What the product does today, with the date it was checked.
Every public statement about Study Manager traces to an entry in our capability ledger — a verification record kept in the engineering repository and updated by whoever measured the code. This page is transcribed from it. “Verified” means executed through the real entry point on the stated date; “available — shown as steps” means the work exists without a screen that names it; “in verification” means a surface is being built; “planned” means it isn't there yet.
Design and sizing
Effect size, power and sample-size reasoning at design time
Continuous, binary, ordinal and categorical outcomes; two-arm and multi-arm.
Verified2026-08-11
Covariate adjustment (ANCOVA)
Two-arm and multi-arm; ρ stated or computed from imported pilot data; continuous outcomes, one covariate slope shared across arms.
Verified2026-09-15
Multiplicity-aware multi-arm sizing
Exact Dunnett for each-vs-control, Bonferroni for all-pairwise; the Design Assistant asks which comparisons will be reported.
Verified2026-09-15
Unequal allocation
Any positive ratio, two-arm and multi-arm, under every multiplicity plan. Refuses treatment arms that differ among themselves under Dunnett.
Verified2026-09-15
Non-inferiority and equivalence
Means (exact noncentral t; exact TOST) and proportions; two arms. Refuses a non-inferiority claim until direction of benefit is stated.
Verified2026-09-16
Cross-over designs
Closed-form and simulation providers with an agreement check.
Verified2026-08-18
Dose-response trend
Linear trend across dose rank with equally spaced scores — not concentration, not log-dose.
Available — shown as steps
Longitudinal / repeated measures
Simulation-based; two arms only.
Available — shown as steps
Guided, dependency-aware design workflow with a readiness review
Verified
Exploratory vs confirmatory declared at planning (desired evidence strength)
Confirmatory studies must pre-declare the minimum meaningful difference.
Available — shown as steps
Survival / time-to-event outcomes
Registry entry only; no provider.
Planned
Blocked designs as a distinct analysis
A blocking factor is recorded but does not yet change the analysis.
Planned
Conduct
Stratified randomised allocation at enrollment
Seeded permuted block on sex and declared balancing factors; litter and arrival weight checked regardless; sequence recorded, never deleted; refused once data exists.
Verified2026-09-17
Blinding modelled and enforced in the record
Distinct unblinding authority; every Data Collection read surface redacts arm identity until an unblinding record exists.
Verified2026-09-17
Data Collection module
Setup from the approved design; worklists by event; bench and record sheets; barcode identity (USB wedge, device camera, scan station, paired phone); CSV/XLSX and ASTM/HL7 imports through quarantine with UCUM unit conversion; QC queue; amendments with controlled reasons; monitoring sweep. 81 tests passed 2026-09-19.
Verified2026-09-17
Notification delivery (email / SMS)
The sweep records what it found and the screens show it; delivery is not built.
Planned
Assess
Post-collection sizing readout
Did the variance and recruitment assumptions hold; what the next study would need at the observed σ. No "observed power," by decision.
Available — shown as steps
Analysis of the collected data under the designed comparison
Engine-computed run saved with a hash of the rows read; two-arm mean difference today; multiplicity-adjusted and covariate-adjusted analyses of collected data in progress.
Available — shown as steps
Planned → Executed → Observed view
The three-column view is being built; the data exists.
In verification
Document
SEND submission package (SENDIG 3.1.1)
19 domains, Define-XML 2.1, Dataset-JSON 1.1, nSDRG, transformation report, eCTD m4 layout; Pinnacle 21 Community FDA 2508.1 at 0 rejects / 0 errors. No PC/PP, SUPP--, RELREC or pooled findings; validated with the Community edition; feature-gated for the founding cohort.
Verified2026-09-17
ARRIVE-oriented readiness and export
Supports ARRIVE reporting; never "ARRIVE compliant."
Verified
Server-side CSV exports with a declared column contract
Enrollment, data plan, collection status, analysis rows; blinding decided server-side.
Available — shown as steps
GLP compliance, regulatory certification, guaranteed compliance
Not claimed.
Planned
Platform
Role-based access and portfolio views
Verified
Tiered entitlements
The infrastructure is live; a packaged tier ladder is not yet described.
Available — shown as steps
Four constrained AI assistants
Objective composer, design drafter, timeline draft, Stat Assistant — each chooses from a closed list and never supplies a number. Contract tests executed 2026-09-19.
Verified2026-09-19
Expert Guidance rail; expert statistician review
Planned
The six-stage lifecycle shown as a whole in the product
Stages are named on every step page; the ribbon and a per-study stage view are in development.
In verification
Evidence that study integrity was upheld, shown as a whole
The record holds every link; a single surface that shows them together is in development.
In verification
The ledger itself, extracted
Generated from the capability ledger on 2026-09-19 — every entry, its section, the dates it cites, and its first sentence. Rows the ledger has struck through are shown as superseded rather than removed, because the corrections are part of the record.
Safe to claim — verified in code
Effect size, power, and sample-size reasoning; longitudinal / mixed-effects mode
Effect size, power, and sample-size reasoning; longitudinal / mixed-effects models; simulation-based reproducibility assessment (engine-backend).
Verified
Continuous, binary, ordinal and categorical outcome support — closed-form provid
closed-form providers, code-verified 2026-08-11.
Verified2026-08-11
Cross-over designs with real closed-form and simulation providers — verified 202
verified 2026-08-18 through the full analysis service, including a closed-form/simulation agreement check.
Verified2026-08-18
Guided, dependency-aware design workflow with readiness/preflight review.
