The NAMs evidence package, explained.

What a NAMs evidence package is, and why one antibody's plan can hold 144 monkeys

One NAMs proposal, from a planned animal study to FDA's answer

1. Pick the study and the askVeltrimab's team asks to replace a planned 6-month monkey study with a 3-month one.
2. Name what the study is there to answerLiver injury late in dosing, in the example. Each planned group of animals gets its question.
3. Grade the evidence for each question, and write the caseStudy reports, lab results, models and published work, each with its limits. This is the step this guide is about.
4. Two people signA toxicologist answers for the science, a regulatory lead for the questions.
5. Ask the regulator, and change the plan only thenA meeting request goes to FDA. Until FDA answers, every animal in the count is a proposal.

Step 3 is the subject of this guide. Veltrimab and its studies are fictional.

Four findings that surprise people new to NAMs

NAM, weight of evidence, context of use, relevant species: the words to know

Fourteen terms are defined in the one-page summary.

The documents in a NAMs evidence package, and the question each one has to answer

In shortNo regulator publishes a template for this package. FDA's formal meetings guidance sets what a meeting request and meeting package need, and FDA's two drafts say what a weight-of-evidence case and a method's validation should cover. Teams build the rest from those. Seven documents do most of the work, and the coverage map is the one that shows where the case is thin.

Under US rules, a sponsor, the company that runs the trial, must give FDA "adequate information" about pharmacology and toxicology, along with who evaluated it and where the records are.3 The rule says the kind, duration and scope of tests vary with the proposed trial. So the regulation sets a standard and leaves the method to the sponsor and to guidance. FDA's March 2026 draft puts it plainly: federal laws and regulations "do not stipulate exactly which nonclinical tests sponsors must conduct nor in what model".5

Seven documents and what a reviewer looks for in each

What is binding on new approach methodologies in November 2026, and what is only proposed

In shortOnly two things in this field bind a US sponsor today: the statute as amended in 2022, and the current regulations, which ask for adequate information. The direct final rule takes effect February 4, 2027 unless withdrawn. FDA's roadmap is a plan; its two NAM drafts are marked not for implementation. FDA Modernization Act 3.0 is a bill. Every citation in your package should say which of these it is.

Sort every citation by what it is before you write a sentence that rests on it. FDA guidance says of itself that it does not "establish legally enforceable responsibilities", and a draft adds that it is "not for implementation".1 Neither is empty. A draft tells you what the agency is thinking and how a reviewer may read your argument. But "FDA requires" and "FDA proposes" are different sentences, and a reviewer sees the difference at once.

Status of each text, read November 3, 2026

FDA Modernization Act 2.0 and 3.0: what changed in the statute, what is still a bill, and the rule FDA issued first

In shortSection 3209 of a 2022 spending law changed the statute from "preclinical tests (including tests on animals)" to "nonclinical tests", and defined that term with animal tests still on the list. FDA Modernization Act 3.0 would make FDA change its regulations to match. Each chamber has passed its own bill, and the bill-status records we read show no enactment. FDA issued a direct final rule on the wording anyway, effective February 4, 2027 unless withdrawn.

People call the 2022 change "FDA Modernization Act 2.0". It became law as section 3209 of the Consolidated Appropriations Act, 2023, headed "Animal Testing Alternatives".6 It made two moves. In section 505(i), the phrase "preclinical tests (including tests on animals)" became "nonclinical tests". A new section 505(z) defined the term.

The definition covers a test "conducted in vitro, in silico, or in chemico, or a nonhuman in vivo test", and lists cell-based assays, organ chips and microphysiological systems, computer modeling, other human or nonhuman biology-based methods, and animal tests.6 Read closely, the law widened the door. It banned nothing and required nothing. Section 3209 also changed the biosimilar route, so a toxicity assessment there "may rely on, or consist of" certain studies.6

What FDA Modernization Act 3.0 would add

FDA's regulations kept the old words after 2022. The 3.0 bill would require an interim final rule within one year of enactment, replacing "animal" references in named sections of Title 21, and adding the statutory definition to four definition sections: 312.3, 314.3, 315.2 and 601.31.7,8 On our comparison, the operative text the Senate passed and the text the House passed are the same, and only the bill numbers and closing lines differ.7,8

