Verification & Critique — R07: Soil vs. Coco vs. Hydroponic Systems

A practitioner-record passage is withheld from the public copy.

1. Scorecards

ChatGPT: 31/31 identity — the R06 “Cockson” misattribution did not recur, so the record is back to zero failures (six of the last seven cycles). The 13 self-flagged partials were exactly what the protocol asks for: missing authors or DOIs declared, not invented, and every one resolved to the described object (R4 Ortiz-Delvasto 2023/24; R11 Allred, Fatzinger & Bugbee 2025; R13 Crispim Massuela et al. 2026; R14 Büser et al. 2025; R23 Weingarten, Mattson & Grab 2024; R25 Punja, Scott & Lung 2022 CJPP 44(1); R29 Kalkušová et al. 2025). One claim-support partial (CAND-09) is a cultivar/outcome swap, not a number error. Its four absence claims are bounded to the documented search as required.

Gemini: the substance is the best it has produced — the A1 taxonomy, the substrate-physics section, the H24 matrix scaffold and the M-verdict structure are genuinely useful — and the bibliographic layer failed again: 2 wrong-object DOIs (one on a fabricated paper), 3 real papers with altered tuples, 2 corpus IDs attached to the wrong sources, and not one of its 16 candidate claims fully supported by its own citation. The new failure mode this cycle is invented magnitudes: kg of rockwool per plant, grams of N per day, lab-failure percentages, cycles per year and yield-per-m² ranges presented in tables with the confidence of measurements and no source at all. Danny’s observation that the output “looks like an upgraded model” is recorded as unverified; whatever the model, the prose improved and the citation discipline did not.

2. Cross-examination — where the two passes agree and disagree

2a. Independent convergence (load-bearing; both verified)

1. No sensory evidence on root-zone system. Both searched and both report zero trained-panel (class b) and zero blinded consumer (class c) studies comparing soil-grown with soilless or water-culture flower (CAND-15; R07-C013). ChatGPT’s wording is bounded (“within the documented search”); Gemini’s “zero published … globally” is over-strong and will be re-worded. Both also find the archive never made the “soil tastes better” claim (archive-audit). Ledger: absence, established — the strongest R07 absence.
2. Root pathogens colonize coir and stone wool. ChatGPT anchors on McGehee & Raudales 2021 (R9 = S-116; Koch’s postulates on hemp seedlings, Connecticut) and Punja & Rodriguez 2018 (R10); Gemini on Punja & Rodriguez 2018 and Punja, Scott & Lung (2021/2022). Two distinct labs on the coir/rockwool finding (rule 12 satisfied); the Punja 2018 recirculation-spreads-inoculum observation is shared. Both reject M9’s premise. Ledger: extends R08-C017 with a second independent survey.
3. “Hydroponics uses 90% less water” is borrowed. Both trace the figure to leafy-greens greenhouse-vs-field comparisons and find no cannabis measurement supporting it as a system-wide statistic (M5). ChatGPT adds two cannabis irrigation studies (Massuela 2026 deficit irrigation, R13; Büser 2025 subsurface drip −18.6%, R14) that change management, not medium. Ledger: M5 unsupported; two measured irrigation-management rows from ChatGPT.
4. Coir is not one material. ChatGPT (CAND-17, from substrate science R19–R21: Carlile, Abad, Verhagen) and Gemini (A2, from Nemati 2021 and Malík & Tlustoš 2025) both establish source- and grade-dependent salinity, CEC and physical behavior from distinct sources. They diverge on the prescriptive rule (2b-5). Ledger: general/adjacent, established.
5. No multi-cycle living-soil vs soilless comparison (ChatGPT gap 8; R07-C014). Inherited from R08-C030 rather than independent, but both re-searched and neither found one. Ledger: R08-C030 re-confirmed in a second search.
6. Delivery architecture is an independent variable — both build on Velechovský 2024 (S-205; same paper, so not independent). ChatGPT’s reading verified in full text: no conclusive effect on inflorescence mass, significantly higher total THC yield under recirculation (CAND-24 = R06-C015 confirmed). Gemini’s numbers are wrong (2b-2). Ledger: R06-C015 stands; R07 row extends it with the corrected comparators.

