Skip to content

Curation log — migraine

2026-08-28 — Seed

  • Created: INDEX.md (18 canonical pages), wiki/overview.md (draft, 25 verified PMIDs), OPEN-QUESTIONS.md (7 questions), and this log.
  • Searches run (live PubMed E-utilities esearch/esummary): ICHD-3; GBD burden; aura, trigeminovascular/CGRP biology and genetics/GWAS; acute-treatment reviews and ubrogepant; erenumab, fremanezumab, galcanezumab, eptinezumab and rimegepant trials; PREEMPT; topiramate comparison; CBT; pediatric guidelines; stroke association; menstrual/pregnancy migraine; biomarkers and chronification.
  • Verification: all cited PMIDs were returned live by PubMed; one initially recalled lasmiditan identifier resolved to an unrelated back-pain paper and was excluded.
  • Next: full build queued per CONDITIONS-ROADMAP.md.

2026-08-28 — Full build

  • Built all 18 canonical wiki pages: classification/diagnosis; epidemiology/burden; trigeminovascular/CGRP biology; aura/CSD; acute and preventive treatment; chronic migraine/chronification; medication-overuse headache; neuromodulation/behavioural care; paediatrics; sex hormones/pregnancy/lactation; comorbidity/vascular risk; guidelines; biomarkers; trials; patient experience; and red flags/safety, plus the expanded overview.
  • Density: every wiki page is 150–400 lines. Seventeen pages carry 25–28 unique PMIDs; the trial-landscape page carries 14 PMIDs plus 39 NCT records. All pages remain status: draft pending the separately required audit session.
  • Literature layer: created literature/BIBLIOGRAPHY.md (375 unique papers, cited-by indexed), six structured landmark notes, guidelines/REGISTRY.md, statistics/STATISTICS.md, and the four-file patient-voice/ layer.
  • Research frontier: replaced seed questions with 16 stable tiered questions and a 14-row Dots not yet connected table.
  • PubMed searches run live: ICHD/classification; diagnostic validation, aura/TIA/seizure mimics and imaging; global/GBD and regional epidemiology; disability, productivity, family and care gaps; trigeminovascular anatomy, CGRP/PACAP provocation and sensitization; cortical spreading depolarization and visual snow; acute NSAID/triptan/gepant/ditan/DHE/antiemetic evidence; conventional, toxin, antibody and gepant prevention; chronic-migraine transition/remission and predictors; medication-overuse thresholds, withdrawal, relapse and prevention; device sham trials, CBT, biofeedback, sleep and exercise; paediatric epidemiology, CHAMP, CBT and guidelines; menstrual migraine, pregnancy, postpartum, lactation and exposure cohorts; psychiatric, sleep, epilepsy and vascular comorbidity; guideline bodies worldwide; genetic, CSF, blood, imaging and electrophysiological biomarkers; stigma, lived experience, cognition, interictal burden, work, family and access.
  • ClinicalTrials.gov searches run live: active and completed migraine studies, pivotal acute/preventive programs, paediatric gepants, pregnancy exposure, menstrual migraine, devices, biomarkers, combination therapy, withdrawal and strategy designs; individual study records were queried through API v2.
  • Verification: final live batch audit resolved 375/375 unique PMIDs through PubMed E-utilities and 39/39 unique NCT IDs through ClinicalTrials.gov API v2. Structural checks found one missing page-reference entry and three bibliography omissions; all four were corrected, then the checks were rerun.
  • Patient organizations: seven public sites were verified across the US, UK, Canada, Europe, Ireland and Australia. Headache Australia timed out and was quarantined rather than listed as verified.
  • Boundaries: no claims or identifiers were written from memory; no unresolved citation is marked as verified. No protected condition, convention, tool or frontend file was changed.
  • Follow-up: a separate citation audit must re-check claim-to-paper fit before any page is promoted from draft to curated; guideline currency and trial status are date-sensitive watch items.

2026-08-30 — Full audit

  • Scope completed: audited all 18/18 wiki pages and all 16/16 literature artifacts (34 migraine Markdown artifacts total), including the 388-paper bibliography, six landmark notes, guideline registry, statistics file, patient-voice layer and open-questions register.
  • Live identifier verification: re-fetched 388/388 retained unique PMIDs through PubMed E-utilities and 40/40 unique NCT IDs through ClinicalTrials.gov API v2. Checked 468 page-reference occurrences and 726 in-body PMID occurrences; all retained page PMIDs are present in both the local reference list and master bibliography. No identifier was accepted from memory.
  • Substantive claim errors found and fixed (6): corrected the status-migrainosus review’s first-author attribution; stroke pooled-CI upper bound (1.83, not 1.82); the one-year MOH strategy interpretation (no significant between-group difference, p=0.377; 11% relapse); the limits of topiramate evidence specifically in chronic migraine with MOH; an unsupported neuromodulation tolerability comparison; and misleading candesartan crossover comparator wording.
  • Citation/index integrity defects found and fixed (45): attached 40 valid but previously unconnected page references to the precise claims they support; removed one valid but irrelevant placebo/psychology paper from the vascular-comorbidity page and bibliography; removed three malformed duplicate bibliography bullets; and repaired one incomplete cited by mapping. Final checks show 0 unused page references, 0 body PMIDs missing from references, 0 bibliography/page-set discrepancies and 0 duplicate retained PMIDs.
  • Currency and stale-absence pass: reran 12 live searches behind negative claims covering multi-class sequencing, disease modification, combination prevention, patient-weighted comparison, pregnancy exposure, predictive biomarkers, gepant/MOH liability, vascular outcomes, gepant-versus-lasmiditan comparison, pediatric infusion pathways, aura-primary endpoints and LMIC implementation. Added current evidence where found and converted remaining gaps to positively stated, dated evidence-gap sentences.
  • New evidence incorporated: added 14 live-verified papers, including TEMPLE (atogepant vs topiramate), comparative CGRP-antibody vascular safety, combination-prevention meta-analysis, pediatric IV-DHE and aura cohorts, a negative suPAR predictor study, and seven 2025–2026 guidelines/consensus documents. Added completed NCT05748483; the trial landscape now contains 40 live-checked registry records.
  • Guidelines and patient voice: expanded the master registry to 27 documents and recorded supersession/method limits through 2026. Rechecked seven public organization URLs on 2026-08-30: six returned HTTP 200; the National Headache Foundation returned HTTP 403 anti-bot; Headache Australia remained network-unreachable and is explicitly excluded rather than labelled inactive.
  • Final validation: all wiki pages remain within 150–400 lines and carry 25–32 unique PMIDs, except the trial-landscape page, which carries 15 PMIDs plus 40 NCT records. Relative links: 0 broken. Unverified-marker count: 0. Unresolved substantive issues: 0. Time-sensitive registry statuses and one public-site access limitation remain dated monitoring notes, not unresolved clinical claims.
  • Promotion: all 18 pages passed and were promoted to status: curated with last-curated: 2026-08-30; the condition roadmap was advanced to audited.
  • Protected scope: no git command was run and no .git/.gitignore was created. No protected condition, CONVENTIONS.md, tools/ or frontend/ file was changed.

