Built all 18 canonical wiki pages: overview; diagnosis/spectrum; epidemiology/course; genetics/neurobiology; mania/mixed states; bipolar depression; maintenance; lithium; antidepressants; psychotherapy/self-management; comorbidity/differential diagnosis; suicide/mortality/physical health; pregnancy/reproductive health; guidelines; biomarkers/digital phenotyping; clinical-trials landscape; patient experience/advocacy; red flags/safety. All remain draft pending a separate audit.
Scale: 2,930 wiki lines; every page 150–191 lines; 286 unique PMIDs across wiki pages; 18 NCT IDs in the trials page.
Built the literature layer:BIBLIOGRAPHY.md with 292 unique PubMed records cited anywhere in the condition; six landmark notes; a 21-entry international guideline/consensus registry; STATISTICS.md with 120 data rows; and a four-file patient-voice layer with method/ethics, nine directly checked public organizations, aggregate themes and annotated sources.
Rewrote OPEN-QUESTIONS.md: 24 stable questions (Tier 1 and Tier 2) plus 12 cross-domain “Dots not yet connected” junctions. Absence statements are explicitly limited to searches retrieved on this date.
PubMed searches run (live E-utilities): diagnosis and diagnostic screening (MDQ/HCL/BSDS, hypomania duration, mixed features, bipolar II/cyclothymia and mimics); prevalence/incidence, treatment gap, course, recurrence, predominant polarity, diagnostic conversion and occupational function; GWAS/cross-disorder genetics, ENIGMA/structural-functional imaging, circadian, inflammatory and metabolic markers; acute mania and mixed-state RCTs/meta-analyses; bipolar-depression RCTs and network meta-analysis; maintenance, discontinuation, lithium concentration/safety/suicide, lamotrigine, valproate and antipsychotics; antidepressant efficacy, treatment-emergent switch, rapid cycling and STEP-BD; psychoeducation, CBT, family-focused therapy, social rhythm and digital self-management; ADHD/anxiety/substance/personality/medical differentials; suicide, cause-specific mortality, cardiovascular/metabolic care; pregnancy, postpartum, lactation, lithium and antiseizure-medication outcomes; guidelines; biomarkers/digital phenotyping; qualitative patient/carer experience and stigma.
ClinicalTrials.gov searches run (live v2 API): condition sweep for bipolar disorder; acute depression, mania, maintenance, neuromodulation, psychedelic/metabolic and precision-treatment records; direct re-fetch of every retained NCT record.
Non-journal sources checked: official guideline pages for NICE, VA/DoD, APA, BAP, RANZCP and JSMD; public organization sites for DBSA, International Bipolar Foundation, Bipolar UK, Bipolar Scotland, Bipolar Australia, Mood Disorders Society of Canada, GAMIAN-Europe, SADAG and Bipolar India. Automated access limitations (HTTP 403/406) are recorded rather than treated as full page verification.
Identifier verification: the final condition-wide live PubMed E-summary check resolved 292/292 unique cited PMIDs; the bibliography reconciliation found no missing, extra or duplicate records, and its first-author metadata matched the live summaries. Direct ClinicalTrials.gov v2 calls resolved 18/18 unique NCT IDs.
Static checks: all 18 wiki pages contain draft frontmatter, TL;DR, Open questions, Related pages and References; all local sibling links target canonical filenames; no unresolved verification or placeholder markers remain.
Could not fully verify: anti-bot responses prevented automated content-level inspection of International Bipolar Foundation and Mood Disorders Society of Canada; their official domains and page identities resolved, and the limitation is disclosed in patient-voice/organizations.md. Currency was not established for every older national guideline; each uncertain status is labeled in the registry.
Next: run the required separate-session audit, re-fetch every claim→identifier pair, check numeric claims against abstracts/full texts, re-run absence searches, correct discrepancies, and promote only fully checked pages to curated.