2b. Divergence (content, not friction — recorded)

1. Yep, Gale & Zheng 2020 — the same paper described as two different experiments. ChatGPT: three root-zone bundles — 11 L peat with synthetic fertigation, 11 L peat with aquaculture effluent, and 3 L peat pots over a DWC aquaponic sump — with the synthetic-peat arm yielding 42–116% more inflorescence; therefore the study cannot support “hydro > soil” because its “hydroponic” arm is a peat container (CAND-01/02). Gemini: “60 L DWC tubs,” pure water culture vs aquaponic solution, cultivars ‘Zenith’ and ‘G1’, water 21.5 ± 0.8 °C, K/Fe supplementation closing the gap (R07-C008). Verifier-A pulled the paper: cultivars were ‘Nordle’ and ‘Sensi Star’, containers were 11 L peat plus a 3 L-pot aquaponic arm, yields differed 42–116%, and there is no 60 L tub, no supplementation arm, no 21.5 °C. ChatGPT is right; Gemini’s C008 is a claim-support NO and its descriptor is fabricated. ChatGPT also dropped nothing: Gemini’s tuple omits Gale. Ledger verdict: CAND-01/02 as written; the Yep study enters as the corpus’s standing example of a system-bundle comparison.
2. Velechovský 2024 magnitudes. Gemini: “+182% total THC in recirculating vs +50.7% in drain-to-waste vs baseline,” P/K/Fe “+93/+43/+940%,” “0.5 m² per plant.” Full text (verify-A): the 182% is recirculating-vs-DTW, not vs control; 50.7% is enriched-vs-control within DTW at week 11 only; augmentation was +83/+39/+860%; density 27.5 plants/m²; fertigation stopped at week 9; substrate temperature not measured. Author list invented. Ledger verdict: use ChatGPT’s reading and the verified full-text figures; Gemini C004 rewritten before use.
3. Rockwool waste magnitude. ChatGPT: no mass-flow inventory exists; M6 stays a direction, not a magnitude (CAND-18, absence, extends R04-C033). Gemini: “0.30–0.50 kg per plant, >90% to landfill” as a modeled claim citing Malík & Tlustoš 2025 — which contains no such figures — while its own data-gap 3 says no mass-balance audit exists. Ledger verdict: absence stands; no magnitude.
4. Coir chemistry — mechanism vs rule. ChatGPT (C7): coir CEC and cation exchange are real and source-dependent; explicitly rejects “coco universally locks out calcium.” Gemini (C005): CEC “40–100 cmol_c/kg,” K/Na-saturated, “mandatory calcium buffering.” The mechanism is agreed (2a-4); the number is in neither cited review at that range, and the prescriptive rule is exactly what the brief’s A1 told researchers to characterize, not stipulate. Ledger verdict: mechanism as general; no numeric CEC; no mandatory-buffering rule.
5. Verdict temperature on M4 / M7 (forgiving soil, unforgiving hydro). ChatGPT: plausible practitioner hypothesis, no cannabis dataset on error tolerance or time-to-symptom (CAND-20 absence). Gemini: “Supported on time-to-symptom; inverted on time-to-remediate,” with a CAPEX/OPEX/failure table giving hours-to-failure by system — uncited. The editor sides with ChatGPT on grade and keeps Gemini’s framing (time-to-symptom vs time-to-remediate are different axes) as a synthesis structure, unsourced. Ledger verdict: M4/M7 unverified; the two-axis framing goes to the system-property matrix as unmeasured.
6. Genotype × substrate. ChatGPT ledgers Burgel 2020 (two genotypes, coir penalty in one) and Allred 2025 (two cultivars, stone wool) as measured interaction evidence (CAND-03, -08, -09). Gemini’s data-gap 5 states multi-genotype trials “are absent.” Ledger verdict: interaction is measured (ChatGPT); Gemini’s absence statement is wrong within its own search.

A practitioner-record passage is withheld from the public copy.

