Protocol design optimization, explained.

Why a trial's plan gets changed after it starts, and why some of those changes could have been seen coming

One Phase 3 protocol, from a clinical team's question to its first amendment

1. The team sets the questionVeltrimab worked in Phase 2 in Crohn's disease, a long-term inflammation of the gut. The next study needs 640 participants at about 180 sites in 22 countries.
2. A clinical scientist drafts the protocolDraft version 0.9 runs to 148 pages. Six departments comment by email and in tracked changes.
3. The draft is reviewed for design riskEach section is read against past amendments, design standards and guidance. This is where a criterion that already failed in another study is caught, or carried forward.
4. The protocol review committee decidesThe sponsor's internal committee meets on October 14 and approves the design, or sends it back.
5. Regulators, ethics committees and sites receive itFrom here, any substantial change is an amendment, with its own approvals, its own wait and sites working from different versions.

Step 3 is the subject of this guide. The people, drugs, studies and figures in this example are fictional.

Four findings about protocol amendments that surprise newcomers

Substantial amendment, summary of changes, critical to quality, estimand: the words to know

Fourteen terms are defined in the one-page summary. ICH is the International Council for Harmonisation, which writes shared guidance for regulators and industry.

Six questions a protocol design review puts to a draft, section by section

In shortA design review asks whether the study can be run as written, and whether a past study shows otherwise. It covers who may join, what each visit asks, what is being measured, how participants are watched over, whether sites can deliver it, and whether the document agrees with itself. Feasibility assessment is one part of it. Writing the protocol is separate work that comes first.

Two guidelines set the tone. ICH E8(R1) says "Study designs should be operationally feasible and avoid unnecessary complexity."6 ICH E6(R3), the good clinical practice guideline, says the sponsor "should not place unnecessary burden on participants and investigators."7 Neither tells you what is unnecessary in your study. A review supplies that judgment, one line of the protocol at a time.

A review needs three kinds of evidence. The first is general: guidelines and published benchmarks. The second is the sponsor's own standards, such as a rule that no visit should exceed a set number of hours. The third is the most useful and the most neglected: the sponsor's own past amendments, each with its reason, its cost and its delay. A reviewer who can say "this exact criterion cost another study of this drug four months" wins an argument that general advice loses.

ICH E6(R3), E8(R1), E9(R1), M11 and the UK's 2026 rules: what each asks of a draft protocol, and since when

In shortGuidelines ask for feasible, clear protocols built around what matters most. Laws decide what happens once you change one. We found no rule that requires a protocol review committee or names a method. Regulatory status below is as of November 2026.

You will meet the phrase quality by design. ICH E8(R1) describes it as ensuring a study's quality is "driven proactively by designing quality into the study protocol and processes".6 In plain words, the thinking happens before the study starts.

FDA guidance is not binding; 21 CFR is. The European Commission's own example of a substantial modification is "inclusion of a new exclusion criteria in the protocol due to a potential liver toxicity described in section x of the IB", the investigator's brochure.21 Where the brochure already described that risk when the protocol was written, this is the kind of amendment a design review exists to catch. Step 4 is the stage at which ICH's final text is recommended to regulators for adoption.6

Protocol amendment rates: 76% of protocols amended, and how much of that is avoidable

In shortThree in four protocols are amended, on average 3.3 times, and it takes most of a year to land each change. The share called avoidable fell from 45% in 2016 to about 23% in 2024. Quote the date with the number. Even the lower share adds up to most of one avoidable amendment for each amended protocol.

The 45% avoidable from 2016 and the roughly 23% from 2024

Both figures are judgments made after the fact, on data the participating companies supplied, and each company draws the line in its own way. A better test asks one question of each amendment: was the information that would have prevented this change available to the sponsor when the protocol was written? A risk in the investigator's brochure was. A competitor's result published two years later was not.

Regulator requests lead the 2024 reasons, along with strategy changes,2 and some regulator requests can be anticipated too: a question about an estimand left to the statistical analysis plan, a separate document, is a good example.