Scope completed: audited all 18 files in wiki/ (2,958 lines), the 299-record master bibliography, six landmark notes, the 120-row statistics reference, the guideline registry, and all four patient-voice artifacts. Also reconciled INDEX.md and OPEN-QUESTIONS.md because they contain condition-level claims and citations.
Live identifier verification: extracted 299 unique cited PMIDs from the final condition and re-fetched every identifier through PubMed E-utilities on 2026-08-30; 299/299 resolved. Extracted 20 unique NCT identifiers and directly retrieved every /api/v2/studies/{NCT} record; 20/20 resolved. No PMID or NCT was supplied from memory.
Claim/reference reconciliation: for every wiki page, compared the PMID set in the body with its References section. Final result: 18/18 exact matches, no missing reference, no uncited reference, and no wiki PMID absent from BIBLIOGRAPHY.md. The final condition-wide and bibliography PMID sets are identical at 299 unique records.
Quantitative audit: rechecked sample sizes, event rates, effect estimates, confidence intervals, P values and endpoint wording against live PubMed records across all pages, six notes and 120 statistics rows. Estimates that differ because of population or method remain separated rather than averaged.
Citation-label errors found and fixed (10): two uses of PMID 37595997 were attributed to Bahji instead of Yildiz; PMID 1586272 was attributed to Gelenberg instead of Keller; two uses of PMID 36542401 were attributed to Depp instead of Goulding; PMID 25554993 was attributed to Proudfoot instead of Barnes; three overview journal labels were wrong (PMID 37595997, PMID 38470559 and PMID 40408762); and the bibliography journal for PMID 38470559 was corrected to Current Psychiatry Reports. The final label/title/identifier pairs were rechecked against the live records.
Stale-absence and currency searches rerun: repeated targeted PubMed searches for clinical biomarker validation, digital monitoring/intervention, bipolar–borderline discrimination, hypomania thresholds, mixed states, lithium hemodialysis, psychotherapy components and guideline biomarker adoption. Repeated the active ClinicalTrials.gov bipolar-condition query (329 returned records before relevance screening), guideline-site checks and multilingual organization searches.
Substantive/currency issues found and fixed (6 clusters): (1) added 2026 monitoring meta-analysis, SmartBipolar RCT, mixed-diagnosis circadian-feedback RCT and implementation synthesis, preserving the distinction between monitoring, targeted intervention and bipolar-specific evidence; (2) replaced the stale 2017-latest Indian adult-guideline statement with the live-verified 2025 update (PMID 41694044) and added the registered German AWMF update/expired prior-validity state; (3) added two directly verified active recurrence-prediction trials (NCT05828056; NCT06204705) and converted the hospitalization-prevention absence into a dated, query-bounded evidence gap; (4) corrected the geographic organization gap by directly verifying four additional organizations in Japan, Colombia, Brazil and Argentina; (5) replaced the categorical bipolar–BPD discriminator statement with current systematic-review and preliminary classifier evidence while retaining the external-validation boundary; and (6) reran the lithium-hemodialysis search and expressed the gap positively as reliance on toxicokinetic, case-level and consensus evidence rather than randomized efficacy evidence.
Literature-layer corrections: bibliography expanded from 292 to 299 live-verified PMIDs; guideline registry from 21 to 23 entries; trial identifier set from 18 to 20; public-organization directory from 9 to 13. Mood Disorders Society of Canada now fetched directly with HTTP 200; the IBPF anti-bot limitation remains disclosed without inferring inactivity.
Final flags: 0 unresolved verification markers; 0 unresolved citation mismatches; 0 unresolved substantive issues. The IBPF access barrier is a transparent source-access limitation, not an unresolved claim because no program-specific assertion is made from the blocked page.
Promotion: all 18 wiki pages promoted from draft to curated; INDEX.md and CONDITIONS-ROADMAP.md updated to audited.
Window: publications and newly indexed records since 2026-08-28 (prompt date_from; last LOG heading used by the scheduler). Identifier retrieval was live PubMed E-utilities esearch/esummary/efetch and ClinicalTrials.gov v2 /studies/{NCT}; no PMID or NCT was written from memory.