Neither bill has cleared the other chamber. Govinfo's status records show S. 355 held at the desk in the House since December 17, 2025, and H.R. 2821 received in the Senate on July 21, 2026, with no public law in either.9,10

Those records were last updated on June 24 and September 19, 2026, and govinfo held no enrolled text when we looked on November 3, 2026. We could not read congress.gov, so we have no check of Congress's own page for those last weeks. An advocacy group said after the House vote that one more Senate vote on the House-passed bill remained.25 If you cite 3.0, cite it as a bill and date your check.

The rule FDA did not wait to issue

On September 22, 2026, FDA published a direct final rule, "Nonclinical Testing Terminology", covering 21 CFR Parts 312, 314, 315, 361 and 601. It swaps "animal" for "nonclinical" in the named sections and adds a definition, and FDA calls the changes noncontroversial.19 A direct final rule works by default. If nobody files a significant adverse comment by December 7, 2026, FDA confirms it and it takes effect February 4, 2027. If someone does, FDA withdraws it within 30 days of the comment period's end and continues through the companion proposed rule.19,27

FDA was careful about what the rule does not do. It "does not eliminate or prohibit animal studies, change evidentiary standards" or add costs or requirements.28 The rule itself says it "adds no new requirements".19 So its effect on your package is one of wording. It removes language that could suggest animal testing is the only way, and it moves no burden of proof.

FDA's animal testing roadmap and the monoclonal antibody draft: from a plan to a proposed 3-month study

In shortFDA's April 2025 roadmap planned to cut routine 6-month primate studies to three months for antibodies with nothing concerning in 1-month studies plus NAM tests. Its December 2025 draft goes further on paper: in general, studies longer than 3 months are not warranted when 3-month data come with a weight-of-evidence assessment. In some cases no toxicology study is warranted. It covers monospecific antibodies outside oncology, and it is still a draft.

The roadmap begins with antibodies and plans to reach other biologics, then new chemical entities.2 Its fourth implementation step is the one most teams quote: reduce routine 6-month primate testing to three months for antibodies with nothing of concern in 1-month studies plus NAM tests.2 In the long term, FDA says it aims to make animal studies "the exception rather than the norm", within three to five years.2 That is an aim. The roadmap also says that "Thresholds will need to be developed" for when testing can be reduced.2

The draft that moved the starting point

The December 2025 draft, "Monoclonal Antibodies: Streamlined Nonclinical Safety Studies", changes what the argument looks like. It covers antibodies that recognize one target, called monospecific, in any indication outside FDA's oncology office. It excludes multispecific antibodies, conjugates and antibody constructs.1 A separate May 2026 draft covers oncology biologics and conjugated products.29 The roadmap reasons from a 1-month study plus NAMs toward a 3-month study. The draft reasons from a 3-month study plus a weight-of-evidence assessment toward no longer study at all. Check which starting point your program fits.

What FDA's draft says a weight-of-evidence assessment may include

Notice where NAMs sit in that list: last, as one optional input. For a standard antibody, much of the case can rest on target biology and existing data. Development and reproduction get a separate line: assessment "should begin with a WoE risk assessment", and an animal study may be warranted when that cannot answer the question.1

"Achieved" in FDA's report, "Draft" on the document

FDA's year-one report of April 2026 rates its roadmap goals. Reducing routine 6-month primate testing is "Achieved", with the expected impact stated as saving "hundreds of non-human primates annually".11 That is the agency's account of its own progress. The legal status of the text it rests on has not changed: a draft, with comments open "at any time" on FDA's page.30

Companies and trade groups have asked FDA to widen its scope and say more about how to document an assessment, and some worry that shorter studies may not support later trials.31 A package should use the draft's reasoning, cite it by date and status, and ask the review division to confirm that it applies.

Validated, qualified, fit for purpose: how FDA judges a NAM, and where ISTAND and organ chips stand

In shortFDA's March 2026 draft names four things to show: context of use, human biological relevance, technical characterization and fitness for purpose. Validation is shown for a use. Qualification is a formal FDA determination, and a qualified tool can then be used in any drug development program for its stated use. It is optional. We found no organ chip qualified through FDA's ISTAND program on November 3, 2026. One liver chip's full package went in during September 2026.

