The cycle provenance page now opens with an at-a-glance panel: the current cycle status, the next scheduled run, your cadence and alert settings, how many new studies are waiting to be reviewed, and the most recent evidence-change alert — with the cycle timeline and frozen snapshots below.
Faster to act on a change
When a cycle re-pools the evidence and a result crosses your alert thresholds, the "Review evidence change" prompt now links straight to that cycle on the timeline, so you land on the before-and-after trajectory rather than a bare results page.
Project cards on the dashboard now show a living-review status chip — whether the latest cycle is running or has failed, and how many new studies await review — so you can see which reviews need attention without opening each one.
Clearer recovery and setup
A failed cycle now explains the problem in plain language and offers a one-click retry when the failure is transient. The automation dialog links straight to search settings when queries still need locking, and its alert-threshold controls have plainer descriptions.
Living reviews keep a complete, per-cycle provenance trail
Every cycle is traceable on its own
Each scheduled cycle now keeps its own record: the extraction work behind it, which pooled outcomes were checked, which (if any) tripped an alert, and the rationale for the verdict. This is saved even for cycles that find no change, so "nothing moved" is a recorded fact rather than a gap.
Roll up and freeze for audit
The provenance of a whole living review — its cycle timeline, each cycle’s results and verdict, and every monitored outcome’s trajectory — can be assembled into a single view and frozen into an immutable snapshot suitable for an audit trail or a health-technology-assessment submission.
Read-only demo account — explore the full app without signing up
One-click access from the sign-in page
The sign-in page now has a "View live demo" button that signs you into a shared, read-only demo account — no email, password, or sign-up.
Explore everything, change nothing
A demo account can browse every page — analyses, screening, extraction, statistical results, forest plots, GRADE tables, and reports — exactly as a normal user sees them. It cannot start new work, edit protocols, upload documents, or change settings, and a read-only indicator stays visible throughout.
Large living reviews now finish their cycles reliably
Big reviews no longer stall mid-screening
A living review with hundreds of studies could previously stall partway through screening, so the cycle never reached extraction or the evidence re-pool. Screening now runs in parallel and each cycle is more resilient — cycles get through and resume cleanly after an interruption.
Sharper living-review change detection
Changes that arrive without new trials are no longer missed
An evidence-change alert no longer requires a new trial. A study retracted, excluded on re-screening, or re-extracted with corrected numbers now also flags — and, regardless of trial count, an outcome flags when its confidence interval crosses the null or the effect moves sharply.
Symmetric thresholds for ratio measures
For risk, odds, and hazard ratios, an increase and the equivalent decrease now count as the same size of change, so the alert threshold behaves consistently in both directions.
Fewer false alerts, clearer emails
A transient glitch can no longer cause a cycle to re-pool by the wrong method and send a spurious alert. Evidence-change emails now open consistently with "Evidence change detected", and an outcome entering monitoring for the first time is recorded in the activity log.
Living reviews pool evidence with your locked analysis method
Slow, cumulative evidence drift is now caught
Change detection now compares each outcome against the last cycle that alerted on it (or its first snapshot), so a slow drift that arrives in small weekly increments accumulates against a fixed anchor and trips once the cumulative move crosses your threshold.
Lockable analysis recipes
From a completed forest analysis you can now lock that outcome’s method — model, effect measure, timepoint, and study exclusions — as a living-review recipe. Future cycles replay that exact method, so the monitored estimate is computed the way you decided rather than by a default. Outcomes without a recipe use a per-outcome model setting.
Recipe management
A new Recipes page lists every monitored outcome with its locked method, lets you set the fallback model for outcomes without a recipe, and flags recipes whose outcome was renamed or removed so they are easy to clean up.
Consistent pooling everywhere
Every living-review cycle now pools each outcome with the same engine and method as interactive statistical analysis, so a change across cycles always reflects new evidence rather than a difference in method.
Trace Lineage — follow any pooled estimate back to its studies and source PDFs
Data lineage for every analysis run
A new Trace Lineage panel on the statistics results opens the full provenance trail for any pooled result: the contributing studies, each study’s effect estimate and variance, and the extracted values and source-PDF snippets behind them. It is available for every completed run.
Pooled-result cards now list their studies
The model-fit summary for a pooled analysis now lists the studies that entered the pool, so you can confirm at a glance which evidence the estimate is built on.
Forest-plot labels restored
Forest plots were rendering without their study labels, axis values, and column headers; they now render with every label, tick, header, and study weight visible.