PubMed searches (pdat and edat, mindate 2026/08/28): diagnosis/spectrum; epidemiology/course; genetics/neurobiology; mania/mixed; bipolar depression; maintenance; lithium; antidepressants; psychotherapy/digital self-management; comorbidity; suicide/mortality/physical health; pregnancy; guidelines/consensus (title); biomarkers/digital phenotyping; RCTs; patient experience; plus a bipolar-title catch-all. pdat counts were inflated by September/December 2026 print issues of earlier e-publications (93 bipolar-title hits). The actionable set was the 12 edat-indexed records plus records with pubmed/epub date on or after 2026-08-28.
Triage dropped (honest empty for these topics): no new mania, maintenance, lithium, antidepressant, psychotherapy, pregnancy, or guideline papers that changed a claim. ADHD perinatal cohort (Vigod, PMID 42663665) is ADHD-centered. ERG letter, exploratory imaging (n=33), pediatric white-matter (n≈20/arm), social-support cross-section (n=105), chatbot case reports (n=2), Cureus case material, and an MDD anhedonia consensus (PMID 42314447; entrez June 2026) were not added. September-issue papers already indexed before 2026-08-28 (Halstead Lancet Psychiatry multimorbidity, PMID 42586082, pubmed 2026-08-12; Yocum prescribing trends, PMID 42578538; Marks cortical thickness, PMID 42568249) were not backfilled.
Survived triage — papers (4): Wu et al. Han Chinese rare-CNV analysis, 3,915 cases / 7,820 controls (PMID 42661060); de Deus et al. ERG systematic review, 32 studies n=1,334, with the quantitative meta-analysis limited to MDD (PMID 42669383); Uelman et al. NESARC-III bipolar I × cannabis-use-disorder severity (PMID 42664567); Dong and Ye incident heart-failure meta-analysis, bipolar RR 1.95 (95% CI 1.31–2.91; I²=80.1%) (PMID 42668911; PMC release 2026-08-01, PubMed entrez 2026-08-30). No landmark note.
ClinicalTrials.gov: all 20 previously cited NCT IDs resolved on direct v2 re-fetch; overall status and results-posting were unchanged (none have results). Added three pipeline-changing records: NCT06929273 (KarXT mania open-label extension BALSAM-3; recruiting; n=450); NCT06372964 (lumateperone pediatric bipolar depression phase 3; recruiting; n=384; last update 2026-08-28); NCT07286175 (adjunctive brenipatide time-to-relapse phase 2; recruiting; n=400). Protocol text does not state brenipatide’s pharmacologic class; the page does not invent one. Seen but not added: NCT06696755 icalcaprant bipolar-depression phase 2 (last update 2026-08-14, before the window); mixed-diagnosis pediatric lumateperone safety (NCT06229210); osteoarthritis radiofrequency ablation false-positive (NCT07789340).
Guidelines: live page checks of NICE CG185 (still last updated 2 Sep 2025), VA/DoD 2023, BAP, RANZCP, JSMD, CANMAT and AWMF BipolarLiving found no completed replacement. APA psychiatryonline.org returned HTTP 403. Registry last-verified date updated; no new entry.
Statistics / patient organizations: three new statistics rows (heart-failure RR; two NESARC-III attempt rates). Organization homepages re-fetched: 11/13 HTTP 200; IBPF remains 403; Bipolar Scotland changed from HTTP 200 (2026-08-30) to HTTP 403 without inferring inactivity. No new organization added. IHME/WHO/NIMH portals resolved but did not yield a new bipolar-specific burden figure beyond existing cited estimates.
Pages edited and demoted to draft:wiki/genetics-and-neurobiology.md, wiki/biomarkers-and-digital-phenotyping.md, wiki/suicide-mortality-and-physical-health.md, wiki/clinical-trials-landscape.md. Bibliography 299→303; statistics 120→123 rows; trial identifiers 20→23. OPEN-QUESTIONS.md OQ-15 and OQ-19 sharpened; no question closed. INDEX.md counts and page statuses updated.
