MED01-P2 — Correction-round prompts (2026-10-06)

Paste each into the same chat the researcher used for its P2 passes. For Gemini that is the rerun chat (second account), not the superseded first chat. Both passes are frozen; each correction is filed beside them as …-P2-<model>-correction-1.md.

Corrections inherit full verification (base Rule 9): every returned identifier and figure is re-resolved by the verifier before anything changes in the ledger. Neither prompt reveals the other researcher’s findings or the verifier’s completeness-probe items. Probe items enter the ledger as verifier-found and are audited in §M10a, not sent to the researchers.

To ChatGPT

Correction round for your MED01-P2 Stage 1 and Stage 2 submissions. Both are frozen as received; reply as a separate, numbered correction note. Do not revise or restate the rest of the passes.

Verification run 1 found every PMID and DOI you gave resolving to the paper you named, all four NCT registrations resolving to the described trials, and the Sativex DIN confirmed. The numerical claims checked at abstract level (CAMS, Ware 2010, Abrams 2007, Kerbrat 2016, Naftali 2013, Geffrey 2015, Tramèr 2001) matched. Three small items:

1. ISRCTN68314063 (Ware et al. 2010). You flagged it as publication-reported and not opened. If you can open the ISRCTN registry record now, give its registered title, registration date, and registered primary outcome, copied from the record. If you cannot open it, say so; do not reconstruct it.

2. NCT00046722 (Abrams 2007). You wrote that the registry record "shows experimentally induced pain as an outcome." State which outcome field that was (primary or secondary) and copy the registered primary outcome wording.

3. CAND-03 (CAMS). Your candidate text reads "cannabis extract Δ −0.32? Precise direction represented above as reported contrast 0.32 …". Restate the comparative result in one clean sentence, using the abstract's own wording for the direction of the estimated difference.

Format: item number · resolution · identifier(s) · "verified live" or "not resolved". Nothing else.

To Gemini

Correction round for your MED01-P2 Stage 1 and Stage 2 submissions in this chat. Both are frozen as received; reply as a separate, numbered correction note. Do not revise or restate the rest of the passes. Do not consult or cite any other researcher.

Verification resolved every identifier live. Many of your papers are real and relevant, but a large share of the identifiers attached to them point to unrelated papers or do not exist, although each was marked "Verified live (resolved title copied)". Several figures also differ from the cited abstracts.

RULES FOR THIS REPLY
- For every identifier you give, you must have opened the PubMed page, doi.org link or registry record in this session. Copy the title exactly as that page displays it.
- If you cannot open the page, write "not resolved". Do NOT supply an identifier from memory or from a search snippet. A wrong identifier is worse than none, and a wrong identifier in a correction round counts as a new citation failure.
- If you cannot support a figure from the source, withdraw it.

A. PMIDs that resolve to unrelated papers or do not exist. For each item, give the correct PMID and DOI for the paper you intended, with the copied title, or write "not resolved".
1. Alger 2002 — you gave 12445899 → resolves to an astrocyte/thalamus paper.
2. Cravatt 2001 — 11752458 → viperin/CMV paper.
3. Huggins 2012 — 22442435 → vasospasm/epilepsy-surgery paper.
4. Collin 2007 — 17355545 → riluzole/ALS paper.
5. Portenoy 2012 — 22442249 → primate testis cell paper.
6. Zammit 2002 — 12446536 → adolescent self-harm survey.
7. Rogeberg 2013 — 23319617 → Holocene gene-flow paper.
8. Budney 2004 — 15516668 → allergic rhinitis paper.
9. Allen 2004 — 15707076 → "SARS: the flying death".
10. Koppel 2014 — 24777948 → soil/plant health paper.
11. Kerbrat 2016 — 27806232 → US health-reform paper.
12. Nutt 2013 — 23764895 → bile-acid transport paper.
13. Grant 2012 — 22615655 → genus Achillea review.
14. Rocha 2014 — 25215638 → cervical carcinoma paper.
15. Geffrey 2015 — 26105080 → no such PMID.
16. Rog 2005 — 16186535 → Charcot-Marie-Tooth quality-of-life paper.
17. Van Gaal 2005 — 15837488 → Langmuir-Blodgett film paper.
18. Pi-Sunyer 2006 — 16474120 → HSV packaging-protein paper.
19. Porter & Jacobson 2013 — 24238796 → resting-state fMRI paper.
20. Englund 2013 — 23108272 → mathematics/white-matter paper.
21. Stith & Vigil 2016 — 27330386 → retinopathy-of-prematurity paper.
22. Tashkin 2013 — 23802294 → "Fathers know best".
23. Tzadok 2016 — 26801975 → Ebola-vaccine news item.
24. Jetly 2015 — 25467990 → postmastectomy seroma paper.
25. Singh & Bali 2013 — 24490074 → polyunsaturated fatty-acid review.
26. Skrabek 2008 — 18266804 → EEG synchronization paper.
27. van den Elsen 2015 — 25969098 → laser-optics paper.

