R04 Verification D — Gemini correction round (rule 9: a correction is a claim until verified)
Date: 2026-09-06 · Inputs: Gemini’s pasted correction response (record of correction) and the Drive export gemini-2026-09-06-corrected.md (landed in raw-inputs/; 176,974 chars — 215 chars longer than the original). Method: Crossref for every corrected tuple; PMC full text for Büser 2025; Semantic Scholar abstract for Wang AE 2019; string scan of the “corrected” export for the failed identifiers and withdrawn magnitudes.
D1. The export is not the correction
The Drive document Gemini calls “corrected” still contains: the Zipper tuple with its dead DOI (line 1352); the “Rehman 2021” sugarcane DOI (line 84); “22.8 kg” in Candidate Claim 2 (line 1250); the fabricated Vandecasteele title (line 1345); the 2,500–3,200-acre / 0.013% denominator (4 lines); “30% to 50%” venting (3 lines); “6.1 L/m²-day” (2 lines); “11%” stations (3 lines); M9 still “Falsified.” Only the Samburova prefix and the npj DOI were fixed in the body. Disposition: as in R03, the pasted response is the correction of record; the export is preserved as-is and is not a citable input. The ledger takes only editor-verified tuples and magnitudes.
D2. Corrected tuples — re-verified
| # | Gemini’s corrected tuple | Crossref object | Verdict |
|---|---|---|---|
| 1 | Wang, Wiedinmyer, Ashworth, Harley, Ortega, Vizuete 2019, Atmos Environ 199:80–87, 10.1016/j.atmosenv.2018.10.049 | Exact | ACCEPT |
| 2 | Cuypers, Vanhove, Bonneure, Tytgat 2017, FSI 277:59–65, 10.1016/j.forsciint.2017.05.016 | Cuypers E, Vanhove W, Gotink J, Bonneure A, Van Damme P, Tytgat J | PARTIAL — and the editor’s own §2 tuple was equally incomplete (four of six authors). Both corrected; S-017 now carries the full list. Logged in critique §5 as an editor error (rule 13). |
| 3 | Vandecasteele, Claerbout, Denaeghel D., Craeye C. 2024, “The repeatability of reusing peat as horticultural substrate and the role of fertigation in nitrogen immobilization in three consecutive strawberry crops,” WM 190:296–305 | Title: “…and the role of fertigation for optimal reuse”; authors Denaeghel Hanne, Craeye Simon | FAIL — fabricated subtitle and wrong initials inside the correction round. The DOI, volume and pages are right; the tuple was completed from memory a second time. |
| 4 | Mills 2025, npj Sustainable Agriculture 3:52, 10.1038/s44264-025-00091-z | Exact | ACCEPT |
| 5 | Samburova et al. 2019, JAWMA 69(11):1267–1276, 10.1080/10962247.2019.1654038 | Exact | ACCEPT |
| 6 | “Rehman et al. 2021” — withdrawn; planting-density statement kept “as standard agronomic knowledge” | — | ACCEPT withdrawal; the density figures (>500,000 vs 1,000–5,000 plants/ha) are now unsourced and do not enter the ledger. |
| 7 | “Zipper et al. 2019” — withdrawn; groundwater-lag concept kept | — | ACCEPT withdrawal; concept enters only via Dillis 2026’s own stated limitation. |
| 8 | Dillis, Grantham, McIntee, McFadin N., Grady 2019, Cal Ag 73(3–4):146–153; n=901 | McFadin Bryan | PARTIAL (initial wrong; ChatGPT’s “McFadin, B.” is right). n=901 confirmed. |
| 9 | Massuela, Hartung, Munz, Erpenbach, Graeff-Hönninger 2023, “Cannabis Hunger Games: nutrient stress induction in flowering stage can enhance cannabinoid concentration and alter the terpene profile in inflorescences of Cannabis sativa L.” | Authors: Massuela, Munz, Hartung, Nkebiwe, Graeff-Hönninger; title continues “– impact of organic and mineral fertilizer levels on biomass, cannabidiol (CBD) yield and nutrient use efficiency” | FAIL — fabricated subtitle and a wrong author (Erpenbach for Nkebiwe) inside the correction round. S-099 is unchanged. |
Net: 5 accepted, 2 partial, 2 new fabrications — the same “right DOI, invented tuple” mode as R03’s correction round (Chouvy “Hashish clubs”; Polson chimera). Sixth consecutive cycle.