Decision noted: PubMed MCP tools were not available; NCBI E-utilities were used as the live equivalent required by the sweep prompt.
Next sweep should chase: posted results or status changes on NCT06951711 / NCT07140913 / NCT06929273 (KarXT mania), NCT07286175 (brenipatide), NCT06372964 (pediatric lumateperone), NCT06696755 (icalcaprant, if updated), and NCT06656416 (ALTO-100 still last verified Jul 2025); any completed AWMF S3/BipolarLiving 2.1, APA replacement, RANZCP replacement, or KMAP-BP sixth edition; whether Wu’s novel Han Chinese CNV loci replicate; a bipolar-specific ERG meta-analysis (de Deus pooled MDD, not BD); and whether Dong’s heart-failure estimate is superseded by a lower-heterogeneity synthesis. Recheck Bipolar Scotland interactively if automated 403 persists.
2026-08-31 — Independent audit of the 2026-08-31 sweep (auditor: Claude; author audited: Grok)¶
Scope: the four wiki pages the sweep edited and demoted (genetics-and-neurobiology.md, biomarkers-and-digital-phenotyping.md, suicide-mortality-and-physical-health.md, clinical-trials-landscape.md), plus every sweep-touched line in BIBLIOGRAPHY.md, STATISTICS.md, guidelines/REGISTRY.md, patient-voice/organizations.md, OPEN-QUESTIONS.md and INDEX.md. Pages the sweep did not edit were not re-audited; they retain the 2026-08-30 audit.
Live identifier verification (this session): all 111 unique PMIDs on the four edited pages plus OPEN-QUESTIONS.md re-fetched by PubMed E-utilities ESummary — 111/111 resolved. All 23 unique NCT IDs re-fetched directly through /api/v2/studies/{NCT} — 23/23 resolved, none with a results section, every page-stated overall status, phase, enrollment and status-verified month matching the live record. The sweep's claim that the 20 previously cited records were unchanged is confirmed. Condition-wide reconciliation: 304 cited PMIDs = 304 bibliography records, and body/References sets match exactly on all four pages (20/34/25/13).
Claim-to-source checks: ~40 claim–identifier pairs read against live abstracts and, where needed, full text. The four new papers' abstracts were fetched by EFetch and each new sentence compared against them; the three new trial records were read at protocol level (title, conditions, interventions, eligibility, primary outcome).
Substantive error found and fixed (1): the Dong 2026 heart-failure claim (PMID 42668911) was attributed to a "seven-cohort synthesis." Seven retrospective cohorts met the review's inclusion criteria overall, but the PMC full text (PMC13525515) states that four contributed the bipolar estimate (McDermott 2005, Chen 2022, Yoo 2023, Lee 2024). Corrected on suicide-mortality-and-physical-health.md, in the STATISTICS row and in OQ-15; added the leave-one-out range (1.73–2.40) and the fact that the bipolar and schizophrenia intervals overlap with no formal between-group test, so the "higher than schizophrenia" ordering is a point-estimate comparison only.
Precision fixes (2): on biomarkers-and-digital-phenotyping.md, the de Deus ERG review (PMID 42669383) pooled only five of its 32 included studies — added to both the table row and the body, since "the only significant pooled parameter" reads stronger without it; and the significant parameter is the mixed-scotopic b-wave implicit time, with its CI lower bound at the null (g=0.23, 0.00–0.46), now stated.
Consistency fix (1): the sweep cited PMID 42314447 in guidelines/REGISTRY.md search provenance without adding it to BIBLIOGRAPHY.md, breaking that file's stated invariant and the INDEX count. Record verified live (Cutler AJ, et al. Psychiatry Res. 2026;364:117225) and added as an explicitly screened-and-excluded, non-bipolar entry. Bibliography 303→304; INDEX counts updated.
