de:wasserversorger-discover

Journal entry appended successfully. The run is complete.

Wasserversorger discovery run complete: population coverage rose from 97.65% → 97.95% (+60 municipalities claimed, residual now 870/1.71M people), roster grew to 2,846 companies (+30 net). Both 30-unit shards resolved cleanly; the mandatory cross-shard collision check caught 4 duplicate mints from shard B (cosmetic name variants like (WKK)/(ZWE)/(FWO) suffixes defeating the reuse grep) — all fixed pre-merge, final collision count back to the standing 27, nothing new introduced. 848 singletons remain for the next batch; the 10-unit/22-municipality multi-member pool and the 27-collision backlog still await the overdue human decisions.

Read CLAUDE.md and journal tail, then SKILL.md in full. On-disk state before starting: 2816 companies, 97.65% population coverage (10013/930 claimed/residual), 908 singletons remaining, multi-member pool unchanged at 10 units/22 municipalities (still awaiting the overdue human decision, correctly left alone), 27 standing domain collisions.

Batch build: ran coverage.mjs/build_gv_layer.mjs fresh, confirmed exact match against the journal. Built the batch by hand (no same-day-exclusion flag exists in build_batch60.mjs yet — 6th run in a row hand-rolling this) to exclude the 10-unit multi-member pool and the 9 checked_at:"2026-08-31" unresolved AGS from earlier runs today. Caught my own bug before dispatch: a first-draft filter excluded same-day AGS only when the whole unit was a singleton (members_total===1), which missed Talheim (08125094, one of the 9) resurfacing as the one residual member of a 2-member unit ("GVV Flein-Talheim") — same AGS, different vkey framing. Fixed by filtering at the member level before ranking, not the unit level. Worth flagging: any future same-day-exclusion flag added to build_batch60.mjs needs to do the same — an AGS can hide inside a multi-member unit even when it was excluded as a singleton earlier the same day. Took the top 60 residual singletons by residual_population (4,307 down to 4,066 — marginal yield still shrinking).

Shard A: 30/30 resolved, 0 unresolved. 16 new companies minted (mostly Eigenbetrieb self-supply: Edling, Schnaittenbach, Wilster, Metten, Empfingen, Meinhard, Neureichenau, Gefrees, Schwarzenbach a.Wald, Bodenwöhr, Twistetal, Schenklengsfeld — plus genuinely new Zweckverbände: Wiesentgruppe/Gößweinstein, Neufahrn i.NB-Oberlindhart, Gruppe Winkelhaid, KVL/Boxberg), 14 reused via pre-dispatch grep -F. Near-misses caught before minting: boxberg.de resolves to the wrong Boxberg (BW, not Boxberg/O.L. Sachsen); stadtwerke-schwarzenbach.de belongs to the wrong same-named Schwarzenbach (a.d.Saale, not a.Wald); gw-n.de is Gemeindewerke Niederkrüchten (NRW), not a Neureichenau lead; Simbach identity (Rottal-Inn vs Passau, two real "Simbach" Gemeinden) resolved by checking the local GV100AD_30062026.txt directly since WebSearch was internally contradictory.

Shard B: 30/30 resolved, 0 unresolved before dedup (see below). Initially reported 18 new mints (16→14 after this run's fix), 12 reused. New host quirks: m.gorxheimertal.de cert mismatch (points to an unrelated abstatt.de cert, worked around via www. host); goerwihl.de 302-redirects to www.goerwihl.de; marktbruck.de self-signed cert (curl -k); niederau.info and rmg-poppenhausen.de/mitgliedsgemeinden/ 404'd on direct fetch despite search snippets implying the page exists (worked around via WebSearch-synthesis corroboration, flagged medium-confidence); no pdftotext/python3 in this shard's environment, blocking 2 Satzung PDFs (Weinbach, Wietzendorf) — worked around via search corroboration. Real "Eging" name-collision trap caught before use.

