de:wasserversorger-discover

Ran phase 2b batch vg8 of wasserversorger-discover: two synchronous shards worked 60 Gemeindeverband units (mostly Baden-Württemberg VVGs plus a Saxony/Brandenburg/SH tail), minting 44 net new companies and resolving 116 of 125 residual municipalities. Population coverage rose 86.2% → 87.68% (+1.24M people, the largest single-batch gain in several runs), municipalities claimed 8486 → 8588, residual now 2,355 municipalities / 10.3M people — first time under 10.5M residual.

Caught and fixed one genuine cross-shard duplicate mint (EVF GmbH & Co. KG, found independently by both shards) before merging, per the mandatory collision check; final roster-wide check confirmed zero new unexplained collisions beyond the 15 known carried cases. Journal updated with full batch detail, host traps, and next-run guidance (skip Kellinghusen, mint new IDs from 10001+).

Scheduled run. Read CLAUDE.md, this journal (tail — full file is 796KB, exceeds the read tool's 256KB cap, so read via tail), and SKILL.md first. Confirmed on-disk state before starting: 1967 companies, 86.2% coverage (from the 15:30 enrich run, itself on top of the 15:47 vg7 discover run).

Batch build: build_batch_input.mjs 61 vg8, dropped Kellinghusen (vkey 010615189, the settled negative — left alone per standing note) from shard A, leaving 30+30 = 60 units. Top of queue was mostly Baden-Württemberg VVG/GVV units (Friedrichshafen, Göppingen, Bruchsal, Lahr, Lörrach, Albstadt, Winnenden, Balingen, Rheinfelden, Bühl, Bretten, Geislingen, Mühlacker, Überlingen, Bad Mergentheim, Tettnang, Bad Saulgau, Lauffen, Bad Schönborn, Graben-Neudorf, Plüderhausen-Urbach, Uhingen, Hemsbach, Eberbach, Sulz am Neckar, Obersulm, Sinzheim, Weinsberg, Aidlingen/Grafenau, Waldenbuch/Steinenbronn, March-Umkirch, Titisee-Neustadt, Welzheim, Staufen-Münstertal, Gundelfingen, Immendingen-Geisingen, Mittlere Fils-Lautertal, Bad Schussenried, Ilshofen-Vellberg, Rheinmünster-Lichtenau, Schliengen, Bischweier-Kuppenheim) plus a Sachsen/Brandenburg/Bavaria/Schleswig-Holstein tail (Untersteinach, Unterspreewald, Bevern, Trave-Land, Eidertal, Schwedt/Oder, Limbach-Oberfrohna, Bad Salzungen, Torgau/Dreiheide, Reichenbach im Vogtland, Crimmitschau-Dennheritz, Weißwasser/O.L., Mittweida, Meerane-Schönberg, Geisenfeld, Zschopau, Stollberg/Niederdorf).

Split into two synchronous subagents (run_in_background: false on both, dispatched in one message, blocked on both before continuing — no background-wait failure), 30 units each, with the mandatory no-Agent-tool line in both prompts. Neither shard recursed into sub-subagents; both ran long (31–40 min) but stayed synchronous throughout.

Shard results: Shard A (BW VVG-heavy + Bavaria/Saxony tail) resolved 56/65 seeded municipalities, minted 26 companies (wv-de-9001–9027, one ID skipped after a disproven claim); 9 genuinely unresolved (3 tiny Bavarian villages — Guttenberg, Kupferberg, Ludwigschorgast — after discarding a WebSearch-synthesis false lead for "Pressecker Gruppe"; 6 small Amt Trave-Land/Amt Eidertal Schleswig-Holstein villages with no findable Wasserbeschaffungsverband). Correctly applied the Betriebsführung-vs-Konzession distinction (badenova holds a full concession for Lahr → recorded as supplier; badenovaNETZE elsewhere is only technical operator → Gemeinde stays supplier of record) and filtered out a wastewater-only Zweckverband (Frohnbach) per the non-supplier filter. Host traps: rzv-glauchau.de and wwa-ho.bayern.de refused all connections (http fallback too) — worked around via WebSearch corroboration and the existing roster record's own documented trap note.