The arithmetic that makes this worth doing: 3.3 amendments multiplied by a 23% avoidable share comes to about 0.76 avoidable amendments on each amended protocol. That is our multiplication of two published figures, and rough, but it says the typical amended protocol carries most of one change a better draft could have stopped.2

Why people say amendments happen, in their own words

A study of 242 approved amendments in 53 UK studies sponsored by the National Health Service (NHS) found "addition of sites" the most common change, with 61 amendments, and "to achieve the trial's recruitment target" the most common reason, with 39.8

Eleven staff at one NHS trust, interviewed for that study, named the root causes of avoidable ones: rushing the first application knowing an amendment would follow, not involving the right people at the start, and finding the plan unworkable once delivery began.8 Those are NHS-sponsored studies, and a commercial sponsor's mix will differ, though a commercial team may well recognize the three causes.

What a protocol amendment costs: $535,000 in 2016 dollars, and everything that figure leaves out

In shortThe figure everyone quotes is a median direct cost from 52 protocols, published in 2016. It leaves out the delay, the approvals and the lost enrollment. A later Tufts report, read only second-hand, credits earlier Tufts studies with nearly $1 million. Compute your own number from your own amendments, and treat these as a cross-check.

Direct cost means the work a sponsor can invoice to the change: redrafting, consent forms, submissions, fees, site retraining, and updates to data collection and the systems that assign treatment. Four costs sit outside it.

Pricing one Phase 3 amendment: direct cost against days of delay

Tufts estimates that a day of a Phase 2 or 3 trial costs about $40,000 in direct cost.26 Lost sales are a much larger and softer number: the same research gives about $500,000 a day in its journal article and about $800,000 in its August 2024 white paper, from one set of 645 drugs.26,27 That figure applies only if the delay moves the launch, and it varies by disease.

Eligibility criteria and screen failure rates: how one rule shrinks the pool you planned on

In shortA Phase 3 protocol averages about 30 rules for joining. About 3 in 10 people screened are never enrolled, and in Crohn's disease trials the figure is about half. Rules multiply: each one looks reasonable alone, and together they remove most of the people you planned on. A review asks each rule for its reason and an estimate of its effect.

A screen failure is someone who goes through screening and is never enrolled. Each costs staff time, tests and a disappointed patient, and contributes no data.

In 17 Phase 2 or 3 inflammatory bowel disease trials, all run by one contract research organization and covering 16,913 screened people, not meeting the disease activity score was the top reason in Crohn's disease, at 23.1% of screen failures, and intestinal infections and tuberculosis screening were other major reasons.9 Reasons like these are visible in earlier trials, which is why they can be reviewed for before a new one starts.

How five reasonable eligibility rules compound

Four eligibility problems reviewers see, and what the evidence says

Broadening has evidence behind it, in one disease. A study emulating completed lung cancer trials on 61,094 patients found that many common criteria had a minimal effect on the hazard ratio, the measure of how much treatment changed survival, and a data-driven broadening more than doubled the eligible pool on average.30 So a strict rule may protect the result less than its authors assume. Some narrow rules protect people or the science, and those should stay. The review's job is to make each one state its reason.

Participant burden in the schedule of activities: the visits and procedures a trial could live without

In shortProcedures grew far faster than endpoints. Nearly a third of procedures and data collected per protocol serve no core purpose or repeat more often than needed. Burden is real and rising, though Tufts' own workshop found no significant link between its burden score and dropout, so measure it for what it is: time and effort you are asking of people.

A newer Tufts study with 15 sponsor companies looked at where those procedures go. Non-core procedures, which support only lesser or exploratory endpoints, make up 16.2% of the Phase 3 total.

Among Phase 3 procedures that do support core and standard endpoints, 12.6% were judged non-essential, meaning they run more often than the result needs.31 Nearly half of those non-core procedures in Phase 2 and 35% in Phase 3 gather exploratory or future-use data.31 These are the first procedures a review questions, and the ones most often defended by someone who remembers why they went in.

Reading one heavy schedule: hours on site at each visit

The sample sponsor's own standard says no visit should exceed 4 hours unless the reason is written down. Weeks 0, 8, 12 and 52 go over with a reason: a long observation period, or sedation and recovery after an endoscopy. Week 4 goes over with none, because two blood samples, the full efficacy assessment and a diary review all landed on the same day.

A home nurse could collect the later sample, and the visit falls to about 4 hours. We found no published rule that sets a 4-hour limit: this is one company's choice, and yours may differ.

