Research and evidence change log

Renderer, navigation, build and service history is recorded separately in the technical change log.

23 July 2026

Completed a second-pass maturity review against registered evidence-mapping, living-review, data-catalogue and government user-needs guidance (SRC-031, SRC-037, SRC-284 to SRC-286). The resulting changes improve disclosure and navigation but do not relabel the project as systematic, exhaustive, living or DCAT-conformant.
Split the complete report catalogue from a generated corpus profile, and added a generated selection/coverage flow that shows the 165 retained candidate decisions separately from the 379-row 2018 publication-cohort overlay and its 343 missing contextual dispositions.
Added a generated evidence-assurance and uncertainty view that names qualified and medium-confidence rows, response-status gaps, selective claim coverage, candidate requirement/test status and open validation gates without creating a quality score.
Registered ten bounded counter-position records (DIS-001 to DIS-010) covering compatible clarifications, material disagreements, a remedy counter-risk and a documentary anomaly. Every record resolves to the row-linked primary and response sources; none adjudicates by preference or treats respondent action as implementation evidence.
Added stable per-mechanism report lists and finding-to-crosswalk routes, restructured the generated Defence crosswalk into a compact summary plus finding-specific sections, and added generated report footers so readers can continue from an individual report to corpus, uncertainty and map views.
Added a living update/release policy and a generated data/citation page with canonical downloads and SHA-256 checksums. No source-content licence, DOI-backed archive, continuous surveillance or fixed update cadence is claimed.
Added a reader-facing governance, privacy and corrections page that labels unpublished ownership/contact, contributor-role, funding, competing-interest and institutional-review declarations explicitly rather than implying absence; it also states the public-record/redaction and Ask-the-Wiki data boundaries.
Simplified the Evidence-base landing page, introduced overview routes for every grouped Methods section, and excluded legacy compatibility routes from Ask-the-Wiki retrieval so moved pages do not compete with canonical content.
Performed a bounded post-expansion saturation review of the seven Findings pages. Added purposive anchors from the 41-report update for inclusive triage questions, deployed alarm configuration, accurate-but-unacted-on EPR content, unresolved inter-agency work, within-Trust record fragmentation, operational location identity, retrospective-entry visibility, qualitative audit loss and connectivity-dependent recording. These are named mechanism/variation anchors rather than a claim to list every coded report.
Preserved conflicting primary accounts rather than resolving them by preference: Melanie Walker distinguishes a local configuration from the supplier default, and Alan Mitchell retains the coroner/supplier disagreement while recording that the medication issue was not causal in that death.
Added a generated response-evidence route and a finding-to-requirement-to-test crosswalk from the canonical included-case, source, claim, requirement and assurance-test records. Recipient statements remain separate from coronial findings, and the existence of a response is not implementation or effectiveness evidence.
Added an explicit AI-assisted workflow and human-accountability disclosure. Generated material remains non-evidential, source registration remains required, automated checks are not described as an independent second review and public redactions remain preserved.
Added visible evidence cut-offs to the Outputs overview so the 15 July BMJ dossier is not mistaken for the 22 July current-corpus synthesis.