Shard B (BW VVG/GVV + Sachsen city tail) resolved all 60/60 seeded municipalities: ~54 to a named company (13 matched to existing roster entries after grep-dedup, 19 newly minted wv-de-9501–9519), 9 correctly recorded self_supplied on explicit source statements (Dettenheim, Uhingen, Albershausen, Sulz am Neckar, Titisee-Neustadt, Eisenbach, Süßen, Immendingen, Hügelsheim), 2 honestly left unresolved (Plüderhausen — operator site parked/dead, not confirmed active; Ernsgaden — inter-Zweckverband contract dispute, explicitly expired Dec 2025, no post-2025 source). Flagged one HR-number mismatch (eneREGIO GmbH HRB 522018 vs. already-rostered "Wasserversorgung Muggensturm-Ötigheim GmbH" HRB 733586, both on eneregio.com) rather than silently merging — linked by domain per the dedup rule but left the discrepancy in notes for human review. Host traps: rzv-glauchau.de, wzv-mns.de, wvv-mauracherberg.de all ECONNREFUSED on direct fetch — worked around via web.archive.org. Caught a copy-paste trap: heuweiler.de's "Strom & Wasser" page verbatim republishes Gundelfingen's Gemeindewerke text (shared CMS template bug) — did not treat as evidence, cross-verified via the real Zweckverband Mauracherberg site instead.

Mandatory cross-shard collision check (run before consolidate, per the skill's standing instruction): found 1 new duplicate mint — wv-de-9019 (shard A, "Energieversorgung Filstal GmbH & Co. KG (EVF)", found via Geislingen an der Steige) and wv-de-9515 (shard B, same entity, found independently via a Lauterstein news article) — identical domain (evf.de), identical HRA 531868/Amtsgericht Ulm. Fixed by remapping shard B's 1 link row (Lauterstein) to the canonical wv-de-9019 and dropping the wv-de-9515 roster row before merging. Re-ran the check after the fix: exactly the same 15 historical collisions as before this batch (9 long-standing pairs + 5 legitimate shared-Amt/VG-portal cases from the vg7 run + 1 pre-existing from the 15:30 enrich run) — zero new unexplained collisions.

Merge: map_roster.mjs: 10009 total link rows (was 9884), 205 needing review (201 unmatched — mostly newly-harvested Ortsteil/no-AGS names from BW member lists and small SH/Brandenburg villages, not a regression — plus 3 expanded_qualifier, 1 ambiguous: Breitenbrunn, 2 Bavarian Gemeinden share the name, still needs a human pick). consolidate_roster.mjs: 2085 shard rows (74 shard-level dupes, last-shard-wins) → 2011 distinct companies (+44 net = 26 shard A + 18 shard B after the evf.de fix). coverage.mjs: population 86.2% → 87.68% (+1.48pp, +1.237M people — the largest single-batch gain in several runs, driven by the many BW VVG units' large residual populations, e.g. Friedrichshafen 69k, Göppingen 61k, Bruchsal 58k). Municipalities claimed 8486 → 8588 (+102). Residual: 2457 → 2355 municipalities, 11.537M → 10.300M people — first time under 2,400 residual municipalities and under 10.5M residual population. build_gv_layer.mjs (fresh for next run): units with residual members 2235 → 2118, multi-member 291 → 232 (covering 469 municipalities, was 589), singletons 1895 → 1886.

For next run: New queue top: Kellinghusen (still the settled negative, skip it), Untersteinach (VGem) (still 3 unresolved after this run — Guttenberg/Kupferberg/Ludwigschorgast, genuinely hard, consider a fresh search angle or mark not_published if a second attempt also fails), Trave-Land (3, the SH villages shard A couldn't resolve), Eidertal (3, same), VVG der Stadt Uhingen (2, partial — Uhingen itself resolved self_supplied but 2 other members of that VVG remain). company_id max on disk now 9519 (sparse: 9001–9027 and 9501–9519 used, 9515 retired by the dedup fix) — next batch should mint from 10001+ to stay clear of every prior range; check roster.jsonl before minting regardless, as always. Open duplicate-mint/collision list unchanged at the same 15 pairs — this run found and fixed its own 1 new pair pre-merge (as intended) and added zero net to the carried pile. Two items still flagged for human review from recent runs: wv-de-1415/7505 (Wiesenbachgruppe), wv-de-3925/8502 (Schwabachgruppe), plus now the eneREGIO/Muggensturm-Ötigheim HR-number mismatch on eneregio.com and Breitenbrunn's name ambiguity. The still-overdue human consolidation pass (scripted domain-collision + near-identical-name sweep across the full 2011-row roster) remains recommended and keeps growing more valuable as the roster grows.