Mandatory cross-shard-plus-existing-roster domain-collision recompute (over all roster.part-*.jsonl shards, never roster.jsonl itself) found 31 — 4 more than the standing 27, all four traced to shard B, all the same reuse-before-mint-grep-miss failure mode flagged in the 2026-08-31 13:00 run's journal entry, now with a new sub-pattern: a parenthetical abbreviation suffix defeats a name grep just like quote/hyphen/suffix variants did before. rmg-poppenhausen.de (new wv-de-19214 "…Rhön-Maintal-Gruppe (RMG)" duplicating pre-existing wv-de-1202, identical name even), zvwkk.de (new wv-de-19208 "Zweckverband Wasserversorgung Kleine Kinzig (WKK)" vs pre-existing wv-de-1739 "Wasserversorgung Kleine Kinzig (Zweckverband)"), eggolsheimer-gruppe.de (new wv-de-19206 "…Eggolsheimer Gruppe (ZWE)" vs pre-existing wv-de-18802 "…Eggolsheimer Gruppe"), fwokronach.de (new wv-de-19213 "Zweckverband Fernwasserversorgung Oberfranken (FWO)" vs pre-existing wv-de-8523 "Fernwasserversorgung Oberfranken"). Confirmed all 4 are the same entity as the pre-existing row (identical domain, land, supplier_kind) before fixing. None of the 4 newly-claimed AGS (Hallerndorf, Hornberg, Untersiemau, Schwebheim) had a pre-existing link, so this was company-level duplication only, no link-level dupes. Fixed all 4 before any merge step ran: repointed the 4 roster-links.part-1356-shardB.jsonl rows to the older/pre-existing company_id, dropped the 4 duplicate roster.part-1356-shardB.jsonl rows (18→14). Recomputed the collision check after the fix: exactly 27, matching the standing backlog, confirming the fix was complete and introduced nothing new.

Merge/finish: map_roster.mjs: 13805 → 13866 roster-links rows (+61, exact = 30+31), needs-review queue 1067 (flat, driven by newly-harvested no-AGS names, not a regression). consolidate_roster.mjs: 3067 shard rows (221 dupes) → 2846 distinct companies (+30 = 16+14, exact, matching the post-dedup mint counts). coverage.mjs: municipalities claimed 10013 → 10073 (+60, exact — all 60 dispatched resolved, 0 unresolved this batch), population 97.65% → 97.95% (+0.30pp), residual 930 → 870 municipalities / 1.964M → 1.713M people. build_gv_layer.mjs: multi-member pool unchanged at 10 units/22 municipalities, singletons 908 → 848 (-60, exact). Final domain-collision check across roster.jsonl + companies.jsonl: 27, matching the standing backlog.

Crossed 97.9% population coverage this run — within half a point of 98%.

For next run: continue singleton batches by residual_population descending — 848 singletons remain, largest still ~6.6k (Schelklingen, still same-day-excluded, now excluded on 3 separate run-days). The 9 same-day-excluded AGS (Talheim included) re-enter the eligible pool once the run date advances past 2026-08-31 — when building that filter, exclude at the member level, not the unit level, since a same-day AGS can resurface under a differently-named multi-member unit (this run's Talheim/"GVV Flein-Talheim" case; caught before dispatch this time, but only by eye). The 10-unit/22-municipality multi-member queue is still awaiting the overdue human decision — do not spend another run on it without new information. Standing 27-domain-collision backlog unchanged (4 different new ones fixed, no net change) — still awaiting the overdue human consolidation pass on the rest. Reinforcing yet again, with a new variant: a name grep before minting is not sufficient — this run's 4 misses were all a matching entity name with a parenthetical abbreviation added or reordered ("(WKK)", "(ZWE)", "(FWO)" suffixes, "(RMG)" already present on both sides). The end-of-batch domain-collision recompute remains the actual safety net and caught 100% of this run's duplicates; keep running it every batch without exception — still worth a human building the small check_collisions.mjs --diff-against-standing script recommended by the prior run, since nothing currently separates "new this run" from the standing 27 except manual diffing. _tasks/netzbetreiber-contact/german-water/ remains absent from this checkout, unaddressed again this run (not touched, out of scope for discovery).