22 July 2026

Registered the supplied 400-page Reg 28 PFD Reference material.docx as SRC-169 by SHA-256 checksum and treated it as secondary synthesis/discovery material rather than source evidence.
Rendered and visually reviewed all 400 pages, extracted approximately 97,700 words, inspected the Word package and recorded the document structure, author metadata, link coverage and AI-discovery URL parameters in a durable reconciliation audit.
Reconciled the document's 60 case profiles bidirectionally against the then-current 71-report corpus. All 60 nominal profile identities/headings mapped to included reports, although two narratives did not match the named cases; the 11 Wiki-only reports each resolved to a registered primary Judiciary source, so their absence from the document did not support a hallucination inference.
Reopened substantive conflicts at the primary Judiciary source. Rejected the document's wrong-case accounts for Darren Carrington and Valerie Gibson, reconfirmed the Wiki's existing EPR mechanism for Marian Grant, retained the respondent counter-position for Theo Tuikubulau and preserved the open Cuthbert Hingert identifier discrepancy.
Corrected two canonical coroner-area metadata errors: Sebastian Daniels to Hampshire, Portsmouth and Southampton, and Jennifer Trigger to North Wales (East and Central). Primary rechecks confirmed the existing Wiki metadata for Greg Hutchins, Steven Welch, Brian Bicat, Jacob Sulaiman, Akhona Moyo, John Spencer and Paula Doreen Hughes.
Added a reader-facing supplied-document audit and an eight-question information-lifecycle worksheet. The latter is explicitly a prospective lens cross-walked to the existing nine codes; it does not create a second canonical taxonomy or recode reports on the document's authority.
Opened independently of the supplied document and registered 11 official recipient-response attachments across six existing reports (SRC-170 to SRC-180). The official Ethan Hanson page supplied a third response omitted from the reference document; every response remains a bounded respondent account rather than evidence that action was implemented or effective.
Completed a reproducible update screen of all 581 Judiciary PFD pages published from 15 July 2025 through 22 July 2026. Eighty-two high-specificity full-text hits yielded seven existing-log duplicates, 38 inclusions and 37 exclusions; three supplementary eligible reports were also integrated. The exact official API boundary, result hashes, terms, dispositions and limitations are retained in research/audits/judiciary-update-screen-2026-07-22.md.
Split primary-source QA rechecked all 41 update additions and corrected setting, recipient, formal-concern, causal-scope and metadata boundaries, including the prospective rather than operative ePRF issue in Colin Brown and the distinct Category 3 error/protocol ceiling in Miles Robinson. This internal QA strengthens the extraction but does not close the independent second-reviewer gaps.
Opened every official attachment then published on the 41 new report pages. Registered 61 response letters across 39 reports (SRC-222 to SRC-245 and SRC-247 to SRC-283) plus one earlier NHS England supporting circular (SRC-246), producing 107 mapped response letters across 63 reports overall. The two pages without located responses remain explicit gaps, and respondent counter-positions are retained without treating proposed action as implemented or effective.

20 July 2026

Implemented the approved second Cloudflare deployment phase as a bounded Ask Wiki tool. The public Zensical search index remains the only runtime retrieval corpus; the generated answer is explicitly non-evidential and must be checked against retrieved pages and their registered primary sources.
Adapted the InterSystems shared-password interaction pattern for Regulation 28. The final design requires the human password for each question, clears it after submission and creates no session. The Pages Function fails closed and verifies an HMAC digest with a separate server-only 64-character pepper, constant-time comparison, fixed-model and request caps, same-origin client requests and no-store responses. The plaintext password is not repository or Pages configuration data.
Added clinical-safety prompt boundaries that preserve the source-to-assurance authority chain, distinguish coronial findings from responses and Defence translations, and prohibit patient-specific advice, unsupported DMICP/product causation and prevalence or trend claims from corpus recurrence.
Confirmed against current Cloudflare documentation that Pages Functions support the Workers AI binding, encrypted secrets are not readable after creation, Wrangler deployment includes the root functions/ directory, Workers AI local inference incurs usage, and static routes should be excluded from Function invocation. No new source record was created because these are deployment controls rather than study evidence.
Runtime validation established that Pages projects reject the Workers ratelimits configuration field even though Wrangler's general schema and type generator recognise it. The unsupported binding was removed before secret creation or deployment; no per-user or WAF rate limit is claimed.
Recorded the residual access-control limitation explicitly: the shared human password is guessable, and the server-side pepper protects the stored verifier rather than preventing online attempts. Identity controls, WAF rate limiting and per-user quotas remain unimplemented future controls rather than present safeguards.