Living reviews now detect when new evidence shifts the pooled estimate
Automatic change detection
After each scheduled cycle, the meta-analysis is re-run on the now-fuller evidence and compared with the previous cycle. When the change crosses your configured thresholds — effect-size change, confidence-interval flip, certainty downgrade, or heterogeneity jump — an alert fires immediately and the analysis overview shows an evidence-change card with a before/after table.
Tunable, conservative defaults
Thresholds are set per analysis in the automation dialog, with conservative defaults so first-cycle and empty-cycle runs never alert. The alert email’s narrative is AI-written, but the decision to alert is deterministic — the AI only ever explains an alert that has already been decided.
GRADE Summary of Findings: automatic certainty with reviewer finalisation
Every run produces a Summary of Findings table
When a statistical run completes, a Cochrane GRADE certainty assessment is derived automatically for every outcome it produced, and the Summary of Findings page renders them as a table with per-factor severity. The certainty derivation is rule-based — no AI in the certainty path — and risk of bias counts only reviewer-validated judgements, so unreviewed AI proposals cannot drag certainty down.
Reviewer finalisation with per-factor override
Open any row to review it: each downgrade factor has its own severity control and rationale field, with the underlying numbers shown, and the three upgrade factors can be recorded too. Final certainty is recomputed, and rows move through a clear draft → under review → finalised lifecycle.
Export and reproducibility
Two HTML export buttons produce self-contained, submission-ready Summary of Findings tables for NICE / HTA appendices. Approving a sign-off freezes every GRADE and risk-of-bias row as an immutable snapshot, and automatic derivation can be turned off per analysis.
Risk of Bias 2.0: automated assessment with dual review
A Risk of Bias stage in the pipeline
The pipeline now runs a Risk of Bias stage after extraction. An AI assessor reads the full text and extracted outcomes and answers the Cochrane RoB 2.0 signalling questions with PDF evidence. The domain judgement itself comes from the deterministic Cochrane decision tables — not the AI — and a human reviewer owns final validation. Both Domain 2 variants (effect of assignment and effect of adherence) are supported.
Dual review for quality judgements
A second reviewer can agree with or override the first reviewer’s domain judgements in a new Risk of Bias review queue, using the same blinded-card pattern as extraction review. Disagreements surface in the unified conflict gate with a per-domain diff.
Risk of Bias matrix
A new Risk of Bias page shows the per-study, per-domain matrix with traffic-light colour coding. A "Generate RoB proposals" button runs assessment across all eligible studies, with a live progress card.
Conservative defaults
Risk of Bias is off for existing analyses and on by default for new ones. Only randomised trials are assessed — the AI skips studies it determines are non-randomised — and every assessment keeps a full reasoning and validation trail.
Living-review foundations: progress monitor, settings, and safeguards
Pipeline progress is visible
While a living-review run is in flight, the analysis overview shows a progress banner with the current stage and per-stage counts, and stops once the run finishes. A failed run shows a red banner with the reason, so you no longer have to dig through logs.
Settings persist correctly
The Living-Review Automation dialog now reads its real saved state, so reopening it no longer silently reverts your cadence and on/off choices.
Cost and permission safeguards
Scheduled runs are checked against usage limits at every stage, require a locked protocol, and skip soft-deleted analyses, so an in-flux protocol or an exhausted quota cannot trigger a paid run. Manual "Run search now" requires edit-protocol permission, and a cleanup job clears any run stuck for over two hours so one crash cannot block future runs.
Notifications and inbox
Living-review emails now route through your notification preferences and digest cadence, while failure alerts always ship immediately. A "Mark visible as read" button clears the new-study badge without forcing an accept/reject decision.
Team plan: per-seat review allowance raised from 5 to 7
More headroom on every seat
Each Team seat now includes 7 pooled reviews per month, up from 5 — so the 3-seat minimum now covers 21 reviews a month. Base and per-seat pricing are unchanged.
Pricing page: live seat-to-review readout
Clearer Team-plan pricing
The Team-tier card now shows a live "reviews per month" total that updates as you move the seat slider, and a capability strip below it summarises what the Team plan includes — roles, comments, sign-off, notifications, conflict adjudication, and audit-ready exports.
A dedicated For Teams page on the marketing site
Team features in one place
A new For Teams page brings the team product together — roles and permissions, comments and @mentions, notifications, two-person protocol sign-off, workload assignments, conflict adjudication, and versioned protocols. The home and product pages were updated to surface team collaboration alongside it.
