de:wasserversorger-enrich

Run complete. Enriched 45 more water suppliers (2289 → 2334 distinct, enrichment.jsonl 2294 → 2339 lines), all verified as genuine active suppliers with clean shard merges and zero collisions; population coverage holds at 99.99% (24 residual municipalities, 11,023 people). Found and flagged 5 new standing items for a human (a wrong official_domain for Stadt Leun, two self-supply-vs-Zweckverband tensions at Wemding/Kleinwallstadt, a stale null website for Wolfach, and a likely resolution of the long-open Vilgertshofen/Fuchstal boundary question); also noted a process gap — this run skipped writing separate Part-3 municipality-link shard files that prior passes produced, logged in the journal with a risk check (spot-verified as low-risk) and a fix for next run.

build_enrich_batch_excl.mjs 45 enrich-20260906-2030: 1106 pending (3375 roster, 2289 enriched, 6 excluded), 0 pending_domain (unchanged — still fully cleared). Population-sorted batch of 45 (5540-5840 range) split 23/22. Ran 2 parallel foreground Agent calls (run_in_background: false), each told not to spawn sub-agents. Both ran synchronously (~19 min / ~12 min wall-clock).

Verification before merge: shard A wrote 23/23, shard B 22/22, both 0 bad JSON, exact match against the two batch files (0 missing, 0 extra), 0 collisions with the existing 2289 records. Appended into enrichment.jsonl: 2294 → 2339 lines (2289 → 2334 distinct company_ids, +45). map_roster.mjs/consolidate_roster.mjs/coverage.mjs re-run for consistency: roster-links unchanged at 22018 (expected, see gap below), population coverage unchanged 99.99%, same 24 residual municipalities.

Non-suppliers found: 0 — all 45 confirmed genuine active drinking-water suppliers.

Field recovery (shard A, 23): name/address/phone/website_url/supply_area_municipalities 23/23; email 21/23; emergency_phone_number 16 found, 3 earned not_published, 4 not_found; opening hours 16/23; contact_form_url 5/23 genuine forms; analysis_url 11/23, multi_zone 9/23.

Field recovery (shard B, 22): name/phone/website_url 22/22; address/supply_area_municipalities 21/22; email 19/22; opening hours 16/22; emergency_phone_number 11/22; contact_form_url only 2/22; analysis_url 9/22.

Process gap this pass — Part 3 (municipality links) not written as separate shard files. Both subagents recorded supply_area_municipalities only as a free-text summary inside the enrichment record's fields, matching current enrichment.jsonl convention, but did not emit roster-links.part-shardA/B-enrich-20260906-2030.jsonl the way prior passes did (confirmed by ls roster-links.part-*enrich-2026090[3-6]* — every prior enrichment run for weeks has one). My subagent prompts omitted that instruction. Checked the risk: spot-checked wv-de-5518 (Zweckverband Holzland) — shard A's freshly-found 4-member list (Hohenpolding, Kirchberg, Steinkirchen, Inning a.Holz) is an exact match to the 4 rows already in roster-links.jsonl from discovery, and in general this batch's _pop_served was computed from existing roster-links.jsonl, so these 45 companies' memberships were already captured pre-enrichment. map_roster.mjs/coverage.mjs came back unchanged, consistent with no new link data either way. Low risk this time, but the next run should explicitly instruct subagents to also write roster-links.part-shardX-enrich-<tag>.jsonl per SKILL.md Part 3 (verbatim member names, member_kind, quartier) — shard A's output was already clean bracketed-array member lists and could be mechanically converted; shard B mostly wrote prose self_only summaries that would need care to convert without misparsing.

New standing items for a human:

• wv-de-19616 (Stadt Leun): roster's official_domain: leun-hessen.de looks wrong — that host now serves a TLS cert for an unrelated (driver's-license-portal-looking) domain. Real site is leun.de. Most of this record's fields are thin/low-confidence because budget went to the domain investigation; needs a full re-enrichment against the correct domain.
• wv-de-2605 (Wemding): search ties the water-emergency number to Bayerische Rieswasserversorgung Nördlingen (regional bulk supplier), in tension with roster's self_only status — flagged, not reclassified.
• wv-de-4292 (Kleinwallstadt): own Notrufnummern page attributes the emergency line to Zweckverband Main-Mömling-Elsava (AMME) — same tension as Wemding, flagged not reclassified.
• wv-de-10515 (Wolfach): roster's website_url: null was stale; wolfach.de resolves fine with a full Impressum/Kontakt site — corrected in this record.
• wv-de-1514 (Vilgertshofen), the long-standing Fuchstal/Vilgertshofen boundary question: own site + a 2018 newsletter (own domain) describe an extraterritorial spring in Fuchstal territory used by Vilgertshofen since 1929, with Fuchstal water only as an emergency backup — reads as resolved (extraterritorial self-supply, not Fuchstal-supplies-Vilgertshofen), but the source PDF itself wasn't read verbatim this pass; recommend downgrading from "open boundary question" to "resolved pending verbatim PDF confirmation" rather than closing outright.

New host/tooling traps: wvv-gmund.de (wv-de-19615) serves a cert for *.your-server.de (same class as the documented bodensee-wasserversorgung.de cert issue). huglfing.de (wv-de-5019): "too many redirects" on root and sub-pages — new trap. braeunlingen.de: one internal page has a malformed 301 (Location header drops www. and the trailing slash) — a server misconfig, not a WAF. fernwasser-franken.de deliberately not fetched this pass, per the documented shared-IP ban with reckenberg-gruppe.de. Shard B again hit the documented "PDF has no text layer" gap (Tacherting, Aying, Ebensfeld, Simssee, Dettenhausen/ASG, Mengerskirchen, Vilgertshofen 2018 newsletter) — no pdftotext/poppler in this session's container; shard A worked around it for some files using the campaign's local .pdftext.mjs (handles a +29 codepoint-shifted embedded font seen before).

Aside: SKILL.md's "Finish" section still names map_ags.mjs/export_csv.mjs, which only ever cover the frozen 25-company pilot (companies.jsonl/links.jsonl) — ran them out of habit this pass and got pilot-only numbers back, harmless but not meaningful. The real current pipeline is map_roster.mjs → consolidate_roster.mjs → coverage.mjs, as every recent journal entry actually uses. Worth fixing the doc.

No merge/export scripts run beyond the above. Nothing sent, no forms submitted, no captchas touched.

For next run: continue with build_enrich_batch_excl.mjs 45 <real batch tag>. Still 0 pending_domain, 1061 other pending (3375 roster total, 2334 distinct company_ids enriched, 6 excluded). Explicitly instruct subagents to also emit Part-3 roster-links.part-shardX-enrich-<tag>.jsonl files this time. All prior standing items carried in every entry back through 2026-08-12 remain open, plus the five new items above (Leun domain correction, Wemding/Kleinwallstadt emergency-number tensions, Wolfach website_url fix, Vilgertshofen downgrade-not-close).