2026-10-07 — Grounding pass on uncited dot rows (grounder: Claude, Opus 5.5)

  • Scope: the "Dots not yet connected" table in OPEN-QUESTIONS.md only. I started with 14 rows, all without a PMID or NCT ID, so the whole table was in scope. logs/grounding-migraine.md resolved no candidate records for any row, because no row's Powers column named an OQ. Every row was therefore grounded against the wiki pages and live searches. Only OPEN-QUESTIONS.md and this log were written.
  • Outcome: 5 rows rewritten (2, 5, 7, 8, 10) and 9 deleted (1, 3, 4, 6, 9, 11, 12, 13, 14). Survivors keep their original IDs, and the gaps are intentional. Each surviving dot states a figure and a PMID or NCT ID, and each junction states an absence with its record count.
  • Records cited in the table, all fetched live and read in this session: PMIDs 38859744, 41315923, 42327799, 42029417, 36797223, 41255098, 36321956, 36967710, 27132088, 28185239, 36205100, 40304557, 15277618, 37400775; NCT06806293, NCT06641466, NCT07535723, NCT04968093, NCT07029126 (all ClinicalTrials.gov API v2). Retraction check: none carries a PubMed Retracted Publication or Expression of Concern type or comment, and each DOI returned an empty Crossref updated-by list.
  • Deleted, with reasons:
  • Row 1 (early frequency rise → withdrawal-after-remission trial). Junction already published. PROMPT randomized 514 topiramate responders to continue or stop after 6 months: migraine days rose 1.19 per 4 weeks on placebo vs 0.10 on topiramate, a difference of −1.09 (Diener 2007, PMID 17988947; PubMed shows ErratumIn, Crossref updated-by empty, not retracted). EVOLVE-1/2 followed galcanezumab for 4 months after cessation, and separation from placebo persisted to months 9–10 (Stauffer 2019, PMID 30942898). A 2026 meta-analysis of 8 discontinuation studies (n=1,012) found MMD +4.43 at 3 months off treatment but still −3.78 vs pre-treatment baseline, and found no gepant-cessation study (Makita 2026, PMID 41026450). The uncontrolled I-GRAINE cohort reported off-treatment ≥50% response rising 25% → 53.8% → 77.8% across three yearly cycles (Barbanti 2025, PMID 39862304). What remains (early prevention in people with rising frequency, to avert chronic transition) is OQ-2 itself.
  • Row 3 (GWAS loci × aura vascular risk). Junction already published, and it is internally contradictory. Malik 2015 found that migraine without aura, not with aura, shared polygenic load with large-artery and cardioembolic stroke (PMID 25934857). Zhang 2024 found migraine with aura genetically correlated with early-onset ischaemic stroke (rg 0.208; MR OR 1.074) but not with stroke at all ages (PMID 39733194). Liu 2026 found broader cardiovascular overlap for MA, including ischaemic stroke, and 160 pleiotropic loci (PMID 42350979). The aura-vs-no-aura discrepancy is a candidate new OQ. The row's "ancestry-diverse, adjudicated" qualifier is a cohort wish, not a junction. Query: (migraine[tiab]) AND (stroke[tiab] OR "cardiovascular disease"[tiab]) AND ("genetic correlation"[tiab] OR "mendelian randomization"[tiab] OR "polygenic"[tiab] OR "shared genetic"[tiab]) → 41 records.
  • Row 4 (CSD explains aura / trigeminal CGRP explains pain). Restatement of OQ-11, which already poses the cortical-event-to-meningeal-pain junction with Fraser 2019 and Uzun 2026.
  • Row 6 (cognitive symptoms × drug cognitive harm). Dot B (topiramate cognitive harm) is uncited in the wiki (preventive-treatment.md line 45 and red-flags-and-safety-concerns.md line 129 carry no PMID), and the junction is already published. Martins 2026 ran objective batteries before and after 3–6 months of CGRP antibody or onabotulinumtoxinA in 90 people, and responders improved more in executive function (PMID 42294952). In a randomized trial, topiramate 100 mg slowed three CANTAB reaction-time measures vs placebo in adolescents (Pandina 2010, PMID 20159428). Topiramate reduced word fluency in an open pre/post series (Romigi 2008, PMID 18217888). Query: (migraine[tiab]) AND (cognit*[tiab] OR neuropsycholog*[tiab]) AND (erenumab[tiab] OR galcanezumab[tiab] OR fremanezumab[tiab] OR eptinezumab[tiab] OR "CGRP monoclonal"[tiab] OR "anti-CGRP"[tiab] OR topiramate[tiab] OR atogepant[tiab]) AND (baseline[tiab] OR before[tiab] OR "pre-post"[tiab] OR longitudinal[tiab] OR prospective[tiab]) → 21 records.
  • Row 9 (CHAMP context response × CBT increment → dismantling trial). Junction is being executed. Powers 2013 already separated CBT skills from equal-time attention, education and monitoring (headache days −11.5 vs −6.8; PMID 24368463). REACH (NCT05889624, recruiting, n≈400) randomizes telehealth CBT with vs without amitriptyline, isolating the drug increment that CHAMP questioned (CHAMP placebo response 61%, PMID 27788026). NCT04715685 (active, not recruiting, n≈260) is a 2×2×2 factorial dismantling of a nurse-delivered mind–body intervention in paediatric migraine. Neither registry record is cited in the wiki or in OQ-7; both should be.
  • Row 11 (sparse pregnancy data × exposure registries → registry with untreated-migraine comparator and infant follow-up). The junction is the design of the row's own Dot B. NCT06150781 (Aimovig Pregnancy Exposure Registry, recruiting, n=2,842) already has an erenumab-unexposed migraine internal comparator, a no-migraine external comparator, and infant outcomes to 52 weeks. Only cross-agent harmonization remains, and that is an infrastructure wish.
  • Row 12 (credible sham × sensation unblinding → sham validation before efficacy testing). Junction already published. Fishman 2023 validated two sham thermal nerve-block devices in a prospective single-blind study before a planned RCT. Sham reduced pain 31% vs 52% for active, and blinding indices were ideal (Bang 0, James 1) (PMID 38137166). Also checked: Huang 2024 (PMID 38245660) and Slotty 2015 (PMID 24812035). Query: (migraine[tiab] OR headache[tiab]) AND sham[tiab] AND (neuromodulation[tiab] OR stimulation[tiab] OR device[tiab]) AND (blinding[tiab] OR "blinding index"[tiab] OR "allocation guess"[tiab] OR "guess"[tiab] OR credibility[tiab] OR "sham validation"[tiab] OR "validation of a sham"[tiab] OR "sham device"[tiab]) → 26 records.