What the evidence on burden and dropout says

Overall burden for patients in Phase 2 and 3 trials has risen steadily since 2011, across 156 protocols from 14 companies, driven by questionnaires, lab and blood work and routine procedures at each visit.32

The tempting next sentence, that lower burden means lower dropout, is less settled. A multi-company workshop on 66 protocols reported that "correlational analysis did not reveal a significant correlation" between the burden score and dropout, and the authors note the sample size may be why.33 Earlier work by the same group linked the score with study conduct and the number of substantial amendments.33 ICH E8(R1) gives the simpler reason to care: investigators and participants can tell you when visits "may be overly burdensome and lead to early dropouts".6 Ask them.

Estimands and intercurrent events in a protocol: stating the exact question before the first patient

In shortAn estimand says exactly what the trial measures, including what counts when something interrupts a participant's course. ICH E9(R1) says the protocol should specify the primary estimand. Published reviews find most protocols do this badly, and a change later usually means an amendment.

Suppose a participant in a Crohn's disease trial gets worse and takes a rescue medicine at Week 8, then is free of symptoms at Week 12. Did the study drug work? That depends on the question. An event like this, which happens after treatment starts and affects how the result reads, is called an intercurrent event.

ICH E9(R1) names five things an estimand must pin down: the treatment, the population, the variable measured, how intercurrent events are handled, and the summary across the group.10 It offers strategies for each event, including treatment policy (count the result regardless), composite (count the event as a failure) and hypothetical (estimate what would have happened otherwise).10

The same trial, two estimands, two answers

How well protocols describe estimands, in two published reviews

The UK review searched 29,212 protocols held by the Health Research Authority and found 122 that mention the word "estimand", under 1% of the database. Of the 81 that attempted a primary estimand, 76 were industry sponsored, and 8 listed an intercurrent event that did not meet the definition, most often missing data.35

The earlier review read protocols published in two journals, in the first year after the guideline, and found none stated the estimand outright.34 Practice has moved since then. These two studies are a warning about what to check.

This matters for amendments directly. E9(R1) says "Changes to the estimand during the trial can be problematic and can reduce the credibility of the trial."10 The M11 template tells sponsors to list changes to the primary estimand or endpoints as a change of strategy in an amendment's reasons.18

Stopping rules and monitoring in a trial protocol: why a risk known from an earlier study belongs in the draft

In shortIf an earlier study of the drug showed a problem, the new protocol should say how it will be watched and when dosing stops. Adding that later is close to the textbook amendment, and it can mean asking hundreds of people to consent again.

A medical monitor is the doctor who watches over participants on the sponsor's side of a trial. In a design review, this person reads the draft against earlier study results and the investigator's brochure, the reference file on the drug. For each risk known so far, two questions follow: is there a test at the right visits, and does the protocol say what happens when the result crosses a line?

An independent data monitoring committee, where one exists, assesses the trial at intervals and recommends whether to continue, modify or stop it.7 FDA's guidance on those committees is still a draft from February 2024.36

The rules treat a late addition as a formal event. Under US rules, "the addition of a new test or procedure that is intended to improve monitoring for, or reduce the risk of, a side effect or adverse event" is one of the listed examples of a change needing an amendment.19 The European Commission's example of a substantial modification is a new exclusion criterion for a liver risk described in the investigator's brochure.21

The same liver rule, added in the draft and added by amendment

Reviewers also watch for the opposite problem. A rule copied from another drug may stop dosing for a change that means nothing here, and every extra test adds visits and blood. Medical judgment decides this, so a good review routes each question to the medical monitor by name and never settles it by default.

Protocol feasibility assessment: screening windows, sample shipping and the sites that have to make it work

In shortEvery added country and site makes a protocol harder to run: both correlate with more amendments, longer timelines and more dropout. Feasibility review asks whether real sites can meet each window, using what they told you last time. The arithmetic is simple and usually unwritten.

Tufts tested 10 design variables against 11 performance measures across 187 protocols. The number of countries and the number of sites were each positively correlated with amendment frequency, longer screening and study duration, and dropout.5 In statistical models, endpoints, eligibility criteria, procedures per visit, countries and sites predicted cycle time, enrollment, retention and amendment frequency, even after controlling for phase and therapeutic area.5 These are correlations: big studies are hard for many reasons. But they name the design choices that are worth arguing about before anyone commits.