D3. Magnitude dispositions — re-verified
| Item | Gemini’s disposition | Check | Verdict |
|---|---|---|---|
| 22.8 kg CO₂e/kg → 61.8 / 110.7; 60.7 → 52.7 per 100 g THC; M2 recomputed 0.15–0.27 mi | Matches PMC full text; arithmetic 0.0618/0.404 = 0.153, 0.1107/0.404 = 0.274 | ACCEPT as modeled inference (grid year/boundary must be stated) | |
| 77/165/388 kg N/ha; 2.8 L/plant-day; 380 L/kg — withdrawn | — | ACCEPT | |
| 5.5 / 3.8 L/m²-day — disclosed as Gemini’s own division of Dillis 2020 monthly volumes by median canopy; withdrawn as verbatim | The disclosure resolves the provenance question: researcher arithmetic on figure data, not a reported statistic | ACCEPT withdrawal; not ledgerable | |
| Büser 2025: 231.23 → 188.19 L/m² (−18.6%), weeds −93.2%, yield +5%, CBD +9% (PMC12247431) | PMC Results: “SDI resulted in an 18.6% reduction in water consumption from 231.23 L m−2 for DI and 188.19 L m−2 for SDI… total irrigation amount of 9971.6 L and 8116.08 L” | VERIFIED verbatim — closes ChatGPT Q6’s partial as well | |
| Wilson 2019 rates: 0.22 gal/ft²-day (Aug), 0.17 (Sept) outdoor; 0.18–0.22 indoor | Article body not retrievable (Cal Ag site returns listing only; DOI landing has no text); abstract not available via OpenAlex | UNVERIFIED — new magnitudes introduced in a correction round; not ledgerable until page-checked | |
| Massuela 17.3% | S-099 values | ACCEPT | |
| Condensate 70% | ETCC | ACCEPT | |
| Mills 2012 $6B / 20 TWh | Secondary | ACCEPT | |
| Medel “11%” withdrawn; Wang ACP 0.34/0.67 ppb; Bauer range; CC11, CC20, CC18, 6.1 L, 30–50%, 1.1%, mm/season — all withdrawn | — | ACCEPT | |
| CC14 retained: “basal emission capacities of 0.5 to 2.5 µg C per gram dry leaf per hour” (Wang AE 2019) | Abstract: “Critical Mass produced the highest terpene emission capacity (8.7 μgC g−1 hr−1) and Rockstar Kush the lowest (4.9 μgC g−1 hr−1)”; β-myrcene 18–60%, eucalyptol 17–38%, d-limonene 3–10% of emissions | FAIL — the range is not the paper’s; the paper reports 4.9–8.7. Correct the ledger candidate: EC 4.9–8.7 µgC g⁻¹ h⁻¹ across four strains, “likely lower than achieved in optimized CCFs.” | |
| CC13 restated: “prohibit all conventional synthetic chemical pesticides” | ENF 24-19 criteria are tolerance-exemption + label non-conflict, not “synthetic” | STILL OVERREACH — ledger uses the DPR wording | |
| CC03: “75% to 90%” outdoor shift (Mills 2025 76% national scenario + 2024b ~90% per-kg comparison) | Two boundaries pooled (sector scenario vs per-kg system comparison) | SPLIT at ledger: R05-C003 (~90%, Canadian per-kg model) and R05-C010 (−76%, US sector scenario) already exist separately | |
| CC10: “1.5× more forest edge per unit area… mean 0.16 ha” | Abstract qualitative; press coverage: 1.5× forest loss and 2.5× fragmentation per unit area; 0.16 ha not located | PARTIAL — ledger states the per-unit-area finding qualitatively with the secondary multipliers flagged as not full-text-verified | |
| CC09: forbearance April 1–October 31 | Dillis 2020 abstract: “state cannabis regulations require forbearance from surface water diversions from April through October”; SWRCB policy page confirms the policy exists (dates not on the landing page) | PASS WITH NOTE (cite Dillis 2020 + Cannabis Policy Attachment A) | |
| M4/M5/M6/M8 recalibrated to “unsupported / not established” | Matches the brief | ACCEPT (M9 in the export still reads “Falsified”; the ledger ignores the export) |
D4. Editor’s own error (rule 13)
| Item | Original (critique §2, item 2; SOURCE_LIBRARY do-not-cite row) | Correction | Basis | Second check | Status |
|---|---|---|---|---|---|
| Cuypers 2017 author list | “Cuypers, E.; Vanhove, W.; Bonneure, A.-J.; Tytgat, J.” | Cuypers E, Vanhove W, Gotink J, Bonneure A, Van Damme P, Tytgat J | Crossref record for 10.1016/j.forsciint.2017.05.016 (six authors) | Gemini’s independent correction reproduced the same four-name list — i.e. it copied the editor’s tuple; the Crossref record is the second check | Applied to critique §2, do-not-cite row, and S-017 |
Root cause: the editor transcribed a partial author list from the Crossref bibliographic-search summary (which prints two names) plus memory, rather than from the full record. Process fix: corrected tuples issued by the editor are pulled from the full Crossref author array, never composed.