  • Row 13 (MOH severity/exposure × CGRP without detox → drug-class-stratified strategy trial). Restatement of OQ-3, which already cites Carlsen 2020 and Sirilertmekasakul 2024 and names opioid/barbiturate dependence. Partial evidence for OQ-3 that was fetched and read: Silvestro 2025, a 200-patient observational study stratifying simple vs complex MOH with and without detoxification under CGRP antibodies (PMID 40457764), and Koonalintip 2025, a 16-RCT network meta-analysis of MOH strategies (PMID 40011869; Crossref updated-by: correction, not a retraction).
  • Row 14 (aura-strongest stroke association × prevention suppresses aura). Restatement of OQ-12, which already cites Zhang 2022 and Gül 2026 and asks whether treatment modifies vascular risk. Dot B has no controlled support in the wiki (only the uncontrolled lamotrigine cohort, Uzun 2026).
  • Absence searches behind the surviving rows (PubMed E-utilities and ClinicalTrials.gov API v2, both on 2026-10-07; Cochrane CENTRAL and SciELO are unreachable, so none of these is a claim of absence everywhere):
  • Row 2: PubMed (migraine[tiab]) AND (CGRP[tiab] OR "calcitonin gene-related peptide"[tiab]) AND (provocation[tiab] OR infusion[tiab] OR "induced migraine"[tiab] OR hypersensitivity[tiab]) AND (predict*[tiab] OR responder*[tiab]) AND (gepant*[tiab] OR atogepant[tiab] OR rimegepant[tiab] OR ubrogepant[tiab] OR fremanezumab[tiab] OR galcanezumab[tiab] OR eptinezumab[tiab] OR erenumab[tiab]) → 15 records. The only provocation-prediction study is erenumab-only (38859744), with its 13-person pilot (30409109). ClinicalTrials.gov query.cond=migraine, query.term=CGRP AND (provocation OR infusion OR provoked) → 38 records, none testing provocation against gepant or ligand-antibody response. NCT05888766 predicts rimegepant acute response from salivary CGRP and evoked potentials, not provocation.
  • Row 5: PubMed (migraine[tiab]) AND ("interictal burden"[tiab] OR "MIBS-4"[tiab] OR "Migraine Interictal Burden Scale"[tiab]) AND (mediat*[tiab] OR "head-to-head"[tiab] OR "active comparator"[tiab] OR topiramate[tiab] OR "direct effect"[tiab] OR "independent of"[tiab]) → 4 records, none a mediation or comparator analysis. The same MIBS-4 terms with randomized controlled trial[pt] → 2 records, both placebo-controlled (36797223, 41255098). The broad MIBS-4 search returned 69 records. ClinicalTrials.gov MIBS-4 OR "interictal burden" → 15 records: the only randomized active comparison is NCT07535723 (not yet recruiting), and NCT04461795 is single-arm open-label fremanezumab (n=40).
  • Row 7: PubMed (migraine[tiab]) AND (stigma*[tiab]) AND (randomized controlled trial[pt] OR randomi*[tiab]) → 3 records, none an intervention trial. The broader … AND (intervention OR randomi* OR trial OR program* OR campaign) → 22 records, none randomized with clinical outcomes. ClinicalTrials.gov term stigma → 1 record (NCT07297901, a breathing-app RCT whose listed outcomes include no stigma measure).
  • Row 8: PubMed (migraine[tiab]) AND (IMPAC[tiab] OR "family burden"[tiab] OR "family impact"[tiab] OR caregiver*[tiab] OR spouse*[tiab] OR dyad*[tiab]) AND (randomized controlled trial[pt]) → 2 records, neither a trial with a family endpoint. The broader treatment/intervention version → 14 records, all observational or unrelated. ClinicalTrials.gov IMPAC OR "Relatives' Stress Scale" OR "family burden" OR "caregiver burden" → 3 records: NCT04968093 (single-arm, n=37, includes the Caregiver Burden Inventory), NCT07029126 (observational, n=308) and one unrelated record.
  • Row 10: PubMed ("menstrual migraine"[tiab] OR "menstrually related migraine"[tiab] OR "menstrual-related migraine"[tiab] OR "menstrually associated migraine"[tiab]) AND ("short-term prevention"[tiab] OR "short-term prophylaxis"[tiab] OR "perimenstrual prophylaxis"[tiab] OR "perimenstrual prevention"[tiab] OR miniprophylaxis[tiab] OR "mini-prophylaxis"[tiab] OR intermittent[tiab]) AND (continuous[tiab] OR daily[tiab] OR "long-term prevention"[tiab] OR topiramate[tiab] OR erenumab[tiab] OR galcanezumab[tiab] OR fremanezumab[tiab] OR atogepant[tiab]) AND (randomized controlled trial[pt] OR compar*[tiab] OR versus[tiab]) → 17 records, none head-to-head. ClinicalTrials.gov query.cond=menstrual migraine → 39 records, none randomizing short-term against continuous prevention.
  • Wiki defects found, NOT fixed (outside this pass's write scope), for the next audit:
  • Misread finding. biomarkers.md (TL;DR and table row) and trigeminovascular-biology-and-cgrp.md §"Human provocation" say CGRP hypersensitivity "enrich[es] erenumab response probability" or "carried predictive information" (Al-Khazali 2024, PMID 38859744). The abstract reports a null: 61% vs 52%, p = 0.498, and "the CGRP-provocation model cannot be used to predict erenumab's effectiveness". Both pages should be corrected and demoted to draft until re-verified. OQ-9 cites the same paper.
  • Misattribution. PMID 25100506 is cited as "MacGregor 2014" in sex-hormones-pregnancy-and-lactation.md, but PubMed lists Silberstein S, Patel S (2014) as authors.
  • Stale OQs. OQ-2 says current trials "mainly measure on-treatment symptom suppression". PROMPT (17988947), EVOLVE post-treatment (30942898) and Makita 2026 (41026450) are post-withdrawal evidence the wiki does not cite. OQ-7 should cite REACH (NCT05889624) and NCT04715685.
  • Stale count. INDEX.md still says "14 unconnected-dots junctions". The table now has 5.
  • Could not resolve: none of the 14 rows was left ungrounded; each was either rewritten or deleted for a stated reason. Rows 7 and 8 rest partly on records the wiki does not yet cite: Sakai 2023 for row 7, and Fofi 2022 and Gantenbein 2025 for row 8. Rows 2 and 5 rest partly on Muñoz-Vendrell 2025, Lampl 2026, Silvestro 2026, Lipton 2023 and Pozo-Rosich 2025. All were fetched and read in this session, but they should be added to BIBLIOGRAPHY.md by a session permitted to edit the literature layer.
  • Independent check still owed: this pass was written by Claude. Under the repository's audit rule, the 5 surviving rows and the 9 deletion rationales should be re-checked by a different engine before tools/verify-dot.sh treats them as assessed.

