de:wasserversorger-discover

Shard files preserved as evidence trail, per the resumability rule. The run is complete.

Outcome: Processed a 60-singleton batch via two synchronous subagents, resolving 58 municipalities (16+18=34 new companies minted, 24 reused via existing roster) and honestly recording 2 unresolved (Frankenblick, Grebenstein). Population coverage rose from 96.24% to 96.62% (1170→1112 residual municipalities), roster grew to 2708 companies, and the domain-collision check stayed at the standing 27 with no new duplicates. The 10-unit/22-municipality multi-member queue remains parked awaiting human review, and a new discrepancy (Velburg/Seubersdorf vs. Zweckverband Laber-Naab membership claims) was flagged for that same pass rather than resolved unilaterally.

Read CLAUDE.md and journal tail, then SKILL.md in full. On-disk state before starting: 2674 companies, 96.24% population coverage (9773/1170 claimed/residual), multi-member pool unchanged at 10 units/22 municipalities (still awaiting the overdue human decision — untouched again).

Same-day-unresolved filter applied via a one-off script (build_batch60_filtered.mjs, deleted after use): the plain top-60 from build_batch60.mjs still included the 6 AGS the 22:00 run had just marked unresolved with checked_at: 2026-08-31 (Schelklingen, Römhild, Weiler-Simmerberg, Berga-Wünschendorf, Gmund a.Tegernsee, Leun) — same day, no new information since, so excluded them and backfilled from the next-ranked singletons instead. build_batch60.mjs still has no built-in same-day exclusion; this is now the third run to hand-roll the same filter (see 08-30 18:00 and 08-31 00:30 entries) — worth promoting to a real script flag rather than reinventing it each time.

Batch: top-60 singletons by residual_population (post-filter), split 30/30, dispatched as shards "shardA-0300"/"shardB-0300", both subagents run_in_background: false in a single message (synchronous per the SKILL banner), mandatory no-Agent-tool line inlined, full method + schema + host-trap list inlined directly in the prompt. Company_id ranges wv-de-18901–18930 / wv-de-18931–18960 (verified against the actual max used ID, 18900, before assigning — the last run's assigned range went up to 18914 but only used to 18900).

Shard A: 30/30 resolved, 0 unresolved. 16 new companies (wv-de-18901–18916), 14 links reused via existing roster (WAZV Lausitz, DERAWA covering two municipalities from one company, TAZ Helbe-Wipper, WAZ Seelow, ZV Isar-Vils, FairEnergie, OVAG, REWAG, Wasserverband Peine, LWG Lausitzer Wasser, Wasserzweckverband Rottenburger Gruppe, HWG Hohenloher). Baar-Ebenhausen produced 2 link rows (genuine Nord/Süd zone split between two existing companies, both reused). New host issues: Limeshain's Satzung PDF was scanned/CCITT-compressed and unreadable without pdftotext (not installed) — resolved via a third-party NGO page quoting the Gemeinde verbatim instead; Wassergewinnung Vierseenland's TLS cert covers *.kasserver.com not its own domain, Impressum unreachable via WebFetch, corroborated via Creditreform/genios instead. Gotchas caught, not mistakes made: ASG (Ammertal-Schönbuchgruppe, existing wv-de-1746) is the wholesale network operator for Dettenhausen/Walddorfhäslach, not the retail claim — each got its own Eigenbetrieb row instead; ewag kamenz is only WAZV Lausitz's Geschäftsbesorger (business manager), not the supplying entity — not credited as the supplier; a "VG Pfaffenhausen" (Unterallgäu) search hit was a name-collision trap, unrelated to Markt Pfeffenhausen (Landshut). Rottach-Egern's supplier is an e.V. (Wasserversorgungsverein), recorded supplier_kind: "other". Schweitenkirchen (Geroldshausener Gruppe) left pending_domain — no confirmed website found.

