Skip to content

Curation log — post-traumatic stress disorder

Newest entries first. Every content-changing session appends an entry: date, what changed, what was searched, follow-ups for next time. See CLAUDE.md.

2026-09-02 — Seeded (Claude)

Created the scaffold: INDEX.md with a 22-page plan (20 topical + 2 standing house pages) and a 12-record anchor table, a seed OPEN-QUESTIONS.md, this log, and the empty wiki/ and literature/ layers. No wiki pages written.

Searches run (live PubMed E-utilities, 2026-09-02). Scoping: post-traumatic stress disorder 60,888; stress disorders, post-traumatic[MeSH Major Topic] 41,226. Per-page topic counts in INDEX.md, including epidemiology 19,588, children/adolescents 6,527, complex PTSD/ICD-11 4,283, military/veterans 3,936, cognitive processing therapy 3,810, prolonged exposure 1,023, comorbidity 923, EMDR 874, dropout/attrition 870, SSRI 664, moral injury 553, prazosin 129, MDMA 129, ketamine/psychedelic 80.

Query-hygiene note. Three initial queries returned implausible numbers through loose automatic term mapping — post-traumatic stress disorder AND moral injury returned 54,812, close to the entire PTSD literature, because "injury" mapped to a very broad term. Re-run as a quoted phrase it returns 553. Affected counts are marked † in INDEX.md. An unchecked count here would have justified a page size the literature does not support.

Anchor records resolved live: 12, listed in INDEX.md.

Scoping decisions. Three borders. The condition owns the disorder, not trauma exposure in general, whose literature is far larger. Comorbid depression and substance use are cross-linked to depression and generalized-anxiety-disorder rather than absorbed. Complex PTSD is in scope and presented as an unresolved nosological dispute, because the DSM-5/ICD-11 divergence changes who is diagnosed and which trials apply to them.

Currency warning written into the seed. Two areas move faster than a seed can hold. The MDMA-assisted therapy phase 3 papers (PMID 33972795, 37709999) report trial results and not the subsequent regulatory outcome; the builder must establish current regulatory and replication status by live search, including registry and regulatory sources, and record the search date. Do not infer a regulatory outcome from a trial result. The same applies to the ketamine and psychedelic literature. Separately, prazosin's positive early trials (PMID 12562588, 17069768) were followed by later re-testing that is not among the anchors and must be retrieved.

Flagged for the build pass. No anchor yet for ISTSS/VA-DoD/NICE guidelines, SSRI regulatory trials, criterion-A and instrument validation, moral-injury primary sources, early pharmacological prevention, or patient-organisation material.

Follow-ups. Build with tools/build-condition.sh ptsd; auditor must differ from the writer.

2026-09-02 — Full build (Codex)

Built all 22 canonical PTSD wiki pages at draft status; created a 345-record bibliography, 6 landmark notes, international guideline registry, statistics sheet, four-file patient-voice layer, 15-question research agenda and updated index. Pages explicitly separate DSM-5 PTSD, ICD-11 PTSD and CPTSD; mixed-anxiety estimates are labelled and only PTSD strata are used; GAD and depression are cross-linked rather than absorbed.

Live PubMed searches. Re-verified all seed anchors and searched diagnosis/CAPS-5/PCL-5/ITQ, CPTSD, epidemiology/risk, neurobiology/fear extinction/HPA/memory, PE, CPT, EMDR, comparative psychotherapy, dropout/nonresponse, pharmacotherapy/SSRIs/SNRI, prazosin/nightmares/IRT, MDMA/psychedelics/ketamine, early intervention/debriefing/hydrocortisone/propranolol, veterans, pediatrics, comorbidity, moral injury, guidelines, suicide/safety and qualitative/preferences. EFetch resolved 360 records; 345 unique records were retained after removing non-specific search hits.

Key corrections to the seed. Retrieved the later prazosin re-test (PMID 29414272). Established current MDMA status from a live-retrieved 2025 review: FDA rejected the initial 2024 NDA and required another phase 3 trial (PMID 39741438); positive phase 3 results were not treated as approval.