2026-10-07 — Independent audit of the dot-table grounding pass (auditor: Grok; author of the changes: claude, Opus 5.5)

  • Scope. All 14 rows were uncited before Claude's pass, so all 5 survivors (2, 5, 7, 8, 10) and all 9 deletions were in scope. Every PMID named below was fetched in this session through PubMed E-utilities and read. Every NCT was fetched through ClinicalTrials.gov API v2. The atogepant schedule that ClinicalTrials.gov does not state was read from the linked Japanese registry jRCT2051240268. Crossref updated-by was checked for every DOI of a record cited in the table or used to justify a deletion.
  • Retraction. None of the 14 PMIDs in the surviving table, and none of the deletion PMIDs re-fetched here, carries a PubMed Retracted Publication or Expression of Concern type. Crossref updated-by was empty for all of them except Koonalintip 2025 (PMID 40011869), whose single update is type correction (doi:10.1186/s10194-025-02031-1), matching the PubMed ErratumIn link (PMID 40240915). Diener 2007 (PMID 17988947) has a PubMed ErratumIn link with no PMID attached, and Crossref updated-by is empty. The abstract is marked [corrected], and the figures in the deletion note match that corrected abstract. Malik 2015 has a Crossref has-review relation, not a retraction. Newly cited in the repairs: Takeshima 2026 (PMID 42018233) and Coffee 2014 (PMID 24450487) have empty Crossref updated-by lists and no retraction publication type. Allais 2011 (PMID 20650999) and Hauge 2009 (PMID 19570717), cited in this entry as near-misses, also have empty Crossref updated-by lists.

Rows checked, and citations that failed verification

Five surviving rows were re-read against the live records. The identifiers exist and are the papers and trials named. Two claims did not survive, both motivated anchoring. Anchoring count: 2.

  • Row 2, figures hold. Al-Khazali 2024 (PMID 38859744): 139 participants received a 20-minute CGRP infusion, then 24 weeks of erenumab; ≥50% MMD response in weeks 13–24 was 60/99 (61%) after a provoked attack versus 13/25 (52%) without (p = 0.498). The abstract's conclusion is that the provocation model cannot predict erenumab's effectiveness, and that it is uncertain whether the null extends to ligand antibodies or gepants. The row had said the authors "state that ligand antibodies and gepants are untested"; that verb was stronger than the abstract, and it is now the authors' uncertainty sentence. Muñoz-Vendrell 2025 (PMID 41315923): 252 patients, 29.7% with ≥50% MMD reduction at 3 months, RR 0.79 (95% CI 0.64–0.97) per additional failed antibody. The two dots are different claims. The junction's "or success after a switch" could be read as denying the switch cohort in Dot B; it now says no study has tested whether provocation predicts response after an antibody-to-gepant switch. The written PubMed query re-ran at 15. A broader provocation query returned 40. ClinicalTrials.gov CGRP AND (provocation OR infusion OR provoked) re-ran at 38. The only provocation-prediction study in those sets is this erenumab trial and its pilot (PMID 30409109).
  • Row 5, population error (anchoring). Lampl 2026 (PMID 42327799): 462 CONQUER participants; MIBS-4 versus monthly migraine headache days rs 0.21 at baseline and 0.32 at month 6. Lipton 2023 (PMID 36797223): galcanezumab mean change −1.9 (SE 0.2) versus placebo −0.8 (SE 0.2) at month 3, all patients. Pozo-Rosich 2025 (PMID 41255098): the −0.9 (95% CI −1.4 to −0.4) is the week-12 between-group difference on the Migraine Interictal Burden Scale, not an arm change. Those three match. Silvestro 2026 (PMID 42029417) did not. The full text (Europe PMC, PMC13108233) says 71/200 (35.5%) at month 3 and 81/200 (40.5%) at month 6 reached fewer than 7 attacks, and 32/71 (45.1%) and 22/81 (27.2%) of those subgroups still had MIBS-4 ≥3. The row had placed 27.2% in "200 people with fewer than 7 attacks." The headache-day sentence is the comparison inside the month-6 subgroup, residual burden versus not: monthly headache days 4.11 versus 3.25 (p = 0.127), not a null treatment effect. The row now says that. The mediation absence holds as a PubMed absence: the written query re-ran at 4, the randomized controlled trial[pt] query at 2 (the two placebo trials already cited), and a mediation-term query at 2, both cross-sectional (PMID 42689355; PMID 41634554), neither a treatment-mediation analysis. ClinicalTrials.gov MIBS-4 OR "interictal burden" returned 18, not the 15 in Claude's log. None of the extra records is a completed randomized active comparison that reports MIBS-4. NCT07535723 is not yet recruiting, three arms (onabotulinumtoxinA plus greater occipital nerve block, either alone), and MIBS-4 is a secondary outcome. Completed single-arm records that list MIBS-4 (NCT04461795, fremanezumab, n=40; NCT04465357, erenumab, n=55) do not falsify the active-comparator sentence. TRIUMPH's Japan interim (Takeshima 2026, PMID 42018233) is observational, lists MIBS-4, and does not give a MIBS-4 estimate; it is named in the row so the absence is not read as "no comparative cohort exists." Lipton's full text was not retrieved (no PMC record), so "neither trial abstract decomposes MIBS-4" is what was verified.
  • Row 7, figures hold. Seng 2022 (PMID 36321956): n=121 from neurology offices; SSCI-8 accounted for 6% unique variance in MIDAS severe disability and 5% in MSQ Emotional Function, and mediated headache frequency and those outcomes. The covariate list in the row matches the instruments named in the methods (frequency, ASC-12, PCS, PROMIS anxiety and depression). The regression equation itself is not in the abstract, and the full text was not retrieved. Sakai 2023 (PMID 36967710): 73,432 Fujitsu employees; 82.9% of those without headache said they would change their attitude; about 14.7 full-productivity days per employee with headache. The dots differ. The written PubMed query re-ran at 3, and none of those three is a migraine stigma-intervention trial (PMID 36164278, 34081779, 26015168). The broader 22-record set is reviews, qualitative work, and the Sakai programme. ClinicalTrials.gov stigma re-ran at 1 (NCT07297901); its outcomes do not include a stigma measure. Europe PMC's loose stigma-and-trial query did not surface a stigma-targeted migraine trial in the first records.
  • Row 8, figures hold. Buse 2016 (PMID 27132088): 4,022 dyads; reduced family participation at least monthly in 48.2–57.4%; 71.7% with chronic migraine felt they would be better parents without headaches. Lipton 2017 (PMID 28185239): 12-item IMPAC from 13,064 respondents. Fofi 2022 (PMID 36205100): 27 dyads; Relatives' Stress Scale fell by 6.5 points (95% CI 0.8–12.2) over 6 months. Gantenbein 2025 (PMID 40304557): 173 patients; 87% IMPAC grade III–IV at baseline. The dots differ (a measured family burden versus uncontrolled treatment cohorts). The written RCT-publication-type query re-ran at 2, neither a family-endpoint trial (PMID 11914403; PMID 8392151). A tighter family-measure query returned 1, the same 2002 epidemiology paper. ClinicalTrials.gov re-ran at 3: NCT04968093 is single-arm mindfulness, completed, n=37, with the Caregiver Burden Inventory as a secondary outcome; NCT07029126 is observational fremanezumab, completed, n=308; NCT05099406 is an unrelated chronic-pain stimulation record. No randomized family-endpoint trial turned up.
  • Row 10, design error (anchoring). Silberstein 2004 (PMID 15277618): 546 women, three-way crossover, 6-day dosing; menstrually associated migraine in 67% on placebo and 41% on frovatriptan 2.5 mg twice daily. Zhang 2023 (PMID 37400775): 14 RCTs, 4,601 patients; frovatriptan 2.5 mg twice daily OR 1.87 (95% CI 1.48–2.38) versus placebo for short-term prophylaxis, the top-ranked short-term option. The OR is an effectiveness odds ratio, not an attack-incidence odds ratio. NCT06641466 matches the row: phase 3, recruiting, 723 estimated, 7-day rimegepant versus as-needed rimegepant, placebo, or standard acute care. NCT06806293 does not. It is phase 3, active not recruiting, 468 actual, atogepant versus placebo, with the perimenstrual window as the outcome. The ClinicalTrials.gov record never says "daily" or "continuous." jRCT2051240268, which lists NCT06806293, states atogepant 60 mg or placebo once daily for 7 consecutive days starting 3 days before menses, for 3 cycles. Calling that trial the continuous arm was the citation attached to a contrast the record does not support. A third phase 3 trial, omitted from the row, is the same design: NCT06417775, ubrogepant once daily for 7 consecutive days starting 3 days before menses versus placebo, active not recruiting, 496, phase 3. The head-to-head absence survived a wider search. The written PubMed query re-ran at 17, all short-term-versus-placebo trials or reviews. A broader hormone query returned 19. Coffee 2014 (PMID 24450487) randomizes frovatriptan against placebo during the hormone-free interval after everyone starts an extended-cycle contraceptive; it is not short-term prevention versus continuous prevention. Allais 2011 (PMID 20650999) is an open-label topiramate subgroup of PROMPT, not a head-to-head. NCT00781456 randomizes a 91-day extended-cycle oral contraceptive against placebo (completed, n=109), not against mini-prophylaxis. ClinicalTrials.gov query.cond=menstrual migraine re-ran at 39, and a second query pairing perimenstrual terms with continuous or extended-cycle terms returned 9, none the missing comparison. The row now describes the three phase 3 gepant programmes as 7-day courses and names Coffee as the nearest design.