14 July 2026

Received the initial scope and five named examples from the project sponsor.
Verified primary Judiciary pages for Stephen Rhodes, Rachelle Ross, Alexander Braund, and Morris Reddington.
Separated verified report content from sponsor-supplied discovery claims.
Ingested the ICMM 2026 project brief unchanged into research/briefs/ and created the rapid-review workspace, protocol, coding framework, datasets, abstract, oral outline and submission checklist.
Checked the current official ICMM event page, guidelines and parent-congress page. Recorded a live deadline conflict: 3 August 2026 on the event page versus 14 July 2026 on the parent page.
Confirmed the statutory distinction between the reporting duty in Schedule 5 paragraph 7 of the Coroners and Justice Act 2009 and the report/response procedure in Regulations 28–29.
Logged 50 unique PFD candidates and completed full-schema extraction for an initial 30-case focused series; generated non-exclusive code counts directly from the CSV rather than hand-maintaining totals.
Corrected Ross: the missing input was national non-responder status, while vendor responses said due/outstanding smear alerts existed.
Corrected Braund: the live SystmOne audit trail retained amendment metadata and prior text; the problem described was visibility in the printed/editable summary used for investigation.
Verified the provisional Derby case as Stephen Stringer, PFD 2024-0555, and avoided attributing the incident to a supplier not named in official materials.
Extracted the Reddington joint response and separated its proposed FHIR delivery remedy from the coroner's finding.
Added Eric Thompson, Stephen Cassidy, Samuel Jordan and Audrey King as representative narrative cases and linked the wider dataset to official Judiciary sources.
Ran a scoped literature-orientation search. Located peer-reviewed PFD collection/thematic-analysis precedents, a 2025 military EHR systematic review and an EHR-interoperability incident analysis; recorded the absence of a located digital-PFD synthesis as a search limitation, not a novelty claim.
Separated a Welsh Government claim about growth in all PFD reports from the unproven proposition that digital-health PFD reports are increasing.
Resolved all 17 eligible-pending candidates against full official reports and available recipient responses. The current pass now contains 47 fully extracted reports from 50 candidates, with 3 protocol exclusions and no unresolved screening decisions.
Populated the working abstract with the current-pass report and non-exclusive theme counts, retained explicit single-reviewer/corpus limitations, and added automated title, body-word, keyword and placeholder checks.