2026-09-02 — Registry, policy and patient-source verification (Codex)

Queried ClinicalTrials.gov v2 live for PTSD and individually re-fetched 11 retained NCT records: NCT03537014, NCT04077437, NCT04959695, NCT04560660, NCT06795659, NCT06549049, NCT05341882, NCT04007666, NCT03711266, NCT05560854 and NCT07430657. Excluded broad condition-query records where PTSD was only an incidental outcome.

Fetched official VA/DoD 2023, APA 2025, NICE NG116, ISTSS 2018, Phoenix Australia/RANZCP and WHO guideline pages. Verified nine public organizations/resources; patient themes use public or PubMed-indexed sources only, are aggregate paraphrases, and retain no private identities.

Follow-up for independent audit. Re-fetch every cited PMID and NCT; check article–claim fit and exact metadata; inspect full texts where abstracts do not support nuance; audit registry–publication concordance; validate all relative links; leave unresolved pages draft.

2026-09-02 — Independent audit (Claude; author was Codex)

Auditor: Claude. Author of the material audited: Codex (build entry above). A different engine wrote every page checked here; no model audited its own writing.

Scope of checking

Object Count checked
Unique PMIDs re-fetched live (esummary + efetch) 385 (345 as built, 40 added during the audit)
Reference-list lines checked against the live record (title, journal, year, volume, pages) 659
Evidence-ledger rows checked against the live record 655
Inline claim→citation pairs read against the source text 571 instances across 22 pages, plus every numeric claim traced to the abstract that states it
NCT IDs re-fetched individually from the ClinicalTrials.gov v2 API 15 (11 as built, 4 added)
Landmark notes re-checked figure by figure 6
Patient-voice organization URLs re-fetched 9 (7 returned HTTP 200 with a matching title)
Guideline issuing-body pages re-fetched 3 attempted

Every one of the 385 PMIDs resolved, and every one matched the author, year, journal and title it was cited under. No fabricated or misattributed identifier was found anywhere in the condition — the build's citation hygiene at the identifier level was clean. The errors were all of a different kind: real records attached to claims they do not support.

Errors found and fixed

A. Claim–citation mismatches (real record, wrong claim). Fourteen citations supported claims their papers do not make. All were corrected by re-anchoring the claim to a live-retrieved source that does support it, and by stating the number the source actually reports.