Rows the auditor deleted: none. Each surviving pair is two different claims. Rows repaired: 2, 5, and 10. Row IDs were not renumbered and no deleted row was restored.

Deletions (9): the stated reasons, checked

The deletions stand. Three reasons were overstated. None was only "this was hard to ground."

  • Row 1 stands. A withdrawal-after-treatment trial is published. PROMPT (Diener 2007, PMID 17988947) randomized 514 people who completed 26 weeks of open-label topiramate, not a responder-only subset: migraine days rose 1.19 per 4 weeks on placebo versus 0.10 on topiramate (difference −1.09, 95% CI −1.75 to −0.43). Stauffer 2019 (PMID 30942898) followed EVOLVE-1/2 for 4 months after galcanezumab; separation from placebo persisted through month 10 for 120 mg and through month 9 for 240 mg (month-10 p = 0.238 for 240 mg). Makita 2026 (PMID 41026450): 8 studies, n=1,012, no gepant-cessation study; monthly migraine days +4.43 (95% CI 2.61–6.25) at 3 months after stopping versus the end of treatment, and −3.78 (95% CI −4.89 to −2.67) after stopping versus pre-treatment baseline. The −3.78 is the pooled post-discontinuation contrast, not specifically the 3-month one (−3.62 versus baseline). Barbanti 2025 (PMID 39862304): off-treatment ≥50% response 25%, 53.8%, 77.8% across three yearly cycles. What the original row still wanted for people with rising frequency is OQ-2.
  • Row 3 stands. The genetic connection is published, and the three papers disagree on aura. Malik 2015 (PMID 25934857): migraine without aura, not with aura, carried the shared polygenic load with large-artery and cardioembolic stroke. Zhang 2024 (PMID 39733194): migraine with aura, rg 0.208 (95% CI 0.038–0.377) with early-onset ischaemic stroke, combined-instrument MR OR 1.074 (95% CI 1.022–1.130), and no association with stroke at all ages. Liu 2026 (PMID 42350979): migraine with aura had the broader cardiovascular overlap, including ischaemic stroke, and the cross-trait analysis found 160 pleiotropic loci across 17 of 21 trait pairs. These are summary-statistic analyses, not an ancestry-diverse adjudicated cohort. That remainder is a cohort wish, as Claude said, not a second unconnected dot.
  • Row 4 stands as a restatement of OQ-11, which already asks for the cortical-event-to-meningeal-pain junction.
  • Row 6 stands against the junction as written. Martins 2026 (PMID 42294952): 90 of 100 completed objective testing before and after 3–6 months of a CGRP antibody or onabotulinumtoxinA; executive-function improvement was greater in responders, independent of treatment group. Pandina 2010 (PMID 20159428): topiramate 100 mg slowed three CANTAB reaction-time measures versus placebo in adolescents aged 12–17. Romigi 2008 (PMID 18217888): open series, word fluency fell. A same-battery comparison of topiramate harm and CGRP-antibody benefit was not the sentence the row asked for.
  • Row 9 stands. Powers 2013 (PMID 24368463): headache days −11.5 versus −6.8 for CBT plus amitriptyline versus education plus amitriptyline. CHAMP (PMID 27788026): placebo ≥50% response 61%. REACH (NCT05889624) is recruiting, about 400, telehealth CBT with versus without amitriptyline. NCT04715685 is active not recruiting, about 260, a 2×2×2 factorial of a nurse-delivered mind–body intervention. The dismantling the row asked for is underway.
  • Row 11 stands. NCT06150781 is recruiting, 2,842 estimated, with an erenumab-unexposed migraine internal comparator, a no-migraine external comparator, and infant outcomes to 52 weeks. Cross-agent harmonization is the remainder, and it is an infrastructure wish.
  • Row 12 stands, with a narrower population than the log implied. Fishman 2023 (PMID 38137166) validated sham thermal nerve-block devices in chronic headache, not in a migraine-only sample, in a prospective single-blind sequence (sham, then active), not a parallel randomization. Sham pain scores fell 31% and in-office active treatment 52%; Bang's index 0 and James's index 1. That is a sham-validation study before a planned efficacy trial. Huang 2024 (PMID 38245660) and Slotty 2015 (PMID 24812035) were re-fetched; neither is retracted, and the deletion does not depend on their figures.
  • Row 13 stands as a restatement of OQ-3. Silvestro 2025 (PMID 40457764) is the observational simple-versus-complex MOH comparison under CGRP antibodies, n=200, with and without detoxification, and no added benefit of detoxification. Koonalintip 2025 is a 16-RCT network meta-analysis; the Crossref update is a correction, not a retraction.
  • Row 14 stands as a restatement of OQ-12. The extra sentence "Dot B has no controlled support" is not true. Tonabersat (Hauge 2009, PMID 19570717) reduced aura attacks from a median 3.2 to 1.0 per 12 weeks versus placebo (p = 0.01) without reducing migraine headache days (3.0 versus 3.0, p = 0.09), in 31 analysed patients. Lamotrigine versus placebo (Steiner 1997, PMID 9137848) did not prevent migraine with or without aura. Neither trial is the long-term vascular-outcome cohort OQ-12 asks for, so the row was not restored. PubMed had no DOI on the Steiner record, so Crossref was not queried for it; its publication types do not include retraction. Hauge's Crossref updated-by list is empty.