15 July 2026

Reconsidered the information architecture against the 47-report corpus, current traceability, official evidence-synthesis guidance, patient-safety thematic-review practice and Zensical capabilities.
Adopted a reader model based on evidence base, findings, Defence application, methods and outputs; classified conference and technical materials as derivatives rather than peer evidence domains.
Registered methodological and information-architecture sources SRC-030 to SRC-037, including JBI, PRISMA, NHS England thematic-review guidance, Judiciary search guidance, evidence-mapping practice and living-review context. These guide transparency and labelling; they do not convert the project into a systematic or scoping review.
Relabelled the product as a targeted evidence map and thematic case synthesis, retained the focused-series boundary and documented explicit inductive, deductive, context-facet and Defence-translation layers.
Added controlled primary workflow, setting, boundary, verification and response-status facets to all 47 included rows without replacing their detailed extraction or non-exclusive mechanism codes.
Generated a complete all-report catalogue, 47 individual report records and a case–mechanism map from the canonical included-case dataset. Empty map cells are explicitly not labelled evidence gaps.
Replaced the former product-oriented theme layer with six mechanism-first findings and a cross-cutting human-factors lens, each recording recurrence, variation, counter-considerations and limits.
Added CLM-015 for Audrey King's hybrid paper–electronic medication workflow, so every representative case now has a stable claim trace.
Established 13 candidate Defence requirements (REQ-001 to REQ-013) and 13 designed assurance scenarios (AST-001 to AST-013). No requirement is validated for Defence and no scenario has been executed.
Separated transferability limits and 12 DMICP evidence gaps from civilian findings; retained wrong-practice population invisibility as an explicit Defence hypothesis rather than a system finding.
Registered recipient-response provenance one document per stable source ID. Thirty-five known official attachments are now row-linked across 18 reports; each generated report page separates those respondent accounts from the coroner's report and warns that a response does not demonstrate implementation or effectiveness.
Split the previously bundled Ross vendor record into separate TPP and EMIS sources and recorded attachment-specific limitations where a response addresses only another concern, supplies policy context, or reports a recipient's contested position.
Completed a bounded historical 2018 publication-cohort backfill: 379 unique Judiciary PFD pages were enumerated; 121 report PDFs yielded direct text and 258 scanned reports (682 pages) were OCR-processed with no extraction failures. The canonical audit records literal official-query zero results, report-text hit counts and the publication-date/report-date limitation.
Recovered the exact 379-row working cohort index and extraction artefacts, published a canonical CSV manifest with official page/attachment URLs, PDF hashes, extraction routes and base query-family hits, and corrected the evidence boundary: 36 cohort rows have retained contextual decisions while 343 do not, so complete enumeration/extraction is not described as complete row-level screening.
Integrated 24 eligible reports dated in 2018, including Karl Willis as an online-prescribing workflow case and Mavis Reeves as a deliberately bounded telecare/health-IT design case. Four pages without a displayed official reference received transparent UNREF-2018-… project identifiers.
Promoted eight strongly matched 2018 report claims into the controlled claim-to-requirement trace, replacing REQ-008's indirect Stringer mapping and retaining REQ-012 as a partial scope extension because offline synchronisation remains unsupported.
Retained eight report-dated-2018 exclusions and five mechanism-relevant reports dated in 2017 as explicit out-of-scope screening decisions. Recorded the later-migrated 2018-0405 report and opened rechecks for other later migrations and the Cuthbert Hingert reference discrepancy.
Queried the canonical graph and five sibling-project evidence trails before extending the DMICP analysis. The internal audit is retained at research/audits/dmicp-sibling-audit-2026-07-15.md; no row-level practice, facility or population data were copied into the wiki.
Registered SRC-130 to SRC-135 for current published DCB0129/DCB0160 and patient-safety context. Distinguished patient-safety learning from DCB clinical-risk management, and recorded that the 2018 DCB releases remain current while NHS England's 2026 national review is in progress.
Cross-walked the 71-report corpus from patient-safety workflows to the existing non-exclusive clinical-safety mechanisms, stable CLM- claims, candidate REQ- requirements and designed AST- scenarios. Counts remain recurrence within this targeted dataset, not prevalence, trend, causation or product comparison.
Corrected the DMICP system description after sibling and primary-source review: official sources support a long-running fixed/deployed Defence EPR; CGI historically described a hosted EMIS PCS-based service; no reviewed public evidence independently proves the current build/topology or a quantified four-nation national-service limitation.
Extended the open-source architecture search after the sponsor described DMICP as an on-premises-hosted, EMIS PCS-based solution within the MCN/Defence environment. Registered current 2025/26 DMICP use, the official 2019 legacy EMIS DPCS platform notice, historical CGI central secure hosting, MODNet access and the public procurement definitions of MCN/MODNet.