These words get mixed up, so use the meanings in the draft from FDA's drug center, the Center for Drug Evaluation and Research (CDER). Validation is "a process by which the accuracy, reliability, and relevance of a procedure are established for a specific context of use". Qualification is a determination that a drug development tool and its context of use "can be relied upon to have a specific interpretation and application".5 Once qualified, a tool is available to any drug development program for that use.32

Qualification is optional. "CDER will review data from both qualified and validated NAMs", and it looks at whether the use fits the case in front of it. The draft also says that, qualified or not, a NAM's "validation is warranted" to judge the data.5

The draft separates three jobs a NAM can do: replace or offer an alternative to a traditional method, fill a data gap, or confirm and complement findings.5 Say which job your method is doing. A method offered to complement an animal study needs a far lighter case than one offered to replace it.

ISTAND, the formal qualification route

FDA made its ISTAND program permanent on July 31, 2025, reporting more than 16 active submissions in early 2026.11 FDA's program page, current September 22, 2026, reports acceptances into the program, including the first in silico tool to help predict liver injury on June 3, 2026.32 Acceptance is an entry point.

One developer filed the full qualification package for its liver chip on September 9, 2026, by its own account, and reports no FDA decision.33 FDA's "first AI tool qualified", in December 2025, scores liver biopsies for clinical trials in a liver disease.11 It says nothing about replacing an animal study.

FDA's reviewers have also published their own account of how particular tests became accepted alternatives, and what they look for in NAM submissions.35 Among in vitro submissions, the 15-year analysis reports that stem cell-derived and sandwich culture models were more common than 3D models and organ chips.34 So the organ chip, the method most headlines discuss, is a small part of what reviewers have seen so far.

That database arrived with the September 2026 rule.28 It is separate from FDA's inventory of development contexts where CDER is open to streamlined programs, which FDA's April 2026 report dates to October 2025 in its text and December 2025 in its timeline.11,37 FDA says the database's entries are illustrative and do not mean a method will be acceptable in a future submission.36

The database holds no antibody application, no organ chip and no chronic toxicity example. FDA's report says so in general terms: validated alternatives exist for some endpoints, such as skin sensitization and mutagenic impurities, "but not others (developmental toxicity, chronic toxicity)".11 That sentence sets the limit for a chronic-study package: its case rests on the weight of several lines of evidence, read together.

EMA's position on non-human primates, and the EU roadmap that leaves biologics out

In shortEMA has not published a roadmap of that name. It works through a rolling non-clinical work plan for 2026 to 2028, a draft reflection paper on non-human primates and a draft on testing requirements. The European Commission's roadmap of June 1, 2026 is a separate policy plan, and it excludes biologicals. Antibody teams should read EMA's draft paper, which accepts waivers case by case and calls monkeys a last resort.

"The EU roadmap" and "EMA's roadmap" are easy phrases to write and both mislead. On November 3, 2026, EMA's page on the ethical use of animals lists no roadmap.38

EMA's plan is a consolidated work plan for the non-clinical domain, 2026 to 2028, whose short-term goals include follow-up on alternatives to non-human primates and integrating NAMs and weight-of-evidence approaches into the regulatory framework.22 A separate draft paper, in consultation from February 13 to June 30, 2025, lists current testing requirements and 3Rs opportunities.39 For 2026 the work plan lists finalizing the non-human primate paper, and it plans to start collecting data on whether ICH S7A needs revising and on the challenges of implementing testing strategies under ICH S6(R1).22

What the draft EMA paper says about antibodies

The draft reflection paper (EMA/CHMP/55697/2025) says non-human primates "should be used for regulatory safety testing only as a last resort".13 Its summary lists waiving 6-month studies for antibodies among important examples, subject to the conditions it sets out.13

It treats weight of evidence as acceptable "on a case-by-case basis" through formal or informal EMA or national procedures.13 It says "A single method will not replace, in a one-to-one context, the use of an NHP for the assessment of risk", so the focus is on weight-of-evidence approaches.13 It is a draft. Comments closed January 31, 2026, and we found no final on November 3, 2026.21

EMA's routes for talking about a NAM

That last row catches people out. FDA's roadmap says it will "offer regulatory relief" to sponsors who submit NAM data in parallel.2 EMA's route explicitly keeps such data out of the decision.13 One set of documents can serve both agencies only if each is told plainly what the data are for.