Guided, dependency-aware design workflow with readiness/preflight review.
Verified
Role-based access and portfolio views.
Role-based access and portfolio views.
Verified
Tiering and entitlement infrastructure — live (`CapabilityTier`, `AppCapabilitie
live (CapabilityTier, AppCapabilitiesService, CapabilityGuard), verified 2026-07-24.
Verified2026-07-24
Customer-specific terminology / lexicon — shipped and live on higher tiers.
shipped and live on higher tiers.
Verified
Two working AI features today: Timeline AI Draft and the Stat Assistant. **Four,
rewrites the scientist's stated objective into a well-formed one using only what they wrote, reports which of six slots it addressed, and by contract may not invent a comparator, dose, timepoint or effect size), design drafter (study_planning_ai_draft_service
Superseded2026-09-19
Covariate adjustment (ANCOVA)
added 2026-09-09. Family ancova_mean_difference, closed form, verified through the real AnalysisService and cross-checked against a Monte Carlo simulation of the actual OLS fit.
Verified2026-09-15
Non-inferiority and equivalence sizing
added 2026-09-15. Three families: non-inferiority for means (exact one-sided noncentral t), equivalence for means (exact TOST) and equivalence for proportions (normal-approximation TOST).
Verified2026-09-16
Unequal allocation
added 2026-09-15 for the six two-arm closed-form families: mean difference, proportion difference, covariate-adjusted, non-inferiority and both equivalence families.
Verified2026-09-15
Claim carefully — verified, with qualifiers that travel with the claim
ARRIVE
say "ARRIVE-oriented reporting" or "supports ARRIVE reporting." Never "ARRIVE compliant."
Verified — with qualifiers
Reproducibility
"supports," "helps assess," "designed to improve." Never "ensures."
Verified — with qualifiers
Custom species/strain catalog
the data model exists; confirm the user-facing surface is complete before promoting it.
Verified — with qualifiers
ANCOVA reachability
resolved 2026-09-11, verified 2026-09-15.
Superseded2026-09-15
Multiplicity-aware multi-arm sizing
the engine sizes the contrasts a multi-arm study will report, with an exact Dunnett adjustment for the vs-control family and Bonferroni for all-pairwise (ca2b4fa, 2026-09-11).
Verified — with qualifiers2026-09-15
Stratified randomised allocation at enrollment
added 2026-09-16, built 2026-09-15 (Ken: "stratification at enrollment").
Verified — with qualifiers2026-09-17
Blinding in the record
code-read 2026-09-16, executed 2026-09-16/17: BlindedTreatment model, a distinct unblinding_authority role deliberately not implied by admin (EMERGENCY-UNBLINDING-DESIGN.md §2.1), enrollment redaction under the blind, and ARRIVE readiness checks for both rando
Verified — with qualifiers2026-09-16
Server-side study exports (CSV)
added 2026-09-18, code-read. study_export_service behind /api/study_{enrollment,data_plan,collection_status,analysis}_export/<c>/<s>/<file_type>: enrollment, data plan, collection status and analysis rows as CSV with a declared column contract per export and b
Verified — with qualifiers2026-09-19
Analysis of the collected data (the designed analysis, engine-computed)
rewritten 2026-09-19 (PLAN-results-analysis-rewrite.md; core 76dc334).
Verified — with qualifiers2026-09-19
Post-collection sizing readout ("did the design's assumptions hold?")
added 2026-09-18, code-read and walked on dev tox28.
Verified — with qualifiers2026-09-18
Data Collection module
added 2026-09-18, built 2026-09-13 → 17 as DATA-COLLECTION-PROTOTYPE, transferred into core-backend (cba574f … 6147a0c) and study-manager-frontend (7555f806, b6cc3e56) on 2026-09-17 and mounted at collection/:studyId; the legacy data/:study-id module and /data
Verified — with qualifiers2026-09-18
SEND submission package (FDA nonclinical, SENDIG 3.1.1)
added 2026-09-17. The SEND Submission Builder in core-backend (app/lib/send_submission/, routes at /api/send_submission/…, flask send_submission, screens at send/:studyId) turns a study's approved design and collected data into the package the FDA Study Data T
Verified — with qualifiers2026-09-17
Tiering as a commercial capability
the *infrastructure* is live and verifiable, but pro has no tier defaults, ARRIVE/SEND sit on no rung, and every gate hides rather than advertising.
Verified — with qualifiers
Not claimed — absent or incomplete
Blocked designs as a distinct capability
a blocking factor currently changes nothing in the analysis; it resolves identically to a two-group design.
Planned2026-09-09
Study-to-study pilot evidence ("pilot evidence from study X" as a source on a clone)
added 2026-09-18, not built. The pilot-evidence import with its expected-value proposal and approval exists from a file; nothing yet offers a completed study on the platform as that source.
Planned2026-09-18
Survival / time-to-event analysis
ModelFamily.SURVIVAL is declared and the capability map carries an entry, but no registered provider claims it.
Planned2026-09-09
Randomisation sequence generation
we ask whether a study was randomised and report the answer; we do not generate or record the allocation sequence.
Superseded2026-09-17
Any regulatory certification, GLP compliance, or guaranteed compliance.
Any regulatory certification, GLP compliance, or guaranteed compliance.
Planned
AI-assisted expert guidance as a general capability
two specific features exist; the broad "Expert Personal Assistance" module is not delivered.
Planned
Ledger last transcribed 2026-09-19. The ledger keeps its own corrections — rows that said “this does not work” and were later found wrong — because a row without a date is a hope, not a claim.