Absence queries re-run in this audit

Written PubMed counts that matched Claude's log: row 2, 15; row 5 exact, 4; row 5 randomized-publication-type, 2; row 7, 3; row 8, 2; row 10, 17. Written ClinicalTrials.gov counts that matched: row 2, 38; row 7, 1; row 8, 3; row 10, 39. The one count that moved was row 5's MIBS-4 or interictal-burden registry search, 18 rather than 15, with no new qualifying trial. Broader queries and the near-misses are named under the rows above. Cochrane CENTRAL and SciELO were not searched; both returned HTTP 403 on the 2026-10-07 probe recorded in SOURCES.md. Europe PMC was used as a second look for family endpoints, stigma trials, menstrual strategy comparisons, and provocation-prediction; the loose queries are high-recall and their leading records were not the missing designs.

Judgement for the novelty gate

Before this audit the five-row table could not be trusted by the novelty gate. Row 10 presented a 7-day perimenstrual atogepant regimen as continuous daily prevention, and row 5 attached a 22-of-81 percentage to all 200 treated patients. After the repairs, the five rows can be trusted by the gate. Each row has two distinct claims, the numbers that were checked match the records they cite, the absences that remain name the query and the source, and no cited record is retracted.

The limit on that trust is the reach of the searches. An absence here is a PubMed and ClinicalTrials.gov absence dated 2026-10-07, plus the broader queries in this entry. It is not an absence from CENTRAL or SciELO. Lipton 2023 and Seng 2022 were checked from abstracts. INDEX.md still says the table has 14 junctions; this audit was not permitted to edit it.

2026-10-13 — Periodic literature sweep (author: Codex; independent audit pending)

  • Window and query record: Searched PubMed E-utilities with publication dates 2026-08-28 through 2026-10-13 for diagnosis/aura, burden, CGRP/biology, acute treatment, prevention, chronic migraine/MOH, behavioural/device care, paediatrics, reproduction, comorbidity/vascular risk, biomarkers, patient experience and guidelines. Each topic also had a "Cochrane Database Syst Rev"[jour] query in the same window; all 13 returned zero dated PubMed records. Broad CDSR checks by publication date and entry date also returned zero. The PubMed topic queries returned 429 distinct IDs; 422 records were resolved in the batch E-utilities fetch, and individual cited records were re-fetched. Exact query strings, translations, IDs, publication types and the unresolved batch IDs are in the sweep manifest. Zero CDSR hits means zero in these dated PubMed searches, not zero reviews in CENTRAL or everywhere.
  • Beyond PubMed: Queried Europe PMC REST for the same 13 topics and a focused TITLE_ABS:migraine preprint/PMC query (32 hits; inspected the first 100 returned records where applicable). Non-PMID records included preprints and conference material, but none was promoted to a citation or wiki claim; no DOI-only record therefore needed Crossref resolution. Queried ClinicalTrials.gov API v2 for 45 IDs already cited in migraine content plus NCT05889624 and NCT04715685 added to OQ-7. Queried the Japanese jRCT listing for jRCT2051240268 and ISRCTN for migraine trials; the jRCT direct page returned HTTP 403, while the ISRCTN59164978 record PDF was opened. Checked AHS/AAN and Taiwan guideline listings, WHO/GBD burden pages, and American Migraine Foundation and Migraine Trust patient-organization material. Cochrane Library/CENTRAL and SciELO/LILACS were not searched because of HTTP 403 access barriers; no conclusion about the absence of trials or regional literature is based on those databases. Unpaywall was skipped because LIFEBANK_CONTACT_EMAIL was unset.
  • What survived triage: The 2026 joint AAN/AHS adult prevention guideline (Potrebic, PMID 42673560) updates the 2012 prevention guideline; its companion systematic review (Pringsheim, PMID 42673559) reports 217 included trials in its abstract and limited, generally low/very-low-confidence active comparisons. Taiwan Headache Society issued an updated adult acute-treatment guideline (Lau, PMID 42723131). The new AMF advocacy page dated 2026-08-31 is useful patient-organization material, not quantitative evidence. NCT07568496 changed from not yet recruiting to recruiting; eight of the tracked NCT records had updates posted since 2026-08-28, none with newly posted results in the API snapshot. Other recent randomized and review records, including app-based biofeedback and acute acupuncture studies, did not warrant page edits on this pass. No new burden or registry statistic displaced the existing quick-reference figures, so STATISTICS.md was left unchanged; no paper warranted a landmark deep note.
  • Retraction and access checks: The three added papers had no PubMed Retracted Publication or Expression of Concern publication type and empty Crossref updated-by lists. Crossref was also checked for the two older records corrected below. Europe PMC marks full text for the new guideline, companion review and Taiwan guideline as subscription-required; the repository changes use the guideline metadata, PubMed abstracts and the AHS's directly retrieved society summary, and explicitly label the review's abstract-level claim. No unsupported full-text detail was inferred.
  • Changed: Added three entries to literature/BIBLIOGRAPHY.md; added two documents to literature/guidelines/REGISTRY.md; revised wiki/guidelines.md and wiki/preventive-treatment.md for the 2026 prevention update; corrected the previously flagged false-positive CGRP biomarker wording in wiki/biomarkers.md and wiki/trigeminovascular-biology-and-cgrp.md against the re-fetched null result (Al-Khazali, PMID 38859744); corrected the Silberstein/Patel attribution in wiki/sex-hormones-pregnancy-and-lactation.md (PMID 25100506); updated the NCT07568496 status and snapshot date in wiki/clinical-trials-landscape.md; sharpened OQ-1 and OQ-7 in OPEN-QUESTIONS.md; and added the AMF page to literature/patient-voice/sources.md. Updated INDEX.md counts/statuses. All six edited wiki pages are draft pending independent audit; 12 untouched pages remain curated.
  • Next sweep: Recheck whether the joint guideline or Taiwan guideline full texts become accessible, whether NCT07568496 and the other recently updated trials post results, and whether direct multi-class prevention sequencing or pediatric CBT component trials report outcomes. Audit the six edited pages with a different engine and recheck the five grounded dot rows. Continue to qualify regional evidence limits until SciELO/LILACS can actually be searched.