Page Claim What was cited Fix
cognitive-processing-therapy.md "Telehealth and massed formats can reduce travel or duration…" PMID 35187740 (mindfulness for nurses' burnout) and PMID 30388297 (Cochrane review of ICU follow-up services) Replaced with four CPT-specific delivery trials: Morland 2015 (PMID 26243685), Peterson 2022 (PMID 35038985), Held 2022 (PMID 35338534), Bruce 2025 (PMID 38946118); both wrong records removed from the page
emdr.md "Working-memory loading predicts reduced vividness/emotionality" PMID 32043428 (EMDR efficacy meta-analysis) and PMID 27726123 (Cochrane pediatric psychological therapies) Replaced with the two dismantling syntheses that measure this: Lee 2013 (PMID 23266601, d=0.41 clinical / 0.74 laboratory) and Mertens 2021 (PMID 34619553, k=53). The two displaced records were moved to sections they do support, with their real numbers
pharmacotherapy.md "Class labels should not imply equal evidence for every molecule" PMID 10761678 (psychosocial treatment review) and PMID 31672337 (cannabinoid meta-analysis) Replaced with the Cochrane pharmacotherapy numbers (PMID 35234292); the cannabinoid review moved to the emerging-agents section where it belongs
pharmacotherapy.md "A negative or small augmentation trial cannot be rescued by a class effect" PMID 38647566 (music therapy meta-analysis) Replaced with Hamner 2019 ziprasidone augmentation RCT (PMID 30640209), retaining Davis 2025 (PMID 39693081)
sleep-and-nightmare-treatment.md "IRT rescripts recurrent nightmares…protocols differ" PMID 25036580 and PMID 24612528 (both prazosin/general PTSD reviews) Replaced with Krakow 2001 (PMID 11476655) and Yücel 2020 (PMID 31855732); the two prazosin reviews re-sited in the prazosin section, where they now document a genuine disagreement with the null trial
sleep-and-nightmare-treatment.md "Trials should measure both sleep and PTSD" PMID 27432823 (prazosin review) Replaced with the two sequencing RCTs: Pigeon 2022 (PMID 34265777) and Colvonen 2025 (PMID 40488726)
dropout-and-nonresponse.md "Session-level change, alliance and avoidance…" PMID 29252162 (mindfulness review) and PMID 29169664 (long-term efficacy meta-analysis) Replaced with Wright 2024 IPD meta-analysis (PMID 39537555) and Maglione 2022 (PMID 34800059)
epidemiology-and-risk.md "Prospective cohorts with pre-trauma measurement…" PMID 29869691 (ambulance-personnel prevalence) Replaced with two actual pre-trauma cohorts: Acheson 2025 (PMID 41103636, n=643) and de Voogd 2025 (PMID 40024495, n=221); the ambulance review moved to the occupational-populations section with its real 11% figure
epidemiology-and-risk.md "Occupational populations should not be used as generic civilian estimates" PMIDs 30422749 and 29609900 (both older-adult/geriatric papers) Re-anchored to Petrie 2018 (PMID 29869691); the geriatric papers now support the age-band claim they actually address
red-flags-and-safety-concerns.md "Adult medication approvals cannot be imported [to children]" PMID 39630422 (pediatric psychotherapy NMA — says nothing about drug approvals) Replaced with the NICE pediatric pharmacological evidence review (PMID 32757557) and Stoddard 2011 (PMID 22040192)
patient-experience-and-advocacy.md "Telehealth helps some barriers and creates privacy/digital-access barriers" PMID 35898808 (self-stigma prevalence) Replaced with Bruce 2025 (PMID 38946118); the self-stigma paper now supports an accurately stated stigma claim
literature/patient-voice/themes.md "Practical access" and "digital care trade-offs" themes Both rested on PMID 35898808 (self-stigma) Re-anchored to PMID 39295155 and PMID 38946118; a separate, correctly stated stigma theme added so the displaced source is used for what it shows
early-intervention-and-prevention.md "Small heterogeneous trials do not establish routine prevention" PMIDs 36402246 and 26808441 (both preclinical/mechanistic reviews) Kept as mechanism citations and added the actual trial synthesis: Astill Wright 2019 (PMID 31819037), 19 RCTs, hydrocortisone RR 0.21 (95% CI 0.05–0.89), everything else null
prolonged-exposure.md "Inhibitory learning, expectancy violation… are competing models" PMIDs 16891563 and 16919525 (2006 neuroimaging reviews) Added Craske 2014 (PMID 24864005), the source that states the model; the imaging reviews re-scoped to the circuitry claim they support

B. Stale absences (the failure mode this audit is designed to catch). Three asserted absences were re-searched live and two had gone stale.

  1. Psilocybin. mdma-and-psychedelic-assisted-therapy.md carried a 2022 review's statement that no psilocybin PTSD trial had been conducted (PMID 35711024). Re-running the search on 2026-09-02 found two published PTSD trials: McGowan 2026 (PMID 40883964; phase 2, non-randomized, open-label, n=22, CAPS-5 −29.9 at week 4) and Armstrong 2026 (PMID 42533157; open-label veteran pilot, n=12, −27.5 points, 95% CI 19.9–35.1, NCT05554094). Rewritten as a dated, positively-stated gap: as of 2026-09-02, 34 participants across two single-arm studies and no randomized controlled psilocybin PTSD trial published.
  2. Prazosin responder phenotype. The build (and OQ-4) treated the responder question as untested. Live searching found the candidate marker — pretreatment standing systolic blood pressure, +14-point CAPS reduction per 10 mm Hg under prazosin (Raskind 2016, PMID 27320368), contested by Manhapra 2019 (PMID 30143236) — and a live registry check found NCT03539614, a phase 3 prazosin-versus-placebo trial built around noradrenergic biomarkers (n=87), with an actual primary completion date of 1 January 2026 and no results publication indexed in PubMed. The gap is now stated as "complete but unreported", not "untested".
  3. ITQ-stratified CPTSD trials. complex-ptsd.md said "trials rarely stratify by ITQ-defined CPTSD". Still broadly true, but a live search found the RESTORE pilot RCT (Karatzias 2024, PMID 38688242) and a moderator analysis (Hoeboer 2021, PMID 33831660) in which CPTSD status neither predicted nor moderated exposure-therapy outcome. Restated precisely: a pilot against usual care exists; the comparison against active trauma-focused therapy in ITQ-defined CPTSD has not been run.