The Commission's roadmap and the pharma package

The European Commission published its "Roadmap towards phasing out animal testing for chemical safety assessments" on June 1, 2026, as a Communication with a staff working document.12 Trade summaries list pharmaceuticals among its 15 domains.42 The Communication itself says "chemical pharmaceuticals", and footnote 20 reads: "biologicals, vaccines, gene therapies; advanced therapy medicinal products and novel therapy veterinary medicinal products are excluded".12 So it says little directly to an antibody program.

The Communication does call EMA's Innovation Task Force, voluntary data submission, scientific advice and qualification the most advanced model of a safe space for such dialogue.12

The EU pharma package would go further. Article 7(7) of the Council's position says an applicant "shall not carry out animal testing in case scientifically satisfactory non-animal testing methods are available".23 That is a proposal. Parliament's file read "Awaiting Parliament 2nd reading" on November 3, 2026, and nothing has appeared in the Official Journal that we could find.24 Directive 2010/63/EU is the law in force, and EMA's draft paper summarizes its limits on non-human primates.13

ICH S6(R1) and the S-series: the flexibility already in the guidelines, and the NAM guideline that does not exist

In shortICH guidelines are the harmonized text that FDA and EMA both apply. ICH S6(R1) says 6-month studies "have generally been appropriate" for chronic biologics, and that shorter or longer ones have supported authorization. Several S guidelines already allow weight of evidence. We found no ICH guideline on NAMs. FDA has called one an "ultimate vision", and EMA plans to start collecting data on S7A and on implementing S6(R1).

A team arguing for a shorter study is working inside the guideline's own words. ICH S6(R1) says duration "should be scientifically justified".14 Its addendum states: "For chronic use products, repeat dose toxicity studies of 6 months duration in rodents or non-rodents are considered sufficient" and adds that longer studies have not generally changed the clinical course.14 The parent text has "in some cases shorter or longer durations have supported marketing authorisations".14 So the guideline writes 6 months as enough, and it has always allowed that less may do.

The dated list a package should cite

FDA's roadmap says it can propose revisions to ICH S6, and calls a NAM guideline at ICH an ultimate vision.2 We found no such guideline or a published concept paper for one on November 3, 2026, though ICH's own pages were not fully readable to us.

There are signs of movement around it. FDA, HESI and the Safety Pharmacology Society held a workshop on ICH S7A and NAMs on July 28, 2026.45 EMA plans to collect data on S7A and on implementing S6(R1).22 Until something is adopted, your package relies on flexibility in the current text.

What month six adds for an antibody, how well animals and liver chips predict, and what nobody has counted

In shortThe largest dataset we found took in 142 antibodies and analyzed 111 for new toxicities. For 71% of those, chronic studies found no toxicities or nothing new. For 13.5%, they found something of concern or that changed a trial, and for 2% the program ended. That is the other side of the argument, in numbers. Animal studies predict the absence of harm better than its presence. The liver chip's 87% comes from small molecules, in a paper mostly written by its maker. No one publishes how many packages are filed or accepted.

In the largest dataset we found, 11 companies shared first-in-human and chronic toxicity data on 142 antibodies, and 111 standard or modified antibodies were used to count new toxicities.46,47 The project joined a national medicines regulator, NC3Rs and pharmaceutical companies, and it proposed a weight-of-evidence model for when a 3-month study is likely sufficient "for some mAbs".48,46 Read both halves of the result. For 79 of the 111, chronic studies showed nothing new. For the other 32 they did, and 15 of those mattered for people or for the trial.47

What we read: for this study, the abstract, NC3Rs' project page and the authors' poster. For the other papers in this section, the abstract only, with the liver chip paper's disclosures. We did not read any of these in full.