B. DOIs that do not resolve, or resolve to a different paper. Correct each with a copied title, or write "not resolved".
28. Collin 2007 10.1177/1352458506070613 · 29. Cravatt 10.1073/pnas.231193898 · 30. Grant 10.2174/187152712800564491 · 31. Henquet 10.1136/bmj.38267.431400.7C · 32. Rogeberg 10.1073/pnas.1220329110 · 33. Wade 2004 10.1177/135245850401000408 · 34. Singh & Bali 10.12659/PCR.889317 · 35. Geffrey 10.1111/epi.13076 (resolves to an RBFOX1-deletion paper) · 36. Huggins 10.1016/j.pain.2012.02.030 (resolves to an rTMS paper) · 37. Rocha 10.1590/0100-3984.2013.1909 (resolves to a peritonitis case report).

C. Other identifiers and attributions
38. Sativex Health Canada DINs 02269926 and 02268571 do not appear in the Health Canada Drug Product Database. Give the DIN as displayed in that database, or "not resolved".
39. "Chaiton et al., 2022" for DOI 10.1186/s42238-022-00133-0, and "Sweileh et al., 2018" for DOI 10.3390/publications6040040: state the first author of each as shown on the publisher or Crossref record.
40. Titles marked "resolved title copied" that differ from the PubMed title: Wilson & Nicoll 2001, Huggins 2012, Portenoy 2012, Englund 2013, Porter & Jacobson 2013. Copy each title exactly from the PubMed page.
41. Every reference is marked "Full text". For each Stage 1 candidate claim's supporting source, state the access level you actually had: full text, abstract only, or secondary report.

D. Canonical counts
42. You stated that requests to NCBI E-utilities were blocked in your environment, yet reported four counts (5,618 / 7,542 / 9,874 / 8,365). A live run of the exact canonical URLs does not reproduce them. State how each figure was obtained (tool, page, date). If they were not obtained from a live run of the exact canonical query, withdraw them.

E. Figures and characterisations that differ from the cited sources. For each, give the location in the source (abstract, or full-text section/table) where it appears, or withdraw it.
43. CAMS (Zajicek 2003): THC vs placebo "CI −0.36 to 2.24, p = 0.14"; funding "UK Medical Research Council"; "up to 78% … correctly identified their treatment group"; capsule content "THC 2.5 mg + CBD 1.25 mg".
44. Ware 2010: "difference −0.7; 95% CI −1.37 to −0.03; p = 0.023"; and that cannabis was given three times daily "for 14-day periods".
45. Abrams 2007: described as a "crossover" trial with between-group "p = 0.016".
46. Portenoy 2012 (CAND-23): state what the study's primary endpoint was and its result.
47. Whiting 2015: MS spasticity "WMD −0.76 (−1.38 to −0.14; 3 trials), though objective spasticity scales showed no statistically significant effect"; AE OR 3.03 and withdrawal OR 2.94.
48. Koppel et al. 2014: you describe its scope as "complementary and alternative medicine in multiple sclerosis" with "Level A … established as ineffective" ratings for the Ashworth scale. Confirm the title and scope of the source those ratings come from.
49. Kerbrat 2016 / BIA 10-2474: "five healthy volunteers … cerebral hemorrhages and necrosis" (Stage 1) versus "one death and four hospitalizations" (CAND-14). Give the participant counts as reported in the source you cite.
50. Naftali 2013: "without endoscopic mucosal healing". Identify which study and year reported endoscopic outcomes.
51. Cesamet (nabilone): "FDA approved … May 15, 2006". Identify the FDA record and state what type of action it was.
52. Pletcher 2012: "N = 5,016".
53. Novotna 2011: adjusted difference "−0.84 (−1.29 to −0.40)".
54. Meier 2012: "average decline of 8 IQ points".
55. Geffrey 2015: interaction "driven by CBD's inhibition of cytochrome P450 CYP2C19".

Format: item number · resolution · identifier(s) with copied title · "verified live" or "not resolved" or "withdrawn". Nothing else.