Worked through: why 28 days of screening was too short

A colonoscopy read by a central reviewer needs a slot at the hospital and time at the reading vendor. A window set from a template ignores both. The same arithmetic applies to a blood sample that must reach a central laboratory within 72 hours at room temperature, when couriers in some planned countries cannot do it. It applies too to training for the staff who score the disease, if that training expires whenever a site opens late.

E6(R3) offers one way out: "Building adaptability into the protocol, for example, by including acceptable ranges for specific protocol provisions, can reduce the number of deviations or in some instances the requirement for a protocol amendment."7

Simple to check, easy to miss: five feasibility questions for a site panel

Synopsis, body text and schedule of activities: when a protocol disagrees with itself, sites follow the table

In shortA protocol states the same facts in several places, and they drift apart across drafts. The fix is cheap and the check is mechanical, yet people do it by eye. Sites act on the table, so a mismatch turns into deviations, departures from the protocol at a site, and those are rising fast.

Six mismatches in one draft schedule

From the fictional VLT-305 draft. The first row matters most: Week 12 is a main result, so a site following the table does the colonoscopy two weeks early.

E6(R3) asks that protocols and operational documents be "fit for purpose, clear, concise and consistent".7 In the NHS study from section 04, interviewees named duplicated text across documents as a cause of inconsistency, and recommended that research staff own document checks.8 The M11 template repeats "Ensure alignment with every other section of the protocol" in more than one section's instructions, because no layout can stop two people typing different weeks.18

A deviation is a departure from the protocol at a site; FDA's December 2024 draft guidance, still a draft, uses the ICH E3(R1) wording, "any change, divergence, or departure" from the protocol's design or procedures.38 In the earlier Tufts sample, deviations involved nearly one-third of patients enrolled, and procedures per visit, countries, endpoints and especially sites were associated with them.37 We found no study that counts how many amendments begin as an inconsistency, so count them in your own summaries of changes.

Can you predict a protocol amendment? Complexity scores, past amendments and what the evidence supports

In shortComplexity correlates with longer trials and harder enrollment, and a structured review can reduce it. Predicting a specific amendment from protocol text has barely been tested: one 2026 preprint reaches moderate accuracy, on eligibility criteria only. Any likelihood on a screen is an estimate. Test it against your own past protocols before you lean on it.

How a section-by-section risk figure is built, and three reasons to treat it with care

A common way to make one number from many findings is to give each a likelihood, then multiply the chances that each does not happen. Take the 12 findings on the fictional VLT-305 draft, from 22% down to 4%. Multiply (1 − 0.22) × (1 − 0.18) and so on through all 12, and you get about 0.26. So the chance of at least one is about 74%. Adding up the 12 likelihoods gives 1.26 expected amendments, and multiplying by a cost per amendment gives an expected cost.

1. It assumes findings are independent. Three eligibility problems may share one cause, and one amendment often fixes several things at once, so the sum overstates the count.
2. It depends on the archive. A likelihood drawn from 148 past amendments means little until you have checked how often those estimates came true.
3. It can run high. Tufts figures imply about 0.76 avoidable amendments on an amended protocol (our multiplication), so 1.26 expected on one draft deserves a question.2

A test you can run. Take 20 past protocols as they stood before their first amendment. Have the method estimate risk with no sight of what happened. Then compare it with what did happen: did the high-likelihood findings show up in real summaries of changes, and how many real amendments did it miss? We found no published study that runs this test on full protocols, so your result will be the best evidence you have.

AI in clinical trial design: what automation can do for a draft, and what FDA, EMA and the EU expect

In shortAutomation suits reading, matching and counting. People keep design decisions, statistics and medical judgment. We found no regulator's rule on AI in protocol review. FDA's January 2025 draft excludes tools for operational efficiency unless they affect patient safety, drug quality or the reliability of study results, and a design tool sits near that line, so keep humans deciding and every change traced.

Three practical limits follow. A model can sound sure about a citation it invented, so every finding should show its protocol line and its source, and a person should open them. It also has no idea what your team said in last week's meeting, so its output is a set of questions for people who do. And a change suggested by a tool can move the study's question or a participant's risk, so statistical and medical suggestions go to the statistician and the medical monitor.

No rule certifies a tool for any of this. What you can get from a vendor is controls that support your compliance, which you then validate for your own use.

Who reviews a draft protocol today, from first circulation to the protocol review committee

In shortSix departments comment by email, one person reconciles the comments, the summary and the table get checked by eye, and a pack goes to committee. Almost nobody rereads past amendments. No public figure says how many days this takes, so the split below is an estimate to replace with your own.

