de:wasserversorger-discover

Phase 2b is underway: population coverage rose from 80.3% to 81.7% (+261 municipalities claimed, roster grew from 1,399 to 1,525 companies). Found a genuine but regionally-limited bulk source (Wasserwirtschaftsamt Aschaffenburg publishes municipality→supplier tables for 5 Unterfranken Kreise — confirmed not statewide after checking 3 other offices) and harvested it alongside a standard 30-unit multi-member Gemeindeverband grind, run as two concurrent synchronous subagents. Domain-collision check found only 2 new, both deliberate/documented (not merge errors); two genuine data conflicts with prior roster entries (Blankenbach/Krombach, Kreuzwertheim) were flagged in the roster and journal rather than silently resolved, and journal is updated with exact next-batch pointers (30 more units queued, company_id range wv-de-4701+).

Chat-triggered (users/108934946863892655718). Phase 2b was approved earlier today (see the DECISION entry above) — this is the first run to actually act on it.

Bulk-source test (SKILL.md's mandated first action), result: partial yes, capped as instructed. Checked whether a Kreis-level Gesundheitsamt/Landkreis register of Wasserversorgungsanlagen exists. Found one: Wasserwirtschaftsamt Aschaffenburg (wwa-ab.bayern.de) publishes, per Kreis it covers, an official "Wasserversorger in Stadt und Landkreis X (Stand 2024)" table mapping every Gemeinde/Ortsteil to its named supplier + phone contact — exactly the bulk source hoped for, and a government primary source (TrinkwV-driven), not marketing copy. Tested whether this generalises: it does not. Checked 2 more of Bavaria's 17 Wasserwirtschaftsämter (Nürnberg, Bad Kissingen) and 1 more from a different Regierungsbezirk (Rosenheim) — none publish an equivalent table, only generic Trinkwasser information pages. So this is a regional one-office win (5 Kreise: Würzburg, Miltenberg, Main-Spessart, Kitzingen, Aschaffenburg — all Unterfranken), not a statewide shortcut. Do not re-test the other 13 WWA offices on the strength of this — the pattern is now confirmed non-uniform. If a future run wants to keep looking for a bulk source, the untested angle is still the Gesundheitsamt/Kreisverwaltung register (never checked), not more Wasserwirtschaftsämter.

Harvested all 5 WWA-AB tables (243 rows) via a synchronous subagent (data pre-fetched and parsed by me first, so the subagent did entity resolution/verification, not live crawling of the source table itself). Ran concurrently with a second synchronous subagent doing the standard phase-2b multi-member unit grind (30 non-Bayern units, the top of gemeindeverband-queue.csv by residual member count) — exactly 2 concurrent subagents, per the skill's hard cap. Both finished cleanly, both self-caught and fixed duplicate-mint mistakes via their own end-of-sweep collision checks before reporting (worth noting as a QA pattern that worked as designed).

Results merged and verified:

• Population coverage: 80.3% → 81.7% (+1.4pp, +1,172,728 people)
• Municipalities claimed: 7,082 → 7,343 (+261)
• Residual: 3,861 → 3,600 municipalities, 16.47M → 15.30M people
• Roster: 1,399 → 1,525 companies (+126: 87 from WWA-AB, 39 from the VG grind — both counts are post-dedup against the existing roster, several near-duplicate mints were caught and reused instead)
• Multi-member GV units remaining: 673 → 619; singletons 1,951 → 1,882
• Re-ran the domain-collision check across every roster.part-*.jsonl shard and the consolidated roster.jsonl (per the skill's explicit last-step requirement) both before and after consolidate_roster.mjs: 5 collisions total, same set both times, 3 pre-existing (sw-augsburg.de, zvo.com, emkendorf.de — known parent/subsidiary or shared-site cases from earlier runs) and 2 new this run (wvv.de: Stadtwerke Würzburg AG + Wassergewinnung Würzburg-Estenfeld GmbH, a subsidiary hosting its Umwelterklärung on the parent's site; vgrd.de: Gemeindewerke Dudenhofen + Verbandsgemeindewerke Römerberg-Dudenhofen, two legal entities sharing one VG site) — both flagged by the subagents themselves as deliberate, not accidental merges. No collision was auto-merged.

Two real data conflicts surfaced, deliberately NOT auto-resolved — need a human/future-run look:

• wv-de-1116 "Gemeinde Blankenbach (Wasserversorgung)" and wv-de-1117 "Gemeinde Krombach (Wasserversorgung)" are recorded as self-supplied from an earlier, unsourced-to-this-run pass. The new WWA-AB table (dated 2024, authoritative) says both are actually supplied by Zweckverband Wasserversorgung-spessartgruppe FWS. Added the new links alongside the old ones with an explicit conflict note on each; did not delete/edit the old rows.
• Similarly, existing wv-de-957 (Zweckverband Marktheidenfelder Gruppe) lists Kreuzwertheim as a member via a WebSearch-synthesised note (not a primary source), but the WWA-AB table says Kreuzwertheim self-supplies via its own Satzung. New company wv-de-4352 (kreuzwertheim.de) minted and flagged against wv-de-957, again without touching the old row. Both need a human pass to decide which record is right (or whether it's a genuine partial/former delegation) — this journal note plus the in-row conflict notes are how to find them again.

Two pending_domain rows got candidate domains found but not applied, out of caution for the merge pipeline: Gemeinde Birkenfeld (wv-de-2627) → gemeinde-birkenfeld.de, Gemeinde Erlenbach b. Marktheidenfeld (wv-de-2628) → gemeinde-erlenbach.de. Worth applying in the next enrich pass.

VG grind shard results (30 units, 123 unclaimed members, 111 claimed / 12 left unresolved): RP (6 units, 24/24), BW (14 units, 56/61 — Hasel, Hausen im Wiesenthal, Biberach, Nordrach, Oberharmersbach left unresolved after real checking, not skipped), Niedersachsen (6 Samtgemeinden, 24/24), MV (1 unit, 4/4), Sachsen (2 units, 8/8). Schleswig-Holstein's Mittelholstein (7 members) was re-investigated and is still fully unresolved — this is now the second confirmed no-result pass on it (first was 2026-08-22), so treat it as a genuinely hard case, not an under-searched one, if it resurfaces at the top of the queue again.

For next run: the other 30 units of this same top-60 batch (shard 2, Baden-Württemberg-heavy incl. VVG Rottenburg am Neckar, Hockenheim, Waldshut-Tiengen, and Brandenburg/Niedersachsen/Sachsen units) were selected but never dispatched this run — company_id range wv-de-4701+ is clear to use for them (this run used up to ~4453; check roster.jsonl's max before minting). Re-run build_gv_layer.mjs fresh first since the numbers above already reflect this run's claims. Multi-member-first ordering continues to hold: 619 units / 1,718 municipalities still ahead of the ~1,882 pure singletons.