2c. Things one researcher found that the other missed (both verified)

• ChatGPT only: Burgel et al. 2020 (genotype × substrate, R2); Schober et al. 2023 by correct authorship (R3); Ortiz-Delvasto et al. 2023/24 (five substrates, water availability as the correlate, R4); Caplan et al. 2019 container-capacity coir (R5); Allred, Fatzinger & Bugbee 2025 (stone-wool osmotic steering, R11); Crispim Massuela et al. 2026 (deficit irrigation, R13); Büser et al. 2025 (subsurface drip, R14); McGehee & Raudales 2021 by name (R9 = S-116); Weingarten, Mattson & Grab 2024 (R23); Vergara & Ruterbories 2026 NY greenhouse economics (R18); Kalkušová et al. 2025 (R29); the substrate-science trio Carlile / Abad / Verhagen (R19–R21); the three growing-media LCAs (R15–R17); NIOSH/OSHA fibre limits and CDPH OCal (R27, R30/R31).
• Gemini only: Malík & Tlustoš 2025 Agriculture 15(18):1955 soilless-media review (real, PASS); Balliu et al. 2021 Horticulturae root-restriction review (real, PASS); Tebdoie et al. 2025 IJAT peat-vs-coir vegetative trial (real; Gemini’s reading inverted — the coir + vermicompost mix had the highest chlorophyll and inflorescence yield was measured; real DOI 10.63369/ijat.2025.21.4.1521-1534); Yep, Gale & Zheng 2020a NaCl-stress paper (real; Gemini’s salinity finding inverted); the PG&E/ETCC condensate report as a named source (= S-007; Gemini’s “70–80%” is not in it — R04-C018 says ~70%); DCC heavy-metal limits (4 CCR §15723, correct).

3. ChatGPT — items requiring correction (rule 9: corrections are claims until verified)

Minor tuple gaps (not failures): R17 registries index “Alejandre EM” not “Migoni Alejandre” (wording note); CAND-04’s Schober abstract obtained via mirror; six claims rest on abstracts behind ScienceDirect/ASHS paywalls and carry that flag in the ledger.

4. Gemini — items requiring correction (correction round requested; every replacement tuple will be re-verified)

Identity (must supply resolving tuples or mark as partial):

1. “Caplan, Stemeroff, Dixon & Zheng (2019). Vegetative-stage substrate composition effects on cannabis. HortTechnology 29(4):469–477. 10.21273/HORTTECH04332-19” — the DOI resolves to Shi et al. 2019, anaerobic soil disinfestation in tomato, HortTechnology 29(6):777–787; no paper with this title or author set exists. Withdraw, or cite the real Caplan, Dixon & Zheng 2019 Acta Horticulturae 1266:55–62 if that is what was meant (it has no Stemeroff and is not a substrate-composition trial). Every use of it (A3, B3, C1, H24 row 4, H27 row 1) falls with it.
2. “Knight et al. (2010) 10.1016/j.forsciint.2010.04.015” — resolves to Krishan, Kanchan & DiMaggio 2010 (limb asymmetry). Real Knight et al. 2010 is 10.1016/j.forsciint.2010.04.022, Forensic Science International 202:36–44. Supply the corrected tuple and re-state what it supports (B3 row: “350–500 g/m² per cycle” — is that in the paper?).
3. Velechovský et al. 2024 — authors are Velechovský, Malík, Šenkyřík & Tlustoš; title as registered under S-205. “Kyselka, Toufarová, Bjelková” do not exist on this paper.
4. Punja et al. 2021 — real object is Punja, Scott & Lung, Canadian Journal of Plant Pathology 44(1):66–81 (2022 issue; 2021 online); “43(sup2):S116–S129” is invented. Also: the corpus ID S-116 is McGehee & Raudales 2021, not this paper; remove the S-116 tag.
5. “Ahmed et al. (2023). Growth dynamics and yield formation… IC&P 203:117172” — the paper is Schober, Präger, Hartung, Hensmann & Graeff-Hönninger 2023; and S-096 is a different Ahmed 2023 (Front Plant Sci 14:1219836). Correct the authorship and remove the S-096 tag.
6. “Vergara et al. (2016) 10.3389/fpls.2016.00019” (cited in D3 for terpene-genotype) — unresolved; supply a resolving tuple or withdraw.
7. “Zheng (2021)” in H27 row 4 carries Nemati’s DOI — a review by Nemati et al. is not “Zheng 2021”; correct or withdraw.
8. Tebdoie et al. 2025 — supply the DOI 10.63369/ijat.2025.21.4.1521-1534 and restate the finding: the abstract reports the coir + vermicompost mix with the highest chlorophyll and reports inflorescence yield; your text says peat won and no yield was measured.
9. Yep, Gale & Zheng 2020a (fpls.2020.01169) — your salinity conclusion is inverted relative to the abstract; restate or withdraw.
10. All DOIs wrapped in google.com/search?q= links — supply bare https://doi.org/ identifiers.