2026-10-20 — Periodic literature sweep (author: Grok; independent audit pending)

  • Window and query record: The 2026-10-13 sweep already covered publication dates 2026-08-28 through 2026-10-13, so topic searches used 2026-10-13 through 2026-10-20 [dp], plus a migraine [edat] catch-up in the same incremental window. Cochrane, guideline publication-type, and retraction queries used 2026-08-28 through 2026-10-20. Thirteen topic queries plus Cochrane-per-topic, RCT, systematic-review, guideline, retraction, and entry-date searches were run live through PubMed E-utilities; two topic queries (biomarkers_all, guidelines_all) first returned HTTP 429 and were retried successfully. Incremental [dp] plus [edat] yielded 51 distinct IDs (47 not in the 2026-10-13 sweep). Cochrane Database of Systematic Reviews queries were zero on [dp] and [edat] for every topic and on the broad migraine CDSR searches (zero in these dated PubMed searches, not zero in CENTRAL or everywhere). Exact query strings, translations, IDs, publication types, Crossref checks and registry snapshots are in the sweep manifest.
  • Beyond PubMed: Europe PMC REST first returned HTTP 503 on all topic queries; a follow-up retry succeeded (TITLE_ABS:migraine in the incremental window, 24 hits; Cochrane journal query 0; HAS_PMID:N 0; preprints since 2026-08-28, 26 hits). No non-PMID record was promoted, so no DOI-only Crossref resolution was required for a new citation. ClinicalTrials.gov API v2 was queried for 53 NCT IDs already in migraine content, plus a LastUpdatePostDate search (21 records), a StudyFirstPostDate search (2 new records), and a ResultsFirstPostDate search since 2026-08-28 (2 records). ISRCTN JSON API returned HTTP 400; the ISRCTN HTML search returned 200 but is JS-gated. jRCT jRCT2051240268 failed TLS handshake. WHO ICTRP portal returned 200 (JS-driven; programmatic search unproven). ChiCTR and CTRI homepages returned 200 without a usable structured result set. Cochrane Library and SciELO returned HTTP 403; LILACS failed to connect. Unpaywall was skipped because LIFEBANK_CONTACT_EMAIL was unset. Patient-organization and society pages were fetched directly (AMF, Migraine Trust, Migraine Canada, EMHA, Migraine Australia, Headache Australia, NHF, Migraine Ireland, IHS, WHO fact sheet).
  • What survived triage: A REFORM analysis found plasma amylin did not distinguish migraine from controls or predict erenumab ≥50% response (Domínguez-Moreno 2026, PMID 42867283; PubMed abstract; Europe PMC isOpenAccess: N, inEPMC: N). The 2026 AAN/AHS prevention guideline and companion review were republished in Headache (PMIDs: 42673606 and 42673583); these are republications, not a second guideline. NCT04396236 (pediatric lasmiditan, n=847) is terminated for a strategic business decision, with results first posted 2026-09-02. Headache Australia’s homepage returned HTTP 200 and was restored to the verified organization table. Nothing else survived: no new Cochrane review, no new society guideline, no new GBD/WHO statistic that could be read from the JS-rendered WHO page, no RCT in the incremental [dp] window, and no landmark paper. A narrative CGRP-antibody discontinuation review (Lai 2026, PMID 42881126), a 38,472-patient utilization cohort (Trenz 2026, PMID 42878096), candesartan concentration data, the MIND methods paper, imaging/glymphatic reports, acupuncture meta-analyses, and 26 preprints were inspected and not promoted.
  • Retraction and access checks: Incremental records promoted here have no PubMed Retracted Publication or Expression of Concern type. Crossref updated-by was empty for PMIDs 42867283, 42673606, 42673583, 42673560 and 42878096. PubMed returned one Expression of Concern in the seed-to-now window (PMID 42870296) for Casili 2020 (PMID 32066464, mouse NTG/dimethyl-fumarate); that original paper is not in this condition’s bibliography. Crossref updated-by on 10.1186/s12974-020-01736-1 is expression_of_concern (doi:10.1186/s12974-026-04087-5).
  • Changed: Added three bibliography entries (394 unique PMIDs). Recorded the Headache republications in literature/guidelines/REGISTRY.md and wiki/guidelines.md, and completed the US-prevention supersession chain through the 2026 joint guideline. Added the REFORM amylin null to wiki/biomarkers.md, literature/statistics/STATISTICS.md and OQ-9. Updated wiki/clinical-trials-landscape.md snapshot date to 2026-10-20 and listed NCT04396236 as terminated. Restored Headache Australia in literature/patient-voice/organizations.md and noted Migraine Ireland’s HTTP 403. Updated INDEX.md. Edited wiki pages remain draft; 12 untouched pages remain curated. No landmark note was written. STATISTICS.md gained one biomarker row and no new burden figure.
  • Next sweep: Recheck whether the AAN/AHS 2026 guideline/review or the Taiwan acute guideline become open-access; whether NCT07568496, NCT04396236, NCT05810038 or NCT07674654 post usable peer-reviewed results; whether NCT07871097 (vestibular-migraine rimegepant) or pediatric CBT/REACH records change status; and whether any preprint from this window is published. Audit the six 2026-10-13 draft pages plus this sweep’s three edited drafts with a different engine. Continue to qualify regional evidence limits until SciELO/LILACS can actually be searched. The 2026-10-07 wiki defects that remain outside this sweep’s write (OQ-2 still omits PROMPT/EVOLVE/Makita post-withdrawal evidence) should be picked up on the next audit, not invented here.

