de:wasserversorger-discover

Phase-2a batch complete: resolved all 60 assigned kreisfreie Städte (Salzgitter down to Herten), adding 46 new companies and reusing 11 existing ones via domain/name dedupe (3 recorded as pending_domain). Population coverage rose 46.57% → 52.31% (+5.74pp), municipalities claimed 5,802 → 5,862, with no new domain collisions. Phase-2a pool now stands at 1,129 units / 23.78M people remaining (next up: Wesel, Neubrandenburg, Euskirchen); journal updated for the next scheduled run to continue from there.

Continued phase-2a exactly where the 13:00 run left off: took the next 60 units of the singleton pool (residual_population >= 10,000, descending) via lib.mjs::parseCsv on gemeindeverband-queue.csv cross-referenced with gemeindeverband-members.csv for AGS — Salzgitter (104,970) down to Herten (60,941). Split into 2 synchronous subagents of 30 cities each (slices p2b/p2c), each with its own non-overlapping company_id range (3035-3067 and 3068-3100) and its own shard files, both blocked on in this turn per the skill's synchronous-subagent rule.

Result: 60/60 resolved, 0 unresolved. 46 new companies (25 in p2b: wv-de-3035..3059; 21 in p2c: wv-de-3068..3088), 11 reused via the required domain/name dedupe against roster.jsonl before minting — more reuse than the 13:00 batch's 4/40, because several of these mid-size cities turned out to already be served by regional multi-city suppliers already in the roster rather than their own Stadtwerke: GELSENWASSER AG (Marl, Castrop-Rauxel — third and fourth cities on that one company, after Gelsenkirchen/Recklinghausen), RWW mbH (Gladbeck, Dorsten, and — new pattern — Kerpen/Bergheim, which RWW took over retail billing for from Bedburg/Elsdorf on 2026-01-01, a live delegation change worth remembering for any future re-check), LWG Lausitzer Wasser (Cottbus), SWN Neumünster (via Amt Nortorfer Land), ZVME Mittleres Elstertal (Gera — legal Versorger per the city's own page even though "Stadtwerke Gera" is the operating brand), NEW AG (Viersen), Avacon Wasser (Lüneburg), Stadtwerke Neuwied (its own city, distinct from an existing Kreiswasserwerk carve-out), WZV Weimar (its own city, on top of the 30 other municipalities already on file), SWU Ulm/Neu-Ulm (Neu-Ulm, second city on that company). 3 pending_domain rows (Dessauer Wasser-/Abwasser GmbH, Wasserversorgung Rüsselsheim GmbH, hertenwasser GmbH — the last a GELSENWASSER/Hertener-Stadtwerke JV with no independent site) — named and evidenced but not importable, flagged for wasserversorger-enrich's priority queue same as prior pending_domain rows. One bonus haul: Trier's SWT Wasser GmbH page named 3 additional supplied hamlets (Gutweiler, Korlingen, Sommerau) beyond Trier itself, captured as extra roster-links rows with ags: null since they weren't in the assigned batch — a reminder that even in the 1:1 "head" regime, an already-fetched page can still yield a bit more.

Traps avoided, worth remembering: two unrelated companies both abbreviated "EWR" — existing wv-de-3033 (EWR GmbH, Remscheid) vs new wv-de-3052 (EWR Aktiengesellschaft, Worms) — flagged in notes to prevent a future false merge on the short name alone. Two TLS certificate errors (Wasserwerke Zwickau, and Viersen's own utility domain) forced reliance on corroborating third-party/register sources instead of the primary Impressum, both downgraded in confidence rather than guessed past. No CAPTCHAs, IP bans, or forms touched — research only, all fetches via WebSearch/WebFetch against the suppliers' own sites.

Ran the Finish sequence: map_roster.mjs (6,419 links, 0 ambiguous/unmatched) → consolidate_roster.mjs (700 distinct companies, up from 654) → coverage.mjs → build_gv_layer.mjs. Population coverage 46.57% → 52.31% (+5.74pp) from 60 rows; municipalities claimed 5,802 → 5,862 (+60, all singletons as expected — the 3 Trier hamlets have no AGS so don't move this counter). Domain-collision check across roster.jsonl + companies.jsonl: same 1 known hit (emkendorf.de, unchanged since 2026-08-19) — no new collisions from this batch, even with the higher reuse rate above (reuse went through the dedupe grep, not a domain accident).

Phase-2a pool remaining: 1,129 units / 23.78M people (next up: Wesel 60,785, Neubrandenburg 60,344, Euskirchen 60,021, Langenfeld (Rheinland) 59,975, Unna 58,333 — the pool has now dropped below the 60k-population band, roughly the midpoint of the original 1,229-unit / 34.9M pool).

For next run: continue phase-2a from Wesel down, same method (2 synchronous subagents, 30 cities each, lib.mjs::parseCsv on gemeindeverband-queue.csv — never a naive split — cross-referenced with gemeindeverband-members.csv for AGS, dedupe-by-domain-and-name against roster.jsonl before minting a company_id). At the current ~40-60/run cadence and a shrinking-but-still-large 1,129-unit pool, expect several more runs before the pool empties or drops below the 10,000-population floor — stop there and report per SKILL.md's explicit phase-2b boundary, do not descend into it automatically.