A ranked, cited
risk report
A finished protocol in; a ranked, cited risk report out, with ICH M11 section detection.
Ranked,
never scored
A read-only diagnostic that runs on the PDF you already have: dose-optimization adequacy, eligibility restrictiveness, endpoint and estimand coherence, the operational triggers of an amendment, and where the document stands against the ICH M11 structure. Every finding cites the sentence it came from.
There is no total, no percentage and no grade anywhere on the report. A number would be ours, and a reader would optimize it.
Design risks · STUDY-A · worst first
One dose level carried into the randomized comparisonHigh · dose optimization
Primary endpoint names no censoring conventionHigh · endpoint and estimand
Creatinine exclusion narrower than the modernized recommendationMedium · eligibility
Assessment with no collection instrument namedLow · operational
Five things a meeting
cannot hold at once
Dose-optimization adequacy
Does the protocol carry more than one dose level into randomized comparison, state a selection rationale, and pre-specify the decision rule? Assessed, not simulated: operating characteristics stay with the tools built for them.
Eligibility restrictiveness
Each criterion read against the modernization recommendations, with the exclusionary consequence stated: narrower than it needs to be, and who it excludes.
Endpoint and estimand coherence
Does the primary endpoint survive to a defined analysis population, censoring convention and assessment schedule, or does the chain break somewhere between them?
Operational amendment triggers
Visit windows that cannot be met, assessments with no collection instrument, criteria no screening field can capture: the gaps the simulator reports, rendered as risks in the document.
ICH M11 structural alignment
Which template headings the document matches exactly, matches loosely, or shows no heading for. Match kinds only, never a verdict that a required section is missing.
Twelve questions,
each with a remedy
A finished oncology protocol is read against each one. Each carries a default risk tier and a reachable remedy, and a test proves the remedy clears it. A question the product cannot answer is not asked.
| Question | Area | Default risk |
|---|---|---|
| Is the comparator the current standard of care in every enrolling region, and is the rationale for choosing it stated? | clinical | high |
| Is the crossover or subsequent-therapy policy stated, and its effect on overall-survival interpretation acknowledged? | clinical | high |
| Is the tumor-imaging interval identical across arms and aligned to the primary endpoint? | clinical | medium |
| Is the schedule-of-events burden (biopsies, imaging cadence, PK density, questionnaires) justified against site and patient capacity? | operational | medium |
| Are central-laboratory, central-pathology and BICR logistics defined: vendors, turnaround, shipment? | operational | medium |
| Is blinding feasible for the dosage form: matching placebo, identical appearance and administration? | operational | medium |
| Is tumor-tissue availability and biomarker-testing turnaround compatible with the screening window? | operational | high |
| Are dose-limiting toxicity definitions and the DLT observation window stated? | safety | high |
| Is the adverse-event grading standard (CTCAE version) specified? | safety | low |
| Is there a DSMB or IDMC with a charter and pre-specified safety stopping rules? | safety | high |
| Are pediatric obligations (RACE Act or PIP) addressed for the molecular target? | regulatory | low |
| Is a diversity action plan or enrollment-goal statement present? | regulatory | medium |
Every template heading,
and its match kind
Three kinds and nothing else: an exact heading match, a loose one, or no heading found. not_detected means “no heading found”. The template’s applicability rules live in its formatting, not its hierarchy, so no screen may report a required-section gap until that is parsed. A placeholder slot nobody must fill is never counted against a protocol.
- 1 Protocol Summary · exact
- 2 Introduction · exact
- 3 Objectives and Endpoints · exact
- 4 Trial Design · exact
- 5 Trial Population · loose
- 6 Trial Intervention · exact
- 7 Discontinuation · loose
- 8 Assessments · exact
- 9 Adverse Events · exact
- 10 Statistics · exact
- 11 Oversight · no heading found
- 12 Appendices · loose
The model finds.
The engine ranks.
The engine locates, ranks and maps every finding. A person opens the review.
Every finding carries its quote
A verbatim passage, located by character offset and, for a PDF, by page. A quote that cannot be located is surfaced as unlocated, never snapped to something nearby.
Ordering is total
Worst first by risk tier, then catalog id, then reference, so two builds of the same inputs give the same list, and the board never re-sorts by anything the report does not carry.
The triage board
A filter rail, finding cards in the report’s own order, a protocol pane that scrolls to the cited page, and the M11 tree colored by match kind with its boundary sentence beside the count. Filters hide cards; they never reorder them.
Open as a review
One click creates a Reviews record with the protocol as its single source, and the append-only decision log begins: findings become owned decisions, and the export gate applies.
Two reads, in words
that cannot overstate
The comparison vocabulary is positional: only_in_previous, in_both, only_in_current. Never resolved, new or unchanged. In our own testing, the same document analyzed twice gave different finding lists while agreeing on the claim types. The finding level does not reproduce across two reads, so no screen may say a design risk went away.
The comparison leads with the claim-type view, which does reproduce, and it is not a gate: there is no status field and no verdict in the response.
“This report describes design risks detected in the uploaded protocol document by an automated reading. It is a statement about what the document says: a risk that is not listed was not detected, which is not the same as a risk that is not present.”
Printed on every report, unchanged.
Found while it is
still a paragraph
| Who | Today | With Design Risk |
|---|---|---|
| Protocol author | A review meeting reads the document once, from the top, and agrees what it remembers. The thin dose rationale is found by the agency reviewer. | Uploads the PDF and reads twelve answers, each cited to the sentence behind it and ranked worst first, before the document leaves the building. |
| CRO feasibility | Discovers the visit window that cannot be met and the assessment with no instrument after the sites have signed. | Sees which windows, assessments and criteria the operational triggers flagged, and prices the bid on what the document actually asks for. |
| Sponsor amendment review | Compares two protocol versions by reading both and trusting the redline. | Runs the amended version against the first and reads the claim-type comparison, in words that cannot overstate it. |
| Regulatory affairs | Checks structure by hand against the template, section by section. | Sees every template heading with its match kind, the count with its denominator, and the sentence that says what the map does not claim. |
Not an authoring suite
Authoring systems see only protocols written inside them. This runs on any finished document, so it is an input to them, not a rival.
Read-only
It runs on a finished PDF: no authoring migration, no IT project.
What it does not do
Stated, not buried.
- No readiness scoreNo total, no percentage, no grade, no color that reads as one. A test checks the rendered page for the words “score” and “%”.
- A risk that is not listed was not detectedThat is not the same as a risk that is not present, and the report says so in its own first paragraph.
- The M11 map reports match kinds onlyUntil the template’s applicability rules are parsed, “no heading found” is the strongest thing any screen may say.
- Operating characteristics are not computedWhether a dose-optimization plan is adequate is assessed from the document; its statistical properties belong to the tools built for them.