The people who comment on a draft protocol, and what each one owns

Where a Phase 3 draft's 23 review days might go

Those four days are the telling ones. Past summaries of changes usually sit in the trial master file, the official record of each study, by study and by date, with no way to ask what happened the last time a criterion like this one went into a protocol. So the memory lives in people, who move on.

In-house committee, benchmark software, outside panel or consultants: four ways to run protocol design review

From a 148-page draft to a signed disposition log: a protocol amendment risk review at work

In shortThis is the example from the top of the page, inside the product. Dr. Maya Chen opens a board of eight drafts. VLT-305 meets the committee in seven days, with 12 findings. She reads one finding beside the past amendment it resembles, accepts the new wording, and watches the estimated risk fall. Other owners decide their own findings, and the draft cannot go to the committee until every high finding has a decision.

1

Eight draft protocols on one board, each section colored by risk

Each column is one draft in committee order, from NWB-118 on October 9 to VLT-307 on November 13. Each has a risk dial and a strip of its sections, colored by the strongest open finding: red for high, amber for medium, blue for low, green for decided. Across the board, 18 of 26 open findings cite a past amendment.

VLT-305 is at 73%, in seven days. The Population cell is selected, with three open findings. The section names are this sponsor's short forms of the ICH M11 headings: "Safety" stands for M11's section on adverse events, for example.18

2

Three eligibility findings, each beside the amendment it resembles

Below the board, §5 Population shows each finding next to its evidence. Requiring two failed biologic classes, 22% likely to force an amendment, sits beside VLT-204 Amendment 2: 4.1 months, $162K, 64 re-consented. The 12-month washout sits beside CRV-301 Amendment 1, with its 112 screen failures. "History of any malignancy" sits beside VLT-302 Amendment 1, where 37 screen failures were treated basal cell carcinoma.

3

One click opens the summary of changes itself

VLT-204 Amendment 2 reads as the team wrote it. The criterion was not required by the target label. Investigators at 31 of 52 sites reported eligible patients being referred elsewhere. It took 52 ethics committee approvals. A reviewer can check the evidence for themselves here, because the finding links to the document.

4

The finding shows its line, its amendment, its standard and its guideline

The review shows the draft with every finding highlighted, and a card for each on the right. Four sources sit in the card for inclusion 3: the protocol line, the past amendment, sponsor standard NWB-STD-014 §3.2, and ICH E8(R1) §3.3. It also notes that the rule clashes with the "prior biologic exposure" grouping in section 4.2.

On the left, the dial reads 73%, with $656K of expected cost and 3.7 months of expected delay. It read 74% at first; accepting the small tuberculosis finding took it to 73%. Part A explains how a dial like this is built and why it is an estimate.

5

She accepts the wording, and the risk falls from 73% to 67%

Maya accepts. "Conventional therapy (corticosteroids or immunomodulators) or at least one biologic class" goes in as a tracked change. The dial moves to 67%, expected cost to $556K and expected delay to 3.1 months. If every suggested fix were accepted, the dial would read 17%.

The arithmetic: an accepted finding keeps 15% of its likelihood, and the expected cost adds the 12 likelihoods and multiplies by $535,000, the 2016 public median for a Phase 3 amendment.4 Both are this sample's assumptions, and a buyer should replace them with their own.

6

Week 12 in the synopsis, Week 10 in Table 2

The lead reviewer compares the synopsis, the body and the schedule for every procedure and visit. A main result, the colonoscopy, is at Week 12 in the synopsis and section 8.2, and at Week 10 in the table. NXL-115 recorded 18 deviations from the same kind of mismatch before its amendment. There are six mismatches in all, each with a button to fix it in the table.

The dosing row is left blank here because the sample's dosing weeks come off the screens.

7

The estimand goes to the biostatistician, the liver rule to the medical monitor

Each finding goes to the person who owns that area. Priya Ranvelle owns the primary estimand, which leaves interrupting events to the analysis plan. The card cites ICH E9(R1) A.6 and CRV-301 Amendment 2, where the agency asked for the strategies after the first participant was in: $610K and 4 months.

Dr. Omar Haddad, the medical monitor, owns the missing liver stopping rule, set beside VLT-204 Amendment 4, which re-consented 211 participants. A safety monitoring finding always reaches the medical monitor and is never downgraded below Medium.