A package that quotes only the 71% will read as advocacy. Say what the 13.5% found, and explain why your antibody does not resemble those cases. That is what the weight-of-evidence model is for.

How well do animal studies predict people?

FDA's roadmap and report state that over 90% of drugs that appear safe and effective in animals do not go on to FDA approval, "predominantly due to safety and/or efficacy issues".2,11 The figure mixes both reasons and it is about whole-program outcomes, so it says little about any one study's value. A narrower dataset is more useful. An industry consortium matched animal toxicology with Phase 1 human findings for 182 molecules, organ by organ.49

The authors read the 86% as support for current testing. The 48% is the figure a critic would quote. For your package the point is narrower: you are arguing about what a specific additional study would add.

The closest public counts, and what has no public figure

Nobody publishes how many NAMs evidence packages are filed, how often they succeed, what one costs or how large this market is. The nearest counts are indirect. FDA says more than 16 submissions are active in ISTAND,11 expects "hundreds" of monkeys saved a year,11 and notes that industry submits "only a fraction" of its NAM data.34 CDER received 1,410 initial investigational new drug applications (INDs), the filings that open human trials, in fiscal 2013,52 which is an old figure.

If you need a baseline, measure your own: hours per package, gaps found after review, questions from FDA, and months from idea to agency answer.

Weight of evidence and model credibility assessment: how a team grades each line, and how far a computer model can carry a claim

In shortNo regulator publishes a grading scale. Teams write their own and should show it: what makes a line strong, and how many independent lines a claim needs. For computer models, FDA's January 2025 draft sets out seven steps, with risk set by how much the decision rests on the model and how bad a wrong call would be. It is written for AI models and it is still a draft. ICH M15 covers model-informed work more generally.

A coverage map is a grid with the questions down the side and the kinds of evidence across. Each cell holds a grade and a citation. The value is in being able to see the thin cell. Here is a fictional grid, with the grades a team might give after reading the reports. The grade is that team's judgment under its own written rule.

Rules a team writes down before grading

These are a team's rules, and they need to be stated as such. Writing "three donors" or "Strong" as though FDA set it is a common overreach, because neither FDA draft sets a number of donors or a scale.1,5

Seven steps for a model, and the scope that goes with them

FDA's January 2025 draft, "Considerations for the Use of Artificial Intelligence To Support Regulatory Decision-Making", lays out seven steps: define the question of interest, define the context of use, assess model risk, plan how to establish credibility, execute the plan, document results and deviations, and decide whether the model is adequate for that use.15 Model risk combines model influence (how much this model contributes relative to other evidence) with decision consequence (how bad a wrong decision would be).15

Mind the scope. The draft is about AI models, and it is draft. Many models in a package are not AI. A physiologically based pharmacokinetic model, which simulates how a body handles a drug, is built from equations and physiology, and ICH M15 covers model-informed evidence like it.16

A team may choose to apply the seven-step structure to every model because it is clear and reviewers will recognize it. Say that you are doing so by choice. FDA's roadmap adds that it may review such simulations to inform first-in-human dosing and to justify waiving animal studies that would normally serve that purpose.2

How to count the monkeys a NAM proposal could spare, and read an exposure margin, without overstating either

In shortFDA's 144 monkeys and $50,000 per animal are real statements with particular origins: a UK page describing a pre-reduction baseline, and a chip maker's calculator. Your own count should start from your plan, tie each group to its question, subtract what a shorter study still uses, and say "could be spared" until a regulator answers. An exposure margin is a simple division that tells you how far animal exposure ran above expected human exposure.

Trace the figures before you reuse them. FDA's roadmap says typical antibody programs use 144 non-human primates, citing a page from NC3Rs, the UK's 3Rs center.2 That same page says its analysis gave the rationale to halve the count "from 144 to 64", through smaller groups, fewer recovery animals and fewer studies.51 So 144 describes a design that the source itself argued could be cut.

The roadmap's "up to $50,000 per NHP" cites a cost calculator on a chip maker's website.2 These numbers are fair to quote as FDA's. They are poor foundations for your own program's business case.

A fictional plan, group by group

Here is the fictional plan from this guide's example. Each group of animals exists for a question. Some questions can be covered by other evidence, and the team tracks them one by one.