Extraction quality: more numeric formats and clinical-scale synonyms
More numeric formats extract successfully
Continuous outcomes reported as median + IQR or median + range now extract end-to-end. Full modified Rankin Scale (mRS) distributions extract — both dichotomised splits and the seven-point ordinal scale — so stroke reviews no longer lose the disability outcome. Event counts extract even when the trial gives no denominator.
Outcomes match clinical scale names
A generic outcome like "anxiety", "depression", or "stroke disability" now matches text that only names the measurement instrument — HADS, GAD-7, PHQ-9, mRS, NIHSS and similar — so reviews that named outcomes generically stop losing trials that used only the scale acronym.
Better recovery when a value is hard to find
When extraction returns nothing for an outcome, it now retries before giving up and falls back to scanning the full Results section. Tables in PMC full-text are read as structured tables, so numbers reported only in tables extract reliably. Every miss records the exact text the system saw and why nothing came out.
More reliable full-text recovery
Better recovery of blocked open-access papers
Open-access recovery now paces its requests so it no longer trips publisher bot-protection, and it can retrieve papers through an alternative retrieval route when the usual full-text endpoints are blocked — making recovery both more reliable and faster.
Manual-upload fallback when a publisher blocks download
Any study whose PDF a publisher blocked now shows a clear notice on its Full-text tab, with direct links to the publisher page, PubMed, and Europe PMC, plus drag-and-drop upload. A new bulk-upload page lets you drop a folder of PDFs and have them matched to the blocked studies automatically.
Home and Product pages refreshed
A cleaner marketing site
The home and product pages were tightened — clearer calls to action, fewer redundant sections, and a single comparison view. Technical backend details were removed from marketing pages in favour of plainer language.
Pricing page redesigned
A clearer pricing page
The pricing page was redesigned around a simple idea — you pay for reviews, not tokens. Each plan shows what a single end-to-end review includes and when it counts toward your allowance, alongside a trust strip (EU hosting, audit trails, bring-your-own-key on Enterprise). Checkout and Team seat pricing are unchanged.
Resources page combines the blog and library
Blog and library in one place
The Resources page now presents the blog and the documentation library side by side, with the library grouped into Getting started, Worked examples, and Methodology.
Captcha-blocked full text now falls back to open-access mirrors
Automatic open-access fallback
When a publisher page is bot-gated, full-text retrieval now checks open-access mirrors — PMC, Europe PMC, Unpaywall, and OpenAlex — before and instead of the publisher, and only reports a paper as blocked when every open-access source is also unavailable.
Add known-relevant papers as search pearls
Seed studies (pearls) on the Search page
You can now add up to five known-relevant PubMed IDs as "pearls" on the Search page. Each accepted ID shows as a chip; invalid entries are flagged and ignored.
Pearls strengthen your query
When you next build queries, each pearl’s subject headings and author keywords feed query enrichment, and every pearl ID is added to the final PubMed search so it cannot be missed. Pearls drive enrichment — distinct from guaranteed-recall seed studies.
Stronger queries for Europe PMC, ClinicalTrials.gov, and OpenAlex
Database-native query syntax
Search queries are now translated into each database’s own syntax instead of reusing PubMed’s — Europe PMC, ClinicalTrials.gov, and OpenAlex each get a properly formed query — so recall no longer collapses on databases that index PubMed-style tags differently.
More robust query generation
Malformed ClinicalTrials.gov queries are cleaned up automatically, OpenAlex queries always carry sensible type, language, and year filters to avoid a preprint flood, and when the AI omits a per-source query the system falls back to a built-in builder so every enabled source is still searched.
AI search strategy now covers Europe PMC, Cochrane CENTRAL, and OpenAlex
One click, multiple tuned queries
The "Generate strategy (AI)" panel adds an "Extra sources" option: opt into Europe PMC, Cochrane CENTRAL, and/or OpenAlex and the AI produces an optimised query for each, alongside PubMed and ClinicalTrials.gov.
Each database queried the right way
Cochrane CENTRAL automatically wraps your PICO terms in the Cochrane Highly Sensitive Search Strategy, OpenAlex uses structured free-text with year, type, and language filters, and Europe PMC accepts PubMed syntax. Each query gets its own preview card with a Test button showing real volume before you run it. PubMed and ClinicalTrials.gov are unchanged when extras are off.