Claim support (must supply a passage or withdraw): 11. R07-C008 (Yep 2020b): “60 L DWC tubs,” ‘Zenith’/‘G1’, 21.5 ± 0.8 °C, K/Fe supplementation closing the gap — none of this is in the paper (cultivars ‘Nordle’/‘Sensi Star’; 11 L peat containers plus a 3 L-pot DWC aquaponic arm; hydroponic arm yielded 42–116% more). Withdraw the claim as written. 12. R07-C004 (Velechovský): 182% is recirculating-vs-DTW, not vs control; 50.7% is enriched-vs-control within DTW at week 11; augmentation was +83% P / +39% K / +860% Fe; 27.5 plants/m², not 0.5 m²/plant; fertigation stopped week 9; root-zone temperature not measured. Restate with the paper’s comparators. 13. R07-C006 (Caplan 2017a): the abstract reports no substrate difference in growth or yield and a pooled optimum of 389 mg N/L; rates were 117–585 mg/L, not 57–389. Withdraw the “substrate-dependent optimum” wording for the vegetative study. 14. R07-C007 (Caplan 2017b): “+110%” and “10.3 → 8.7% THCA” are not in the abstract; two substrates, not one; cultivar ‘Wappa’. Supply the passage or mark abstract-level. 15. R07-C002: neither Velechovský nor Yep measured flowering duration across systems. Withdraw as measured; if retained, it is general plant physiology with a different source. 16. R07-C001: “7–14 days faster canopy fill” is not a figure in Schober 2023. Supply the passage or drop the number. 17. R07-C010: “0.30–0.50 kg wet rockwool per plant; >90% to landfill” — not in Malík & Tlustoš 2025. Withdraw or source. 18. R07-C011: “45–125 g N and 15–40 g P per day per 1,000 plants; 150–250 mg/L NO₃-N in runoff” — cited to a Water Board order that contains no such measurements. Withdraw or source. 19. R07-C015: “30–50% TYMC failure rate for living-soil crops” — Punja, Scott & Lung is a Pythium etiology paper with nothing on compliance testing. Withdraw or source. 20. R07-C016: “4.8–5.4 vs 3.5–4.2 cycles/yr” — “synthesized”; no source. Withdraw. 21. R07-C005: “CEC 40–100 cmol_c/kg” — not at that range in Nemati 2021 or Malík & Tlustoš 2025. Supply the passage or drop the number; the mechanism survives. 22. R07-C009: the coir/rockwool species list belongs to McGehee & Raudales 2021 (S-116), which you did not cite. Re-anchor. 23. R07-C012: “70–80% transpired … 60–75% withdrawal reduction” — S-007 supports ~70% recovered as condensate and R04-C018 says recoverable ≠ reused; the 60–75% figure is not in the report. Restate as the R04-C018 inheritance. 24. The H25 yield map (g/m²/cycle and g/m²/yr by system), H26 flow map (L/kg, kg salts, kg media), F2 discharge, F5 embodied-carbon table, and G1 economics table carry ranges with no citations. Each row is a claim; supply sources per row or label the tables “illustrative, unsourced” so they are not read as measurements.

Standing note for the correction round (v1.4 rule 9 and the R04/R06 precedent): every replacement tuple and every supplied passage is re-verified (verify-D). A corrected export that re-fabricates inside the round — as R06’s did for McKernan — is kept as a record and not used as a source. Where you cannot resolve a tuple, “partial citation — not fully resolved” passes review; a confident wrong one does not.