Source-access fix (1): Bipolar Scotland, recorded by the sweep as HTTP 403, returned HTTP 200 on the audit re-fetch (title "Bipolar Scotland - Scotland's Bipolar Charity"). The directory and source note now record the successful retrieval and identify the 403 as a request-level anti-bot response. Re-checked with the same method: ibpf.org still 403 (limitation correctly disclosed), psychiatryonline.org still 403.
Absence and currency re-checks: re-ran the ClinicalTrials.gov active-record query behind the trials page's hospitalization-prevention gap — no active record randomizes an alert-plus-response pathway with hospitalization prevention as the clinical-utility endpoint; the nearest hits are a transitional-care feasibility study (NCT07213492, primary outcome feasibility/acceptability) and a probiotic relapse trial (NCT03349528), so the dated, query-bounded gap statement stands. Re-fetched NICE CG185: last updated 02 September 2025, exactly as the sweep recorded.
Verified without change: Wu 2026 (PMID 42661060) — 3,915 cases / 7,820 controls, rare-deletion overrepresentation, only 3q29 and 15q11.2 replicating among 12 European loci: matches the abstract verbatim in substance. Uelman 2026 (PMID 42664567) — 33.71% (21.64–48.36) vs 16.65% (13.17–20.83), n=753 / 35,556, correctly framed as adjusted cross-sectional rates. All three new trial rows match their records, including BALSAM-3's mania/mixed-mania population and the sweep's correct refusal to invent a pharmacologic class for brenipatide (the protocol states none; it does describe a self-injected agent with diabetes-related exclusions).
Verdict: the sweep was sound in its identifier handling — no fabricated or mismatched citation, no stale absence, honest recording of what it could not fetch. The single real defect was the cohort-count overstatement, which the abstract alone would not have caught. Promotion: all four pages promoted draft → curated; INDEX.md returned to 18 curated pages. 0 items remain flagged.
Next sweep should also chase:NCT05065294 (psilocybin therapy for bipolar II depression, active-not-recruiting) surfaced in the audit's re-query and is not yet cited on the trials page — consider it when next revisiting the bipolar II portfolio gap; and whether Dong's four-cohort heart-failure estimate is superseded by a larger, lower-heterogeneity synthesis.
Scope and disposition: extended all 18 canonical wiki pages without restarting or deleting prior verified content. Every touched page was set to status: draft and last-curated: 2026-09-03 for independent re-audit.
Evidence-base result: condition-wide PubMed coverage increased from 304 to 393 distinct records (21.8 per wiki page across the condition); 382 distinct records now appear directly across wiki pages. The master bibliography was reconciled to all 393 condition-wide PMIDs, including four earlier records cited only as screened context in this log.
Searches run: 90 live topic-specific PubMed E-utilities searches—five per page—covering functional burden and recovery; diagnostic validity across age groups; incidence/onset/course; genomics, circadian biology, bioenergetics and neuroprogression; acute mania and mixed states; bipolar depression; maintenance and discontinuation; lithium concentration, renal effects, interactions and pharmacogenomics; antidepressant efficacy/switch/continuation; psychotherapy components and lifestyle; ADHD/OCD/autism/epilepsy differentials; suicide and physical-health pathways; pregnancy/postpartum/lactation; guideline grading and disagreements; circulating/imaging/polygenic/digital biomarkers; experimental treatments; stigma, recovery, adherence and employment; and emergency recognition/agitation/prediction. Additional targeted live searches resolved long-acting injectable maintenance evidence and lithium-response GWAS evidence.
Identifier integrity: live PubMed ESummary resolved 393/393 PMIDs present anywhere in the final condition; selected new records were also retrieved with EFetch for abstract-level claim extraction. Direct ClinicalTrials.gov v2 /studies/{NCT} calls resolved 29/29 NCT identifiers with no errors. No PMID or NCT was written from memory.