Recorded the sponsor's subsequent clarification that MCN means MOD Core Network and that “on-premises hosted” means a centrally hosted DMICP server estate inside Defence/MCN infrastructure, with medical centres and users accessing through MCN/MODNet client terminals or devices—not practice-local servers. This resolves the terminology and intended topology distinction in the supplied context; no located public source independently confirms the complete current topology or exact DMICP-to-MCN path.
Recorded the sponsor's confirmation that the live DMICP clinical application is EMIS PCS/DPCS. This promotes the current product family from historical lineage to sponsor-supplied operational context, while leaving the exact live version, Defence adaptations, component baseline and independent architecture confirmation open.
Recorded the sponsor's current national-service boundary: e-RS is separate browser access to an HSCN/Spine web application rather than DMICP integration; EPS, GP Connect, NHS App, Summary Care Record and the Scottish, Welsh and Northern Irish national platforms are unavailable to Defence. Retained PDS as the documented current DMICP integration and pathology as dated interface evidence. Isolated local systems/manual routes are not used to imply meaningful national-platform availability.
Closed the sponsor clarification sequence with DMICP's clinical-safety regime recorded as unknown: the sponsor cannot presently confirm DCB0129/DCB0160 or a Defence equivalent, a named Clinical Safety Officer, Hazard Log or Clinical Safety Case. No inference of absence, inadequacy or non-conformity is made.
Replaced the earlier broad connectivity formulation with a bounded integration/access account: registered Spine PDS integration, dated pathology-report messaging, browser-only e-RS access, the named unavailable national services, the paper-based NHS–DMS transfer limitation and dated CQC workflow evidence. Brawdy also records age-related DMICP slowness, outages and newer-technology compatibility as widely affecting DPHC. The CQC material is labelled as inspection evidence under the DMSR-invited programme; isolated examples evidence local work as done but are not treated as national-platform availability, a representative comparison, DCB assessment or complete interface inventory.
Triangulated the PFD-derived hazard questions with CQC's 2022/23 and 2024/25 Defence medical-centre findings, preserving their inspection/date boundary and avoiding a DCB-conformity or product-causation inference.
Added a two-page thematic output under Outputs that maps PFD evidence to DCB0129/DCB0160 controls and to bounded DMICP/CORTISONE assurance questions. It remains derivative of the canonical evidence, claims, requirements and tests.
Developed a BMJ Military Health Analysis article on foreseeable harm pathways in UK Defence digital care. Registered current journal-formatting and AI-disclosure guidance (SRC-159 to SRC-161) and additional sociotechnical, interoperability and Defence-specific literature (SRC-162 to SRC-168).
Kept the manuscript below the current 2,000-word Analysis limit with a 250-word unstructured abstract, the required three-part key-message box, one table and 30 Vancouver references. It does not publish the 71-report project corpus as original results.
Framed six safety-critical seams—visibility, work ownership, interoperability/identity, integrity/provenance, resilience and governance—and translated them into an assurance agenda for DMICP coexistence and the CORTISONE transition.
Preserved the publication boundary: public coronial and CQC evidence identifies foreseeable scenarios, not DMICP-attributable deaths, product-wide defects or DCB non-conformity. The present DMICP clinical-safety regime remains unknown, not absent.
Added a transparent BMJ-compliant AI-use statement and recorded outstanding human authorship, institutional review, competing-interest and MOD publication/security-clearance requirements.
Added a reader-facing BMJ publication dossier: an evidence/argument map covering all 30 references, a public/internal/unknown DMICP baseline, a seven-case PFD crosswalk, methods and limitations, and a submission-readiness gate tied to the current Word checksum. The raw sibling audit remains repository-only and SRC-005 remains author-supplied context rather than public authority.
Manually recoded all 71 included reports using an orthogonal care-pathway schema. Each row now records a unique non-exclusive participant set, information/work origin and expected-action destination, non-exclusive origin/destination care contexts, care-setting span, sector boundary, failure point, classification confidence and a report-specific rationale.
Split care-setting span from sector boundary after a structured QA challenge showed that the earlier single scope field conflated cross-setting, cross-sector, patient-to-service and national-to-local pathways. Added an explicit primary/secondary care-context roll-up while retaining diagnostics, ambulance/urgent response, pharmacy, secure care and private care as distinct contexts when a primary/secondary assignment would be artificial.
The current descriptive recoding identifies primary care among the information origins in 14/71 reports, secondary care in 28/71 and multiple care settings in 45/71. Origin-context counts are non-exclusive and all aggregates remain recurrence within the targeted dataset, not incidence, prevalence, causal weight or service-performance comparisons.
Regenerated the complete catalogue, case–mechanism map and 71 report pages from the canonical CSV. Classification-specific confidence is high for 67 rows and medium for four; the work remains single-reviewer coding pending the independent check and adjudication recorded in VAL-020.