C. Currency corrections.

  • Ketamine. Both pharmacotherapy.md and the MDMA page asserted only that "ketamine evidence for depression is not PTSD evidence", while PTSD-specific randomized evidence exists and disagrees with itself. Added Feder 2021 (PMID 33397139; n=30, CAPS-5 −11.88, d=1.13) and Abdallah 2022 (PMID 35046508; n=158, null on PCL-5 and CAPS-5) side by side, with the population and dosing differences stated rather than averaged.
  • MDMA regulatory status. "A 2025 review reports the FDA rejected the initial 2024 application" is now dated to the decision itself (declined August 2024) and corroborated by a second 2026 review (Morland 2026, PMID 41820235), with a 2026 meta-analysis of 8 RCTs added (Fares-Otero 2026, PMID 41825162; SMD −1.19, 95% CI −1.95 to −0.42, most trials at high risk of bias in outcome measurement).
  • New drug class. The TSND-201 (methylone) phase 2 RCT (Jones 2026, PMID 41706459, NCT05741710) had reported since the build and is the only PTSD drug trial in this base that separates a rapid-acting drug effect from a therapy package; added to pharmacotherapy, MDMA and trials pages and made the basis of new question OQ-17.
  • NICE NG116. The registry recorded the April 2025 surveillance as current. A live re-fetch of the NG116 evidence page found a July 2026 exceptional surveillance listed as well. Its decision document returned HTTP 403, so the outcome is recorded as unresolved rather than assumed.
  • Harper 2026 direction. The build cited the CPT-versus-PE dropout trial four times without ever stating its result. The direction (PE 52.31% vs CPT 45.77%, separating after session 3) is now given on all five pages that use it.

D. Overstatements softened. Three claims went beyond their sources and were rewritten to what the source shows: the "diagnostic labels may feel reductive" claim on patient-experience-and-advocacy.md (Beck 2016 is a metaphor analysis; Benfer 2023 is a conceptual Open Forum — both now described as such); the moral-injury treatment-comparison claim, where Dobos 2023 (PMID 36630050) is a bioethics argument with no PubMed abstract and is now labelled as an ethical argument rather than evidence of effect; and the blinding claim on the MDMA page, moved from a two-page JAMA Viewpoint to the 2026 state-of-the-science review that enumerates the five methodological debts.

E. Metadata. One reference and one ledger row carried the literal placeholder "[Not Available]" as the title of PMID 37615227 (a Danish-language review); replaced with the record's actual English title.

Searches re-run during the audit (live PubMed E-utilities and ClinicalTrials.gov v2, 2026-09-02)

Psilocybin × PTSD (all designs, and restricted to RCTs); MDMA-assisted therapy 2025–2026; ketamine × PTSD randomized trials; imagery rehearsal therapy and CBT-I for trauma-related nightmares; prazosin responder/predictor/biomarker (plus a registry sweep of all prazosin × PTSD studies); CPT telehealth and massed formats; massed and intensive-outpatient PE; PTSD dropout predictors and attrition meta-analyses; sertraline/paroxetine placebo-controlled PTSD trials; pediatric PTSD pharmacotherapy; ITQ/CPTSD randomized trials; eye-movement dismantling and dual-task working-memory studies; pre-trauma and pre-deployment prospective cohorts; inhibitory-learning exposure literature; PTSD × SUD psychological interventions; telehealth acceptability and barriers; VA/DoD 2023 guideline synopsis. Individually re-fetched NCT03537014, NCT03539614, NCT03711266, NCT04007666, NCT04077437, NCT04560660, NCT04959695, NCT05341882, NCT05554094, NCT05560854, NCT05741710, NCT06549049, NCT06795659, NCT07430657 and NCT00532493.