8

The committee pack builds itself from the decisions

Maya clicks "Generate pack". The writer collects the decisions and reasons, draws risk by section, writes the summary and the questions for the committee, and checks that every citation resolves. The pack for Dr. Sam Patel, the committee chair, reads 74% at draft and 67% now: two accepted, one raised for the committee, nine still open, ten past amendments cited.

9

High findings hold the draft back until each one has a decision

Maya tries to send it. The screen stops her and lists what is still open, each with its owner and likelihood: the 28-day screening window, the 12-month washout, the primary estimand. A rejection needs a written reason. Accepting, rejecting or raising each one to the committee all count as a decision.

10

The protocol lead signs with her name, the time and the meaning

Once the high findings are decided, she signs the disposition log, the record of how each finding was decided. The record shows the signer, the time and the meaning: "I confirm these dispositions as protocol lead". The chair then takes the committee's decision on the protocol itself.

Select any screen to open it larger. The "likelihood" and "amendment risk" figures are sample estimates. They have not been validated against real outcomes, and your own back-test is the evidence to rely on. See section 12.

The amendment archive, the decision log, the design quality dashboard and your review rules

In shortAround one draft sit the screens a clinical development team uses all year: the runway of committee dates, every protocol's stage, an archive of past amendments, the log of decisions, a portfolio dashboard, and settings that hold your thresholds, owners, rules and sources.

Six reviewer agents, the named owners who decide, and what goes in and comes out

In shortSix specialist agents read, compare, estimate and rank. None can decide a finding or sign. A named owner decides each finding, the protocol lead signs the log, and the committee chair approves the protocol. The draft remains your controlled document.

What each protocol reviewer agent does, and where it stops

Who decides a finding, who signs the log and who approves the protocol

From one draft and your past amendments to a board, a pack and a signed log

Records and systems a protocol review rollout connects to

Audit trail, signatures and credentials a clinical quality team will ask about

Every agent step, decision, pack and signature is recorded with who, what and when. The signature shows signer, time and meaning, which is what 21 CFR 11.50 looks for in a signed record.52 These are controls that support your compliance with Part 11, FDA's rule on electronic records and signatures, and validating them for your use stays with you. The agents never edit the controlled document: an accepted suggestion goes in as a tracked change, and the draft stays yours.

Botminds' information security is managed to ISO 27001 and audited under SOC 2 Type II. No product can be certified for good clinical practice or Part 11, and we make no such claim. The solution runs at enterprise scale across many drafts and therapy areas, with roles, maker-checker approvals and an exportable audit trail.

What avoidable protocol amendments cost a sponsor each year, and where to find your own number

In shortPublished figures frame the scale: a median direct cost per amendment, a wait of most of a year, and a daily cost of a running trial. Your own summaries of changes hold the real numbers. You can pull most of them in a week.

Seven amendment costs to find in a week, starting with your last 20 summaries of changes

Worn-out sites, lost lessons and a difficult second consent: what the totals leave out

What catching amendments in the draft returns: direct cost, trial days and reviewer hours

In shortPut in your own numbers. The main value is a share of avoidable amendments that never happen, and that share is an assumption until your back-test measures it. Reviewer hours are a small and easily proved second line. Defaults use published figures where they exist and illustrative ones elsewhere.

Eligibility, schedule, estimands, consistency: where a design review saves money

Protocols a year, avoidable amendments and reviewer hours on your own figures

How ready your clinical development team is: past amendments, standards, owners and a committee that will use it

In shortPick the answer closest to your team today. Eight questions score 0 to 2, and the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

Twelve questions for any protocol optimization software vendor, us included

In shortEvery tool looks good on a clean sample. These questions test what a demonstration hides: whether it reads your history, whether its likelihoods came true, whether each finding can be checked, and who stays in charge of your document.

Testing protocol review software in ten weeks: a blind back-test, planted errors, one live draft

In shortStart with protocols you already signed, because you know how they turned out. Hide what happened, see what the tool finds, and count hits and misses. Plant errors to test the consistency check. Then review one live draft alongside your usual process, and decide on measured numbers.

Hits, misses, false alarms and how long owners took to decide

Protocol optimization questions answered: amendments, screen failures, estimands, burden and AI

In shortThe questions people ask most, each answered so it can stand alone.