2026-10-27 — Periodic literature sweep (author: Claude, Opus 5.5; independent audit pending)

  • Window and query record: The 2026-10-13 and 2026-10-20 sweeps had covered 2026-08-28 to 2026-10-20, so the 13 topic queries used 2026/10/20:2026/10/27[dp], with a migraine [edat] catch-up for the same window. Cochrane, guideline publication-type and retraction queries used the full 2026-08-28 to 2026-10-27 window. All queries ran live through PubMed E-utilities (the PubMed MCP server was disconnected this session). Results: 77 distinct PMIDs, 73 not in either prior manifest ([edat] 72, [dp] 27, RCT 2, SR/MA 1). Cochrane Database Syst Rev: zero for every topic, zero for broad migraine by [dp] and [edat], and zero for (headache OR migraine) by [edat]. That is zero in PubMed, not zero in CENTRAL, which could not be searched. Guideline publication type: 2 records (42673606, 42673560), both already registered. Retraction/EoC: 1 record (42870296, the Casili 2020 expression of concern logged 2026-10-20; not in this bibliography). Exact strings, translations, IDs and per-record checks are in the sweep manifest.
  • Beyond PubMed: Europe PMC REST returned 200 on the first try: 29 window hits, 0 without a PMID, 30 preprints since 2026-08-28, 0 Cochrane, 0 guideline-titled. Crossref updated-by was empty for all six promoted DOIs, and none carries a PubMed Retracted Publication/Expression of Concern type. ClinicalTrials.gov API v2: all 62 NCT IDs from the 2026-10-20 manifest re-fetched. The only change was NCT06893874 (machine-prescription pragmatic trial; still COMPLETED; update posted 2026-10-21; enrolment now 169). It is not cited in the wiki, so nothing to edit. LastUpdatePostDate search: 14; new registrations: 1 (NCT07891585, laser acupuncture, not yet recruiting, not promoted). Results posted since seed: 2 (NCT04396236, NCT05810038; both already logged). ISRCTN's XML API returned 200 this time (HTTP 400 last week): 163 migraine records, none migraine-specific updated in the window. WHO ICTRP returned 200 but is JS-only and was not searched programmatically. Cochrane Library and SciELO returned 403, so LILACS/regional literature was not searched. Unpaywall skipped (LIFEBANK_CONTACT_EMAIL unset). Full texts opened: PMC13684737 (GWAS preprint) and PMC13695261 (GON RCT).
  • Guidelines, statistics, patient voice: No new society guideline in PubMed or Europe PMC. The IHS guideline page (200) showed nothing new. ehf-org.org now serves unrelated gambling spam (the domain looks lapsed or hijacked). No migraine markdown links to it; it appeared only as a probe URL in the 2026-10-20 manifest. Future probes should not treat it as the EHF site. AMF and Migraine Trust returned Cloudflare 403. Migraine Canada (200) showed a social post and no campaign. Nothing new for patient-voice/. New statistics: Africa GBD 2023 secondary analysis, plus England (CPRD) and US (MIND claims) care-pattern figures.
  • What survived triage (6 records):
  • (1) A multi-ancestry migraine GWAS meta-analysis, now PubMed-indexed with PMC full text (Overstreet 2026, PMID 42922860; medRxiv preprint, not peer reviewed). It reports 311 European-ancestry and 316 trans-ancestry lead variants, 229 putatively novel. Per the full text, the jump from 123 is partly a looser locus definition. Phenotyping includes headache phecodes in MVP, and African ancestry alone gave 2 GWS variants. Last week's lane saw this preprint via Europe PMC without a PMID and did not promote it.
  • (2) CGRP-therapy BP/BMI trajectories in a 6,361-patient matched EHR cohort: erenumab's slope did not differ from controls, and atogepant reduced BMI (Phan 2026, PMID 42928575).
  • (3) US MIND claims cohort: 46.1% of newly diagnosed adults filled an opioid within 3 years (Cohen 2026, PMID 42929214; Pfizer-funded).
  • (4) England CPRD cohort: 57.1% of headache presentations undifferentiated; 36.5% of people with migraine prescribed a preventive (Watson 2026, PMID 41513289; Pfizer-funded).
  • (5) Africa GBD 2023: 180.4 million prevalent cases (Lu 2026, PMID 42920904).
  • (6) A 48-patient RCT of perimenstrual GON blockade (Akşahin 2026, PMID 42920151; NCT07576309, retrospectively registered). Primary-outcome table checked against full text.
  • Inspected and not promoted: epilepsy-subtype meta-analysis, null on primary (Dörtkol 2026, PMID 42910536; it would only touch the curated comorbidity page). Eptinezumab 300 mg escalation in 57 partial responders (42904798). Rescue-medication claims cohort (42925254). Persistent-aura case-based SR (42931151). Homeopathy open-label RCT (42922208). Erratum notices 42932855 (Makita placebo-response MA, not in this bibliography) and 42900214 (flunarizine VM paper, not in this bibliography). Imaging, mechanistic, case reports and off-topic hits. No landmark note was warranted.
  • Changed:
  • wiki/biomarkers.md: preprint GWAS sentence with its caveats.
  • wiki/preventive-treatment.md: BP/BMI paragraph in Special safety boundaries.
  • wiki/sex-hormones-pregnancy-and-lactation.md: GON row in the short-term prevention table.
  • All three were already draft, stay draft, and have last-curated: 2026-10-27.
  • STATISTICS.md: 4 rows. BIBLIOGRAPHY.md: 6 entries (394 → 400 unique PMIDs).
  • OPEN-QUESTIONS.md: OQ-12 gains the BP-trajectory evidence (intermediate marker, not events). OQ-16 gains the England/US care-cascade leak figures. Neither question is answered.
  • INDEX.md updated.
  • Not touched: the 12 curated pages, guidelines/REGISTRY.md (no new document) and wiki/clinical-trials-landscape.md (no cited-trial status change).
  • Next sweep:
  • Check whether Overstreet 2026 is published in a peer-reviewed journal, and update the preprint flag.
  • Check whether NCT06848075 (elismetrep phase IIb, COMPLETED) posts results; the phase 3 elismetrep trials NCT07645924/NCT07674654 are already in the landscape page.
  • Carry over from last week: NCT07568496, NCT07871097, REACH/pediatric CBT status, and AAN/AHS 2026 open-access status.
  • An independent engine should audit the six drafts. OQ-2 still lacks the PROMPT/EVOLVE/Makita 2026 post-withdrawal evidence flagged on 2026-10-07; leave it to the audit.
  • Keep stating that regional (SciELO/LILACS) literature and CENTRAL are unsearched.