What was deepened: added actual effect sizes and confidence intervals for cognitive remediation, youth mania treatment, valproate, lumateperone, ketamine versus ECT, enriched maintenance therapy, LAIs, antidepressant continuation, lithium concentration and renal decline, psychoeducation, comorbidity prevalence, smoking/physical activity, postpartum relapse, lactation exposure, biomarkers and passive relapse prediction. Added explicit contradictions where symptom benefit did not yield function, internal digital performance lacked external validation, continuation reduced depression but numerically increased mania, and lithium efficacy gradients coexist with renal risk.
Guidelines and literature layer: added mixed-presentation CANMAT/ISBD ranks, ACOG modified-GRADE/good-practice distinctions, historical BAP/CANMAT provenance and a NICE-methods critique; expanded STATISTICS.md by 20 source-locked rows; sharpened four open questions and added two cross-domain junctions.
Deliberately left alone: canonical page structure, verified pre-existing prose and citations, the six landmark paper notes, patient-voice organization files, and non-bipolar conditions. No speculative biomarker, psychedelic, ketogenic-diet or digital-monitoring efficacy claim was promoted beyond the design and population actually studied; mixed-diagnosis estimates remain labeled as such.
Next audit: independently re-fetch and check every new claim–PMID pair, especially author labels and exact numeric transcription; confirm that guideline rank terminology matches each source's own grading system; and validate that pooled mixed-diagnosis ketamine/ECT and digital estimates are not read as bipolar-specific effects.
2026-09-03 — Independent audit of the evidence-broadening pass (auditor: Claude; author audited: codex)¶
Scope. The 2026-09-03 deepening pass by codex touched all 18 canonical wiki pages plus INDEX.md, OPEN-QUESTIONS.md, BIBLIOGRAPHY.md, guidelines/REGISTRY.md and statistics/STATISTICS.md. This audit targeted the new material. The pre-pass per-page citation sets were reconstructed from the author session's own file inventory, giving an exact new/old split; previously audited prose (2026-08-30 and 2026-08-31) was not re-litigated, but every inline author/year label on all 18 pages was machine-checked against live ESummary records.
Citation breadth — the measure this pass existed to move. Before: 297 distinct PubMed records across the 18 wiki pages = 16.5 per page (mean 21.3 citations per page). After: 382 distinct records = 21.2 per page (mean 26.3 citations per page). Eighty-six records were new to their page and 85 were new to the condition; no prior citation was dropped. The additions were spread deliberately — 4–6 new records on every page, none reused across more than one page except Simjanoski 2023 (PMID 37263531), which was already cited on genetics-and-neurobiology.md. This was genuine widening, not padding with the condition's existing landmark set; page length rose only ~17 lines per page (2,958 → 3,267 wiki lines) while the distinct-record base rose 29%.
Live identifier verification (this session). All 393 unique PMIDs anywhere in the condition re-fetched by PubMed E-utilities ESummary: 393/393 resolved. All 86 page-new records additionally retrieved by EFetch and read at abstract level; two read at PMC full-text level (PMC6380965, and the earlier-audited PMC13525515 context). All 33 unique NCT IDs re-fetched directly through /api/v2/studies/{NCT}: 33/33 resolved, none with posted results; overall status, phase, enrollment and status-verified month matched every page-stated value. No identifier was written from memory.
Reconciliation. Body-vs-References PMID sets match exactly on 18/18 pages. Condition-wide cited set = bibliography set = 393, with no duplicate, missing or uncited record. All wiki links resolve to canonical filenames; every page carries frontmatter, TL;DR, Open questions, Related pages and References; page lengths 166–212 lines; zero [unverified] or placeholder markers.
Claim-to-source checks. Roughly 110 new claim–identifier pairs were read against the live records. The large majority transcribed exactly, including hard cases the author handled well: the Hou 2016 lithium-response hazard ratio, whose direction is confusingly reported in the source (correctly attributed to the alternate alleles); the Vita 2025 lithium mania-response estimate that "barely crossed the null" (RR 1.35, 1.00–1.83, p=.049, very-low confidence); the Kishi 2016 rapid-cycling sensitivity analysis; and the Imaz 2019 lactation figures reproduced with the source's own denominators.