5. Editor-caught items (not on either verifier list)

1. Gemini’s export is formatted as the synthesis — “Research Synthesis — R07,” “Report Version: v1.0,” a “Permanent Drive Mirror” line — and its first Drive “export” was a 3 KB outline that described sections it had not written. Caught on retrieval. Raw input is raw input; the file is used as such and the version stamp is ignored.
2. Gemini’s genotype-absence contradicts the record it should have found: Burgel 2020 and Allred 2025 are multi-genotype substrate/steering trials; ChatGPT ledgered both. Caught in the convergence map.
3. Gemini attaches the brief’s own vocabulary to unsourced numbers — “root-zone system descriptor,” “air-temp-only,” “system-orphaned” appear on claims whose numbers are invented. The eleventh field does not launder a magnitude. Caught reading Section K against verify-A.
4. The R06 do-not-cite DOI 10.1080/07060661.2021.1935947 (a Punja tuple Gemini fabricated last cycle) is not the DOI either researcher uses now (…1954695, which resolves to Punja, Scott & Lung). Confirmed distinct; the do-not-cite entry stands.
5. Caplan 2017b container size: the R06 practitioner-era record says 6 L; Gemini says 7.6 L (2 gal); ChatGPT does not state it. Abstract-level; logged for the ledger boundary line as “6–7.6 L (abstract-level; confirm)”.
6. Gemini’s “Section J gap 3” and “R07-C010” contradict each other (no mass-balance audit exists / here is the mass per plant). Caught on read.
7. ChatGPT: its R25 (Punja, Scott & Lung) is dated 2021 in text and resolves to the 2022 print issue — a date convention, not an error; ledger uses “2021 online / 2022 print.”

6. Inherited-corpus checks

• R06-C015 (Velechovský; delivery architecture): confirmed in full text by verify-B — no conclusive effect on inflorescence mass, significantly higher total THC yield under recirculation. Upgraded: the R07 row adds the verified comparators (+83/+39/+860% augmentation; 27.5 plants/m²; fertigation ended week 9; root-zone temperature not measured → air-temp-only).
• R08-C017 (pathogens from coir and rockwool; S-116 McGehee & Raudales): re-confirmed; second independent survey (Punja, Scott & Lung 2021/22) added.
• R08-C030 / R06 keystone absence: re-searched by both; stands.
• R04-C018 (~70% condensate recovery; recoverable ≠ reused): stands; Gemini’s 70–80% and 60–75% are not in S-007.
• R04-C033 (no media mass-flow inventory): stands; ChatGPT’s CAND-18 re-confirms; Gemini’s magnitudes are unsourced.
• R06-C008 / R08-C007 (Caplan 2017a/b optima): consistent with the abstracts — vegetative optimum pooled at 389 mg N/L (2017a), flowering optima 212–261 with substrate-dependent top-rate behavior (2017b). No patch. Gemini’s “substrate-dependent vegetative optimum” misreads 2017a.
• S-096 / S-116 / S-205 identities: SOURCE_LIBRARY entries correct; Gemini’s tags wrong, not the library.

A practitioner-record passage is withheld from the public copy.

8. Correction round — verify-D result (2026-09-10; verification/R07-verify-D-corrections.md)

24 items: ACCEPT 20 · PARTIAL-ACCEPT 3 (items 1, 11, 16) · REJECT 1 (item 10). Nine revised claims: ACCEPT 7 · PARTIAL-ACCEPT 2 (Claims 1, 2) · REJECT 0. Gemini’s submission is filed at raw-inputs/gemini-2026-09-10-corrections.md; its re-uploaded .docx (same Drive id) is a record, not a source.

• Identifiers: all 13 new or corrected DOIs resolved identically on Crossref and OpenAlex — 0 wrong-object, 0 nonexistent, 0 fabricated authors or titles. Caplan, Dixon & Zheng 2019 Acta Horticulturae 1266:55–62 (10.17660/ActaHortic.2019.1266.9) exists with that title; Knight et al. 2010 author list and …04.022 DOI exact; Velechovský “Šenkyřík, J. B.” and PMC10847352 exact; Tebdoie six-author tuple exact; McGehee & Raudales 10.3389/fagro.2021.706138 exact. The identity layer of this round is clean — a first for a Gemini correction round in this corpus.
• Restatements verified: item 12 Velechovský (182% recirculating-vs-DTW under the enriched regime; 50.7% DTW week 11; +83/+39/+860%; fertigation to week 9, demineralized water from week 10, harvests sampled to week 13; 27.5 plants/m²); item 13 Caplan 2017a (117–585 mg N/L; UNIMIX 1-HP vs UNIMIX 1; no substrate difference; 389 pooled); item 14 Caplan 2017b (57–283; ‘Wappa’; U2-HP/U2; 212–261 vs 283; U2-HP +11% yield, +20% THC); item 17 Malík & Tlustoš rockwool disposal qualitative only; item 21 Nemati/Malík coir Na/K exchange and Ca/Mg buffering as mechanism; item 22 McGehee species from coir and rockwool; item 23 S-007 ~70%.
• Rejected: item 10 — the submission states all search-wrapped DOI links were stripped; six remain wrapped (items 1, 2, 5, 8; revised Claims 3, 6). A false completion statement is itself a finding.
• New errors introduced inside the round: Yep 2020b restatement says “β-pinene” — the abstract says pinene with no isomer; “42–116% greater than aquaponic treatments” — the abstract says “than the other rootzone systems” (both non-hydroponic arms); revised Claim 2 gives an “11-week flowering cycle” — the paper ran 7 days veg then sampled the generative phase weekly through week 13; revised Claim 1’s substrate list changed between rounds (peat-perlite / coir-perlite / potting soil → rockwool / peat-perlite / coir) with no passage, and “high-CBD hemp cultivar” and “CBD insensitive” remain unverifiable behind ScienceDirect; several descriptor fields were added without a source (1.5 L pots / Aquathrive / manual DTW on Claim 4; HPLC on Claim 5; automated drip on Claim 6; benches / precision drip on Claim 7).
• Residual strings: every withdrawn number appears only inside its own withdrawal sentence; none survives in the nine revised claims. The R06 failure mode (withdrawn numbers living on in prose) did not recur in the pasted object.