Search-strategy generator moved to the Search page
Build and refine queries where you run them
The "Generate strategy (AI)" panel — search-breadth modes, the outcomes-in-search toggle, query testing, and the structured-plan editor — now sits at the top of the Literature Search page, right above the query box. The Configure page still holds the protocol.
A faster path to a first query
Analyses that go straight to the Search page get an inline plan form — Population, Intervention, Comparator, Outcomes, year range, study design — so you can fill it, click Generate, and have the queries appear in the search box ready to run or edit.
Discover related papers — citation chasing via Semantic Scholar
Find what your Boolean query misses
A new Discover page lets you pick known-relevant papers as seeds and walk the Semantic Scholar citation graph from them — references, citing papers, both, or recommendations for the seed set. Candidates are ranked by how many of your seeds they connect to, so the papers bridging your evidence base surface first.
A Cochrane-recommended technique, automated
Citation chasing — backward and forward — is recommended in the Cochrane Handbook; done by hand it means clicking "cited by" for every included paper. Here it is one click per analysis, and selected papers flow into screening through the same de-duplication and provenance as every other source.
MAUDE, ERIC, and Cochrane CENTRAL added to Additional Sources
Three more free databases
Additional Sources adds three free databases: MAUDE (FDA medical-device adverse-event reports), ERIC (~2M education research records), and a Cochrane CENTRAL approximation that wraps your PubMed query in the Cochrane Highly Sensitive Search Strategy.
Pearl growing
A new Pearl Growing tab lets you paste 1–10 known-relevant PubMed IDs; the system fetches their subject headings and author keywords and surfaces terms shared across several pearls as ready-to-paste query additions.
Search beyond PubMed and ClinicalTrials.gov
Additional sources
A new Additional Sources panel lets you search free providers alongside PubMed and ClinicalTrials.gov — OpenAlex (250M+ works) and Europe PMC (~40M records). Results are automatically de-duplicated against studies already in the analysis, and each study records which sources it was found in for PRISMA reporting.
Refinement suggestions
A new Suggest button decomposes your query by PICO, runs the Population, Intervention, and Outcomes clauses separately to show how many records each matches, flags the most restrictive clause as the bottleneck, and offers copy-to-paste refinement chips.
Export to RevMan 5
Cochrane authoring handoff
A new "Export RevMan (.rm5)" button downloads the full analysis — included and excluded studies with reasons, and extracted outcomes — as a RevMan 5 file.
Outcome types mapped correctly
Dichotomous, continuous, and time-to-event outcomes each map to the matching RevMan outcome type. Single-arm measures, which RevMan cannot represent, are listed in the published notes rather than dropped silently, and risk-of-bias judgements are left to be completed in RevMan.
Import and export citations in RIS and BibTeX
Seed a review from your reference manager
A new "Import RIS / BibTeX" button accepts files exported from EndNote, Zotero, Mendeley, PubMed, Embase, Web of Science, and Ovid. Files are previewed before anything is saved, and duplicates of studies already in the analysis are flagged.
Export included studies
New "Export RIS" and "Export BibTeX" buttons download the included-study set for import into a reference manager or a LaTeX manuscript, preserving full bibliographic detail and identifiers.
Sign in with Google or Microsoft
Faster sign-up with your existing work account
New "Continue with Google" and "Continue with Microsoft" buttons on the sign-in page — no extra password or account to create. New accounts start on a free plan, ready to begin a project right away, and email-and-password signup remains invitation-only.
Team collaboration quality-of-life
Choose a workspace for each project
Pick Personal or a team when you create a project, and move a solo project into a team later with a "Move to team" card on the analysis overview.
Assign and track screening work
Split unassigned studies evenly across your team with a named preview before committing. An "assigned to me" filter hides everything else, and in dual-review mode each study card carries a peer-status badge — Agreed, Awaiting peer, or Conflict.
Role-aware workspace
Viewers and reviewers see the Configure chat read-only, with a banner explaining why. Your role shows in the analysis sidebar, admins can change a teammate’s role, and comment threads now sit under each study and each outcome.
Protocol freeze, version history, and workload assignments
Split screening workload across your team
Lead reviewers can assign studies to specific teammates per phase and review slot, so each reviewer sees only their own queue. Assignment is opt-in per analysis, with an auto-balance option and per-reviewer progress bars, and reviewers get an in-app alert when new work is assigned.