Defects found and fixed (8).
antidepressant-controversy.md and red-flags-and-safety-concerns.md both stated that Melhuish Beaupre 2020 (PMID 32134853) estimated antidepressant-associated mania at 14%. That figure is background context in the review's opening, not a pooled estimate it derived. Both sentences rewritten to describe what the review actually did (24 clinical and 10 genetic studies; five clinical factors meta-analysed) and to label the 14% as cited background.
bipolar-depression.md attached the Durgam 2025 adverse-event rates (PMID 39946099) to the 200-participant bipolar subgroup; the abstract reports them for the combined MDD/bipolar mixed-features population. Now labelled, with the absent mania/hypomania treatment-emergent events added.
bipolar-depression.md cited Shi 2025 (PMID 39549887) without noting that the paper carries a published expression of concern (J Affect Disord. 2025;378:350–351). Disclosed, and the vague "high heterogeneity" replaced with the actual I² values (75% response, 90% remission, 99% pooled symptom difference).
comorbidity-and-differential-diagnosis.md quoted Salazar de Pablo 2025's 3.72% bipolar occurrence after stimulant exposure (PMID 40900605) without its confidence interval — a number stripped of the context that makes it meaningful. The interval (0.77–16.05 across four studies, a twentyfold span) and the two other intervals are now stated.
maintenance-and-relapse-prevention.md presented Kishi 2016's RR 0.63 (PMID 27207910) as a pooled estimate "across seven LAI trials overall". Seven trials were identified (n=1,016); RR 0.63 comes from the risperidone-LAI-versus-placebo comparison specifically. Corrected, and noted that the report gives P values rather than confidence intervals for those estimates.
red-flags-and-safety-concerns.md attributed to Pacciardi 2019 (PMID 30721526) a caution against generalizing inhaled-loxapine safety to unsupervised community use. The review argues the opposite — it advocates community self-administration on the basis of preliminary results from the first ten participants of NCT02525991. Rewritten to state the review's actual position, with the respiratory contraindication quantified from the PMC full text (symptomatic bronchospasm 53.8% in the asthma trial, n=52, and 19.2% in the COPD trial, n=53, versus 11.5% and 11.1% on placebo) and with NCT02525991 verified live: completed 2019-12-31, n=323, status verified Nov 2020, no results posted as of today — so unsupervised use remains a proposal.
patient-experience-and-advocacy.md credited Jagfeld 2021 (PMID 34139411) with organizing bipolar experience around the CHIME processes. CHIME is pre-existing; the review's contribution is the bipolar-specific POETIC adaptation that makes "Tensions" an explicit component across 12 qualitative studies. Corrected, with the review's own demographic limitation added. Its reference line also named the wrong journal for Astill Wright 2025 (PMID 41105870) — J Affect Disord. → J Med Internet Res. 2025;27:e71525.
psychotherapy-and-self-management.md ended its lifestyle paragraph with curation-process commentary ("the new comparison here is deliberate recurrence rather than evidence-base widening"). Removed — that is a note about the build, not content about bipolar disorder. The substantive Simjanoski finding was kept.
Stale absence found and corrected (1).clinical-trials-landscape.md concluded its ketogenic-diet paragraph with "Registered, adequately controlled bipolar trials are needed before efficacy claims." A live ClinicalTrials.gov query today returns several bipolar-specific registered ketogenic studies. The sentence now names them with live-verified detail — NCT06221852 (randomized, open-label, first-episode, n=50, recruiting), NCT07121894 (single-arm bipolar depression, n=30, recruiting) and NCT06920940 (adolescents, n=80, recruiting) alongside the already-tracked NCT06081426 — and states the absence precisely: none is blinded, all are ongoing, and no completed adequately powered randomized efficacy trial with a mood-episode endpoint exists.
