de:wasserversorger-discover

Journal entry written. Run complete.

Discovery run processed a 60-unit singleton batch: population coverage rose from 99.86% to 99.87% (10,650 → 10,670 municipalities claimed; residual dropped from 293 to 273), roster grew from 3,202 to 3,212 companies. Notable finds: 26 of shard B's units turned out to be uninhabited gemeindefreie Gebiete (zero population, not real gaps), one two-pass-old false negative (Thierschneck) was resolved, and a domain-collision counting discrepancy from prior runs was diagnosed and fixed (canonical count is 30, unchanged by this batch). Full details are in the journal entry dated 2026-09-01.

Continued phase 2b singleton sweep from the 271-unit pool left by the prior 13:00 scheduled run. build_gv_layer.mjs → 281 units with residual members (10 multi-member/22 municipalities, still on hold pending the overdue human decision; 271 singletons). build_batch60.mjs took the top 60 by residual_population. Max in-use company_id was 19866; assigned shard A wv-de-19900–19949, shard B wv-de-19950–19999.

Dispatch: two synchronous general-purpose Agent subagents, both run_in_background: false, both carrying the mandatory no-Agent/no-Task-tool line verbatim. Both completed inline (~1286s / ~464s), neither recursed into background sub-subagents.

Shard A (30 units): 19 resolved, 11 unresolved, 10 new mints (wv-de-19900–19915, sparse — several ids in range unused), 6 links reused existing companies. Self-caught issue: shard initially re-minted 6 duplicates of already-rostered companies (Zweckverband Wismar, Rombachgruppe, ZWA Saalfeld-Rudolstadt, VGW Selters, WVEK/KNE, BEWA/Thüringer Becken) before running the mandatory domain-collision check; caught and fixed before writing final output — the workflow the SKILL.md's last-step recheck exists for, working as intended. Namesake traps avoided: "Zweckverband Wasserversorgung Schwarzachgruppe" (Nürnberg/Schwabach) vs. Perasdorf's actual Niederbayern Schwarzach; "Amt Rantzau" (Kreis Pinneberg) vs. the batch's different Rantzau in Amt Großer Plöner See.

Shard B (30 units): 1 resolved (Thierschneck, reused existing wv-de-1178 ZWE Eisenberg — overturns a two-pass-old false negative that only ever checked JenaWasser), 26 not-applicable, 3 genuinely unresolved after a documented 3rd attempt each (Niederirsen, Göttin, Kollmoor — see their notes in roster-links.part-shardB-20260901-1552.jsonl for the specific candidates ruled out each pass). New finding: 26 of shard B's 30 assigned units were gemeindefreies Gebiet/Gutsbezirk — 22 in Niedersachsen, 4 in Hessen — confirmed population 0 directly from GV100AD's satzart-60 field (uninhabited forest districts, e.g. Nordseeinsel Memmert, Insel Lütje Hörn). Recorded as company_id: null with high-confidence "not a real water-supply gap" notes, same pattern as the earlier gemeindefreies-Gebiet-M-V precedent. Worth flagging for a future run: the singleton queue still contains a real cluster of these zero-population non-gap entries mixed in with genuine residual work — they cost a shard slot each without being fixable, so if they keep resurfacing a one-time filter in build_gv_layer.mjs (exclude satzart-60 zero-population units from the queue entirely) would free up real batch capacity. Not applied this run — flagging only, since it touches a shared script.

Domain-collision discrepancy resolved: the "28 vs 30" mismatch flagged in the last several journal entries was a methodology artifact — the collision-check script variant used at 13:00 fell back to website_url when official_domain was null, inflating the count to 63 when re-run today. Restricting to official_domain only (ignoring rows with no resolved domain, i.e. pending_domain rows) gives the canonical 30, matching every prior run's number exactly, and this batch introduced zero new collisions under that method (the one new-mint collision, amt-leezen.de picking up wv-de-19911, is the documented legitimate shared-Amt-domain pattern, not a duplicate). Use official_domain-only for this check going forward — website_url fallback is not comparable across runs.

Merge: map_roster.mjs: total links 15439 → 15499 (+60, exact). consolidate_roster.mjs: 3202 → 3212 distinct companies (+10, exact). coverage.mjs: municipalities claimed 10650 → 10670 (+20, exactly the 19+1 resolved — no bonus links harvested this run). Population 99.86% → 99.87% (+0.01pp), residual 293 → 273 municipalities / 115.9k → 106.5k people. build_gv_layer.mjs: multi-member pool unchanged at 10 units/22 municipalities (still on hold), singletons 271 → 251 (-20, exact).

For next run: continue singleton batches by residual_population descending — 251 remain, but a meaningful chunk of the pool is the zero-population gemeindefrei/Gutsbezirk cluster described above (not real gaps, just queue noise) — expect batches to keep surfacing some of these until Niedersachsen/Hessen's cluster is exhausted or the filter is added. Standing backlog untouched this run: the 30 domain collisions (now with a resolved counting method, see above), 6 duplicate-mint name pairs, St. Peter/St. Märgen AGS-swap, wv-de-10532 hq_ags mismatch, and the overdue human decision on the 10-unit multi-member pool — all still awaiting a consolidation pass. _tasks/netzbetreiber-contact/german-water/ remains absent from this checkout (5+ runs now).