Content added because the audit exposed a hole, not because it was asked for

  • epidemiology-and-risk.md had no population prevalence figure at all. Added US lifetime 6.1–8.3% and global 3.9% (Flesaker 2026, PMID 42089634), plus perinatal community-versus-high-risk strata that demonstrate the denominator problem the page argues for.
  • pharmacotherapy.md had no quantitative SSRI effect. Added the Cochrane numbers (66 RCTs, n=7,442; response RR 0.66, 95% CI 0.59–0.74; withdrawal RR 1.41, 95% CI 1.07–1.87) and the PE-versus-sertraline head-to-head (PMID 30516797).
  • clinical-trials-landscape.md listed registry status without enrollment, concealing that a "completed" esketamine-plus-PE study (NCT06795659) enrolled 4 participants. The table now carries phase and enrollment for all 14 trials.
  • comorbidity.md asserted that integrated and sequential SUD approaches "should be compared" without citing the comparisons that exist (PMID 37971855; PMID 35558682).
  • children-and-adolescents.md had an uncited Medication section and no acceptability data; both now sourced (PMID 32757557, PMID 22040192, PMID 34377359 — pediatric dropout 11.7%, notably lower than the adult figures).

Promotion result

21 of 22 pages promoted to status: curated. Every citation on those pages resolves, matches its record, and supports the claim it is attached to; every quantitative claim was traced to the source text; every asserted absence was re-run and re-dated.

guidelines.md remains draft, for a reason specific to it: three of its load-bearing facts rest on issuing-body web pages rather than PubMed records, and those could not be independently re-verified from this session. healthquality.va.gov failed at the TLS layer; istss.org returned HTTP 200 but renders its content client-side, so the 361-RCT / 208-meta-analysis / 125-recommendation counts could not be read from the raw HTML and remain as the build session recorded them. The VA/DoD guideline itself has been re-anchored to its peer-reviewed synopsis (Schnurr 2024, PMID 38408360), which independently confirms the 34-recommendation figure and the GRADE/PICOTS process, so that claim is now solid. The residual uncertainty is the ISTSS counts and the unknown outcome of NICE's July 2026 exceptional surveillance. literature/guidelines/REGISTRY.md is likewise held at draft and carries both notes inline.

Still flagged for the next sweep

  1. NICE NG116 July 2026 exceptional surveillance — retrieve the decision document. Highest priority; it may change a first-line recommendation.
  2. NCT03539614 — completed 1 January 2026, unpublished. Its result decides OQ-4 (prazosin responder phenotype) and either resolves or entrenches the prazosin contradiction.
  3. ISTSS 2018 counts — re-verify from a source that is not client-rendered, or replace with a PubMed-indexed equivalent.
  4. Two organization URLs — combatstress.org.uk (HTTP 403, Cloudflare) and the canada.ca PTSD portal (HTTP/2 failure) could not be re-fetched; neither non-response is evidence the resource has changed.
  5. APA 2025, Phoenix Australia, RANZCP and WHO registry entries were not re-fetched in this audit; they were verified only by the build session.
  6. Structural note for whoever next builds a condition. Roughly 60% of each page's line count is boilerplate repeated verbatim across all 22 pages (the "Minimum extraction frame", "Claims this page does not make" and "Evidence-updating triggers" tables), and most pages' citation counts are carried by an Evidence-ledger table that restates each record's title rather than making a claim about it. Median inline-cited records per page is 10 against 30 reference lines. The citations are real and the ledgers are accurate — this is a density problem, not an integrity problem — but a future build should spend those lines on content instead.