Absence re-verified and re-dated (1). The trials page's hospitalization-prevention gap was re-queried today across three formulations. It stands: no active record randomizes an alert-plus-response pathway with hospitalization prevention as the clinical-utility endpoint; the nearest records remain NCT07213492 (primary outcomes feasibility and acceptability) and NCT03349528 (probiotics). The bullet is re-dated to 2026-09-03 and now names those nearest records.
Pre-existing defect found in passing and fixed (1).PMID 33845326 was cited on comorbidity-and-differential-diagnosis.md and in BIBLIOGRAPHY.md as "Villarroel J, et al. Machine learning for the differential diagnosis of bipolar and borderline personality disorders. Psychiatry Res.". The record is Bayes A, et al. "Differentiation of bipolar disorder versus borderline personality disorder: a machine learning approach." J Affect Disord. 2021;288:68–73. This survived the 2026-08-30 and 2026-08-31 audits. Author, title and journal corrected in both files; the claim's numbers were correct but its 73% overall accuracy is now shown with the asymmetry that makes it interpretable (84.1–87.8% for bipolar disorder versus 50.0–57.7% for borderline personality disorder — chance-level for the latter in a small, bipolar-II-weighted sample).
Checked and confirmed without change. BEAM-BD (PMID 37530824) hazard ratios and event counts; Elmosalamy 2025 bipolar-II monotherapy counts and rates; Vega-Núñez, Rowland, Barendse, Song and Fang on the biomarkers page; Rhee 2022 and Rhee 2025 ECT estimates; Ferentinos, Lai, Kwon and Betzler on comorbidity; Alcaíno, Duffy, Youngstrom, Pavuluri and Dols on diagnosis; Solmi, Pedersen, Esaki, Lee and Harvey on epidemiology; Kuang, Liang, Yan, Passos and Muneer on genetics; Yatham 2021/2006, Jauhar, Goodwin and ACOG on guidelines; Hsu, Macaron, Arnold, Al-Soleiti and Hou on lithium; Nestsiarovich, Mari, Kishi 2021 and Bai on maintenance; Vita, Fornaro and Fountoulakis on mania; Dean, Samamé, Silva Ribeiro, Afzal and Bora on overview; Tse, Latifian, Favre, Loots and Astill Wright on patient experience; Imaz, Newmark and Costoloni on pregnancy; Tan, Bond, Hautzinger and Frank on psychotherapy; Lublóy, Gonzalez and Thompson on red flags; Vancampfort, Pietschnig, Hu, Braithwaite and De Hert on suicide/physical health. All 20 new STATISTICS rows carry source, population and method matching their records; both new REGISTRY entries and both new "Dots not yet connected" junctions match their sources.
Verdict. The deepening pass did the thing it was asked to do: it widened the evidence base rather than lengthening pages against the same references, and it was disciplined about not over-reading mixed-diagnosis, internally validated or enriched-design estimates. Its real failure modes were a reversed source stance (Pacciardi), a background figure presented as a derived estimate (Melhuish Beaupre, twice), a subgroup/whole-population mix-up (Durgam), a pooled-versus-pairwise conflation (Kishi 2016), one undisclosed expression of concern, one stale absence, and one leak of build commentary into page prose — none of which the abstract alone would catch without deliberately re-reading each source against each sentence.
Promotion. All 18 wiki pages promoted draft → curated. INDEX.md updated to audited. 0 items remain flagged.
Next sweep should chase: posted results for NCT02525991 (self-administered inhaled loxapine outside hospital; completed 2019, still unposted); completion/results of the bipolar ketogenic trials NCT06221852, NCT07121894 and NCT06920940; whether the expression of concern on Shi 2025 (PMID 39549887) resolves to a retraction or correction; NCT05065294 (psilocybin for bipolar II depression), still uncited on the trials page; and whether Fang 2026's call for closed-loop intervention trials is answered by any registered alert-plus-response protocol with a hospitalization endpoint.