run (27 known before → 51 total now) and is a strong signal that the wv-de-205xx/206xx id range specifically needs a human dedup pass, and that whatever discovery method produced it should gain a name/area collision check against the existing roster before minting a new id.

wv-de-20602 (Gemeinde Meinheim) resolved to vgem-altmuehltal.de, the shared Verwaltungsgemeinschaft-Altmühltal site, not a Meinheim-only domain — accepted per the campaign's own guard for VG-shared sites (a named "Gemeinde Meinheim" subpage on the VG's own official site, Impressum-equivalent), but flagged since other municipalities in the same VG could independently claim the same domain later; no such collision exists yet. Confidence throughout is medium/low (no water-specific content found for Meinheim itself, tier-4 analysis) — a genuine thin result, not a resolution error.

Roster row edits this run: hand-updated official_domain + domain_status: pending_domain → resolved directly on the 33 roster.jsonl rows above (matching the precedent set by the 2026-08-30 05:30 run) — not via consolidate_roster.mjs, which rebuilds roster.jsonl purely from roster.part-*.jsonl shards and would have silently discarded these hand edits since they exist nowhere in a shard file.

Other roster-data notes: Stahlwerk Annahütte (wv-de-20523) confirmed as a genuine drinking- water supplier despite looking like an industrial-only entry (real HRA 4666, one of Ainring's 4 named public suppliers). Ettenstatt (wv-de-20594) and — more weakly — Meinheim (wv-de-20602) have no water-supply content anywhere on their own sites despite a roster eigenbetrieb classification; worth checking whether either is actually served by an outside Verband instead of self-supplying. Grassau/Rottau (20544/20545) and Surgruppe/Otting-Pallinger-Gruppe (20535/20505) population/address overlaps were checked and are benign (distinct legal entities, shared Geschäftsstelle in the Surgruppe/Otting-Pallinger case). Stadtwerke Traunstein's roster-supplied email ([email protected]) looks stale against the current Impressum ([email protected]).

New host trap: DNS shows reckenberg-gruppe.de, gnotzheimergruppe.de, buechelbergergruppe.de and pfofeldergruppe.de all resolve to 78.46.40.200 — the previously-known hard-ban host, now confirmed shared by five campaign companies (with fernwasser-franken.de), not two. Budget under 7 fetches across all five combined, not per-domain.

Field recovery: name 45/45, address ~41/45, phone ~41/45, website 33/45 (= domain resolution count), email lower (JS/image-obfuscated Impressum emails hit repeatedly on small-Gemeinde sites, e.g. Ettenstatt), opening hours the weakest fields as usual for small Zweckverbände/Eigenbetriebe. Analysis tier: mostly tier-3/4 for the tiny Wasserbeschaffungsverbände that stayed pending_domain, mixed tier-1/2 for the resolved municipal/Stadtwerke rows. Combined ~126 fetches, ~65 searches across 45 companies.

Merge/finish: appended both shards directly to enrichment.jsonl (1065→1110 lines, 1060→1105 distinct company_id, +45 exact). Appended roster-links.part-shard{A,B}-20260902- 0030.jsonl (69+50=119 rows) and ran map_roster.mjs (16644→16763 roster-links rows, +119 exact). coverage.mjs: population 99.96% → 99.96% (unchanged — expected, since 23 of the 33 resolved companies serve territory already claimed by their pre-existing duplicate), residual 37 municipalities / 35,813 people (roughly flat). pending_domain in the pending tier: 108 → 63 (all 45 batch rows left the pending tier by getting an enrichment record, regardless of whether the domain itself resolved — same mechanic noted in prior runs).

For next run: continue by population served — build_enrich_batch_excl.mjs 45 picks up where this run left off; 63 pending_domain rows remain in the pending tier and discovery keeps minting more. Priority for a human: the wv-de-205xx/206xx duplicate-mint cluster above (23 new collision pairs in one batch) is now the single biggest addition to the standing domain-collision backlog (51 total) and the most concrete lead yet on why the backlog keeps growing — worth checking whatever discovery pass produced the WWA Ansbach/Traunstein-sourced ids for a missing name/area dedup step before it mints more. Also needs a human decision: wv-de-20597 (Rastberg-Gruppe, merged into a new Hesselberg-Gruppe 2026-01-01) and wv-de-20524 (Standortverwaltung Bad Reichenhall, likely not a public retail utility). Standing backlog otherwise unchanged: 6 duplicate-mint pairs (pre-existing, excluded from batches by id), GIPS- platform email cipher (Balingen/Gaggenau/Hemer/Wedel, still uncracked). SKILL.md's Finish section still names the stale map_ags.mjs/export_csv.mjs pair instead of the live map_roster.mjs + coverage.mjs path — flagged across many consecutive enrich runs now, still unfixed.