Protocol freeze and sign-off
Five milestones — protocol locked, screening complete, extraction complete, analysis frozen, manuscript locked — let a review enforce each stage before the next opens. A lead reviewer requests a sign-off, an admin approves it, and approval snapshots the protocol into an immutable version and advances the lock. Once a stage is locked, further edits to it are refused.
Version history
Every sign-off snapshot is listed in chronological order, and a diff view shows what changed in the protocol, eligibility criteria, and extraction form between any two versions.
Unified notifications with @mention routing
One inbox for every team event
Mentions, invitations, adjudication pings, and the living-review digest now share one notification bell and one email pipeline. @-tagged teammates get both an in-app notification and a digest-eligible email.
Notifications on your terms
Per-user preferences route each kind of notification to immediate email, digest email, or in-app only, with a daily, weekly, or off digest cadence. Invitations and conflict-adjudication requests always send immediately.
Anchored comments and @mentions
Collaboration threads on every object
Leave comments on analyses, studies, individual screening decisions, extracted outcomes, and extraction reviews — every object in the review pipeline has its own thread, with one level of replies.
Mentions and moderation
@mention teammates by name or email; mentions resolve against the analysis’s team, so non-members are dropped. Threads can be resolved and re-opened, authors can edit their own comments, and admins and lead reviewers can remove inappropriate content. Comment access follows the same team-role permissions as the parent analysis.
Team roles enforced across the app
Role-based access control
Team roles are now enforced on every action, not only on whether you can see an analysis. Admins can do everything; lead reviewers can edit protocols, adjudicate, and invite collaborators; reviewers submit screening decisions and extraction reviews; viewers are read-only. Personal analyses are unaffected — a solo owner keeps full control.
Review mode selector — AI as Reviewer A
Choose your screening review mode
A new Review mode selector offers Single reviewer, Dual: AI + me, and Dual: two humans. In "AI + me" the AI pre-screens every study as Reviewer A and you act as Reviewer B; "two humans" disables the AI pre-screen entirely, recommended for Cochrane-style reviews. Existing analyses keep their current behaviour.
Optional unblinding
An optional "Show AI verdict before I decide" setting — available only in "AI + me" mode — reveals the AI’s verdict as you open each study. It is useful for triage, but it defeats independent review for inter-rater reliability, and the interface warns you each time it is on.
Faster statistical analysis, extraction QC, and review adjudication
Statistical analysis
Faster, more reliable meta-analysis — results appear in seconds.
New sensitivity-analysis tools: leave-one-out, a comparison of alternative heterogeneity estimators, and the Knapp-Hartung adjustment.
New publication-bias tools: Egger’s regression, Begg’s rank correlation, trim-and-fill, and funnel plots with automatic small-study caveats for k < 10.
Forest-plot images are now saved and available directly in evidence previews.
Extraction quality control
Automated checks inspect every extraction for common issues — missing effect sizes, impossible or implausible values, and subgroup-vs-overall confusion — and re-run extraction with targeted hints before surfacing the result.
Improved study-design classification for more accurate arm identification.
Review and adjudication
Per-outcome Accept / Reject decisions replace batch approval, so reviewers choose exactly which extracted outcomes to keep.
The full quality-control history is visible in the review panel, and AI-assisted adjudication recommends the most likely correct value, with a confidence score, when extraction candidates conflict.
Also fixed
Fixed broken Analysis tab and sidebar links, and a login crash when an unrecognised email address was entered.
Batch extraction, supplements, and Search Agent
Batch extraction
Auto-Extract All Outcomes processes every eligible study in parallel instead of one at a time.
Confidence badges (green / amber / red) flag the reliability of each value, a quality dashboard scores each extraction, and a Review Queue collects low-confidence or conflicting extractions for human approval with a side-by-side conflict view.
Companion documents and supplements
Supplementary materials — PDF appendices, Excel tables, Word documents — are fetched automatically and read into the extraction.
The pipeline also discovers companion publications (secondary analyses and follow-ups) for a complete trial narrative.
Full text
Bulk Upload accepts a drag-and-drop of up to 50 PDFs with automatic study matching, and a fallback search checks open-access sources when the primary PDF is unavailable.
Search Agent
An AI Search Agent builds and runs optimised PubMed and ClinicalTrials.gov queries from your PICO criteria, with real-time precision feedback that screens a sample of results before you commit.
Extraction accuracy
Arm-swap detection flags possibly mislabelled treatment and control arms, and a duplicate hazard ratio is no longer counted across multiple outcomes.