Disposition: PROVISIONAL holds lift on C005 (Balliu — untouched by the round, verify-A basis), C012 (Tebdoie — item 8 now matches the abstract; stays abstract-level), C019 (Yep 2020a — direction now correct), C021 (scaffold — item 24 declares every table range illustrative; cells stay unmeasured), and on the S-224 citation inside C004/C030. Gemini’s revised Claims 3–9 are ledgerable after editorial wording (they duplicate rows already carried from ChatGPT’s verified claims; they add no new row, only a second reading). Revised Claim 2 enters only through C013’s verified full-text comparators (the “11-week” timeline is not used). Revised Claim 1 (Schober 2023) stays held — not ledgerable as measured from Gemini; the corpus row rests on ChatGPT’s verified CAND-04. Do-not-cite additions: “β-pinene” attributed to Yep 2020b; “11-week flowering cycle” attributed to Velechovský 2024; the six wrapped links.

7. Disposition

• ChatGPT: critique approved for ledger; corrections B-09 and B-19 applied at ledger stage; no correction round. Identity record 31/31.
• Gemini: correction round requested on items 1–24. Surviving contributions pending the round: the A1 taxonomy and H24 matrix as scaffolding (general; no unsourced numeric cell is ledgered); the substrate-physics and coir-exchange mechanism (general, no number); Malík & Tlustoš 2025 and Balliu et al. 2021 as verified reviews; Tebdoie 2025 and Yep 2020a as real papers whose findings Gemini inverted (ledgered from the abstracts, not from Gemini); the time-to-symptom vs time-to-remediate framing for the matrix (unmeasured); the absence claims C013/C014 re-worded as bounded. Every Gemini-only row enters the ledger PROVISIONAL until verify-D, per the R06 precedent.
• Danny: one judgment call. With 0 of 16 Gemini claims fully supported and the correction burden this large, the editor still recommends running the round (its two verified reviews and its inverted-but-real papers are worth recovering) but with the R06 rule in force: the corrected export is a record, not a source, until each item is re-verified.
• Next: ledger v0.1 with IDs R07-C001+ (Gemini’s own “R07-C0xx” labels are discarded; the ledger assigns). New SOURCE_LIBRARY registrations from S-219: ChatGPT’s 19 unregistered tuples (R1–R5, R7, R8, R10–R12, R15–R17, R20, R21, R23, R25–R27) plus Malík & Tlustoš 2025, Balliu et al. 2021, Tebdoie et al. 2025 (real DOI) and Yep, Gale & Zheng 2020a if not already registered — ~23 new sources. Reuse map from verify-B (R6→S-101, R9→S-116, R13→S-182, R14→S-180, R18→S-011, R19→S-107, R22→S-205, R24→S-012, R28→S-104, R29→S-185, R30/R31→S-215). Do-not-cite log additions: 10.21273/HORTTECH04332-19 as a Caplan 2019 citation; 10.1016/j.forsciint.2010.04.015 as Knight 2010; “Velechovský, Kyselka, Toufarová, Bjelková” author string; “Punja 2021 CJPP 43(sup2):S116–S129”; “Yep & Zheng 2020b 60 L DWC / Zenith / G1” descriptor.