Skip to content

True First protocol assurance

Surface what needs a decision before study launch.

HumanTrue examines the protocol as an interdependent operational specification, not a collection of isolated paragraphs. It does not judge the science or rewrite the protocol. It gives the sponsor source-linked questions in time to decide what was intended.

Cross-document verification

Clinical meaning depends on what connects.

HumanTrue checks eligibility, endpoints, dosing, visits, withdrawal rules, analysis definitions, and supporting manuals as a connected dependency graph.
  • Contradictions between synopsis, body, tables, and footnotes
  • Ambiguous thresholds, timing, events, and stopping conditions
  • Missing or unreachable operational paths
  • Cross-document drift across protocol, SAP, ICF, and manuals

Using our AI pipeline, we've analyzed thousands of publicly available clinical trial protocols. We've found inconsistent language, ambiguities, and knowledge gaps. All of which compromise the ability to run a trial.

INCONSISTENCY

Contradictory · Oncology · NCT02953782

How many patients, exactly?

The synopsis says 15-24 patients will be enrolled in Phase 1b, but the sample size section says 32.

Protocol p. 20
An estimated 15 to 24 patients will be enrolled in Phase 1b, depending on dose escalation and expansion.
Protocol p. 27
An estimated 32 patients will be enrolled in Phase 1b, depending on dose escalation and expansion.
INCONSISTENCY

Contradictory · Oncology · NCT02723955

Inconsistent dosing for the same drug combo

The same safety run-in section gives two different starting doses, 24 mg and 80 mg, for the same drug combination.

§ 4.1.1.3.2
The dose ... in each chemotherapy combination will be 24 mg...
§ 4.1.1.3.2
An initial three participants will receive 80 mg ... in combination with chemotherapy
AMBIGUITY

Lexical · Oncology · NCT02953782

Disease progression — which kind?

The synopsis uses "disease progression" as a treatment-stopping trigger, but that term can mean any radiographic progression or only iRECIST-confirmed progression.

Protocol pp. 8, 20
Treatment with study drugs may be continued until ... disease progression. ... Added confirmed tumor progression according to iRECIST to the reasons for withdrawal from study treatments.
§ 10.3.1.1
... iRECIST requires the confirmation of progression and uses the terms iUPD (unconfirmed progression) and iCPD (confirmed progression). Confirmatory scans should be performed at least 4 weeks, but no longer than 8 weeks after iUPD.
INCONSISTENCY

Temporal · Oncology · NCT02178358

Imaging cadence collision

The synopsis says post-treatment imaging happens every 2 months, but the efficacy section says every 6 weeks.

Protocol p. 4
Patients who have discontinued study treatment without progression will continue to be followed for progression. Every attempt should be made to gather all information every 2 months (radiological scans and survival)...
§ 10.1
Patients who come off therapy and do not have objective progressive disease should be followed every 2 months after discontinuation from study treatment until death, including radiologic examinations every 6 weeks...
INCONSISTENCY

Contradictory · Oncology · NCT02723955

17 cycles or 18?

The synopsis sets the duration of appx. 2 years to be 17 cycles, but the treatment-duration section sets it at 18.

Protocol p. 17
For participants enrolling in HNSCC dose optimization cohort the maximum treatment duration is approximately 2 years, or up to 17 cycles for both GSK3359609 (feladilimab) and pembrolizumab.
§ 4.2
The maximum treatment duration for participants enrolled in HNSCC Q6W dosing cohort is expected to be approximately 2 years, up to 18 cycles for both GSK3359609 (feladilimab) and pembrolizumab.
INCONSISTENCY

Contradictory · Rheumatology · NCT01237587

Named 50%, timed at 30%

The endpoint is labeled "time to first 50% pain reduction," but the calculation qualifies patients on their 50% reduction date while measuring the clock from their earlier 30% reduction date (i.e., the qualifying event and the timed event are different).

§ 12.2.7
Time to event variable: 6b. Time to first 50% reduction in BPI average pain score
§ 12.2.7
Derivation and Details: 6b. For the patients with a 50% reduction at a visit in the treatment phase, time = days from the date of the visit that the earliest 30% reduction is observed to the randomization date.
AMBIGUITY

Semantic · Oncology · NCT02723955

Replacement rule, reversed

At a dose level with only one participant, the replacement rule says to enroll a replacement only when fewer than one participant drops out meaning a replacement is triggered when the participant completes the period, and no replacement happens when the participant actually drops out.

§ 4.3
In Part 1A and Part 2A, if at any dose level a participant prematurely discontinues study treatment before the completion of the 28-day DLT observation period ... a replacement participant may be enrolled ... if <3 participants complete the DLT observation period at that dose level.
§ 4.3
In those Part 1A dose levels that have a single participant enrolled, a replacement participant will be enrolled if <1 participant prematurely discontinues study treatment before the completion of the 28-day DLT period.
AMBIGUITY

Vagueness · Oncology · NCT05218499

Contraception duration mismatch

The protocol enrollment criteria and amendment log give different durations for the required contraception rules during the study.

§ 3.3.2
Women of childbearing potential ... and men able to father a child must be ready and able to use ... birth control ... until 6 months and 12 days after last dose.
§ 4.2.2.3
The contraception requirements for male patients ...3 months and 10 days. For female patients ... 6 months and 10 days

Validation in layers

Examine the source. Test the model. Demonstrate fidelity.

  1. Protocol integrity

    Identify consistency, ambiguity, logic, and feasibility issues before digitization.

  2. Schema and semantic QA

    Validate USDM structure, controlled terms, Biomedical Concepts, and sponsor extensions.

  3. Round-trip fidelity

    Confirm that the structured model faithfully reproduces schedules, footnotes, criteria, and source meaning.

  4. Human consensus

    Add Level 2 expert review and certification for regulated or high-consequence outputs.

Amendment governance

A change never stops at the protocol.

On average, a phase 3 trial carries 5-7 substantial amendments, each affecting dozens of downstream artifacts, decisions, and regulated systems. See what changed, what depends on it, what must be regenerated, and whether every affected artifact still agrees with the source.
HumanTrue amendment analysis showing a source-level visual comparison between protocol versions
Visual deltaSee paragraph, table, requirement, and version changes together.
Dependency map connecting protocol concepts to pharmacy, lab, SAP, EDC, IRT, ePRO, and site outputs
Dependency mapSee which documents, systems, instructions, and decisions inherit the change.
HumanTrue targeted regeneration interface for updating affected downstream content
Targeted regenerationUpdate affected sections against the revised source.
HumanTrue re-verification results showing review status after an amendment
Re-verificationConfirm that every affected artifact still agrees with the protocol.

Why it matters

Trial documents now move at machine speed. HumanTrue makes sure errors don't.

AI accelerates content creation and systems configurations, so that studies can be built in a fraction of time. But an inconsistency that once slowed a single site now spreads to every connected system at machine speed - unreviewed, everywhere at once.

  • 76%

    of Phase I-IV protocols now require at least one substantial amendment

    Tufts CSDD

  • ~45%

    of those amendments are considered avoidable

    Tufts CSDD, 2016

  • $141K-$535K

    direct cost of a single substantial amendment, by complexity

    Tufts CSDD

  • ~3 months

    average added timeline per amendment

    Tufts CSDD

20,450

First-pass candidates across over 1,000 public protocols.

The current analysis focuses on ambiguity and inconsistency, including contradictory doses, incompatible timing, reversed replacement rules, and endpoint derivation conflicts.

Read the Protocol Benchmark

Start with one protocol

Find what your current process cannot see.

Request a Working Session