Shard B: 28/30 resolved, 2 unresolved. 18 new companies (wv-de-18931–18948), 10 links reused via existing roster (Stadtwerke Rosenheim, Rheingauwasser, FairEnergie, Kreiswerke Main-Kinzig, Stadtwerk Tauberfranken, KWA Meininger Umland, WAZV Eichsfelder Kessel, Zweckverband Wenzenbacher Gruppe, WVV Rotenburg-Land, Zweckverband Sengenthal-Deining — the last upgraded medium→high confidence with a fresh primary source). 14 of 18 new companies are self_supplied Eigenbetriebe (Velburg, Waakirchen, Wegscheid, Grävenwiesbach, Ludwigsau, Stühlingen, Guxhagen, Kirchdorf a.Inn, Ranstadt, Großaitingen, Dietzhölztal, Schönau a.Königssee, Eltmann, Helsa, Seubersdorf), each actively checked against an external Verband first, matching the roster's existing convention of giving self-supply its own company row rather than a null. Two genuine unresolveds: Frankenblick (Thüringen — official site frankenblick.eu had a DNS resolution failure in this environment despite appearing in search results; Fernwasserversorgung Franken, Rennsteigwasser and ZWAS Suhl all checked and ruled out as wrong-region) and Grebenstein (Hessen — Wasserwerte and Satzungen pages checked, no operator named, no Zweckverband found). New host issues: wazv-mittlere-wesenitz.de (Stolpen) exists only as an email domain, ECONNREFUSED on http/https — pending_domain despite high-confidence name/legal-form; stadtwerk-kuelsheim.de has a kasserver.com cert mismatch, used the working stadtwerk-tauberfranken.de instead (same company). Flagged for human reconciliation, not overridden: Velburg and Seubersdorf i.d.OPf. both have primary-source evidence (own Ortsrecht/Wasserversorgung pages) of self-supply for the Kernort, while existing roster row wv-de-7561 (Zweckverband Laber-Naab) lists both as members via indirect Gremien-page corroboration — likely Verbandsversammlung-representation vs. actual-supply-area mismatch, documented in notes on both new rows rather than resolved unilaterally.

Validation before merge: all 4 shard files well-formed JSONL (16+18=34 roster rows, 31+30=61 link rows — the 61 not 60 is Baar-Ebenhausen's two-zone split); shard A ids 18901–18916, shard B ids 18931–18948, zero overlap. Duplicate-mint check across both new shards + full existing roster (all roster.part-*.jsonl + roster.jsonl): exactly 27 collisions, all six already-known duplicate-mint pairs, none touching this batch's 34 new companies.

Merge/finish: map_roster.mjs: 13385 → 13446 roster-links rows (+61, exact), needs-review queue 938 → unchanged in count reported (938, same churn source as before — new rows without direct AGS matches). consolidate_roster.mjs: 2929 shard rows (221 dupes) → 2708 distinct companies (+34, exact). coverage.mjs: municipalities claimed 9773 → 9831 (+58 = 60 dispatched − 2 unresolved, exact), population 96.24% → 96.62% (+0.38pp), residual 1170 → 1112 municipalities / 3.139M → 2.824M people. build_gv_layer.mjs: multi-member pool unchanged at 10 units/22 municipalities, singletons 1148 → 1090 (-58, exact). Final domain-collision check on merged roster.jsonl (2708 companies): 27, none new.

Crossed 96.5% population coverage this run.

For next run: continue singleton batches by residual_population descending — 1090 singletons remain, largest ~5.4k (Frankenblick, Grebenstein just joined the unresolved pool, checked_at 2026-08-31 — wait a day before re-including, per the same-day-filter precedent now used three runs running). The 10-unit/22-municipality multi-member queue is still awaiting a human decision (4+ failed automated methods across several prior runs) — do not spend another run on it without new information. Standing 27-domain-collision backlog unchanged (six known duplicate-mint pairs), still awaiting the overdue human consolidation pass — this run adds one candidate worth a human look when that pass happens: Velburg/Seubersdorf i.d.OPf. vs. wv-de-7561 Zweckverband Laber-Naab (self-supply evidence vs. membership-list evidence, documented above). build_batch60.mjs still has no same-day-exclusion flag — third run in a row to hand-roll it; worth a human fix. resolve_singleton.mjs's missing _tasks dependency remains unaddressed and was skipped again without spending time re-debugging it, per the 08-30 22:00 precedent.