A gross count treats the whole 48 as spared. But the ask is a 3-month study, which needs animals of its own. Show both figures, and say which one your 3Rs report uses. Your count also has two states: "could be spared", while you are asking, and "agreed", after FDA answers.

An exposure margin, worked through from a report

A margin compares exposure at the highest dose with no adverse effects in animals, the NOAEL, against the exposure expected in people. Regulators and reviewers read two measures: peak level (Cmax) and total exposure over time (AUC). The arithmetic is one division. The figures below are from the fictional report.

A large margin supports the case, and it has limits. If the human dose rises, the margin shrinks. If anti-drug antibodies cut exposure in some animals, the average hides them. And for an antibody with a NOAEL at the highest dose tested, the margin shows how far you looked, while the true limit stays unknown. A package says all of this.

Common NAM submission errors, from citing a draft as law to a single-donor run, and what they put at risk

In shortFDA publishes no list of rejected NAM proposals, so this list is our reading of what the texts warn against. The worst errors are about status and scope: a draft cited as a requirement, or a bispecific argued under a monospecific draft. The risks are a wasted meeting, months of delay, an animal study booked too late, and in the worst case a harm that the dropped study would have shown.

What is at stake, for a program and for people

For people. The reason for all of this is that a medicine might harm someone. FDA itself cites TGN1412, an antibody that caused a life-threatening cytokine release in volunteers in 2006 after appearing safe in monkeys.2,11 That case argues for human-based tests. The 2 of 111 antibodies whose programs ended after chronic findings argue for caution in the other direction.46,47 FDA's roadmap says any new method must be "equal or superior to the animal test it replaces in protecting patients".2

For the program. FDA can put a trial on clinical hold when it lacks what it needs. In fiscal 2013, 125 of 1,410 initial INDs (8.9%) went on hold in the first 30 days, with manufacturing reasons cited most, then clinical, then toxicology.52 That figure is dated, covers all INDs, and is unrelated to NAMs. It gives a sense of scale.

For the calendar. A meeting that ends without agreement can cost a quarter. FDA's August 2026 guidance sets the clock: from receipt of the request, FDA schedules a Type C meeting within 75 days.17 If FDA says no after a monkey slot has gone, the booking starts again. EMA notes the shortage of non-human primates, worsened during the COVID-19 pandemic.13

FDA's December 2025 antibody draft names "for example, Type B meetings".1 The final guidance adds that INTERACT, for early questions before a pre-IND meeting, can cover "the use of New Approach Methodologies".17 The choice depends on where the program stands.

What regulators expect when AI touches a NAMs package: FDA's draft, EMA's paper, and where GMP Annex 22 does not reach

In shortFDA's AI draft covers models whose output supports a regulatory decision, and it says it does not address drafting a submission where that does not affect the reliability of results. So AI used to read and draft sits outside it, and an AI model offered as evidence sits inside it. Either way, a named person stands behind the conclusion. Draft EU GMP Annex 22 concerns manufacturing and does not govern this work.

Two uses of AI meet in this work, and they carry different duties. One is a model offered as evidence, such as a machine-learning immunogenicity predictor. FDA's January 2025 draft covers that, with its seven steps.15

The other is software that reads reports, fills the coverage map and drafts prose. The same draft says it "does not address" AI "used for operational efficiencies (e.g., internal workflows, resource allocation, drafting/writing a regulatory submission) that do not impact patient safety, drug quality, or the reliability of results from a nonclinical or clinical study".15 That last clause matters. A tool that drafts is out of scope only while it does not change what the evidence says, and a citation it invented would change that.

The limits. Software can find passages, line up figures, spot a missing control and draft a paragraph with a citation. It cannot decide that an argument is good enough. US rules ask for the identification and qualifications of the individuals who evaluated the results and concluded it is reasonably safe to proceed.3 That is a toxicologist's name on a statement, and no tool can sign it.

Who builds a NAMs package today, from nonclinical toxicologist to regulatory lead, and where the weeks go

In shortA program toxicologist and a regulatory lead assemble the case by hand, from study archives, lab reports, model reports and papers. No public survey measures how long it takes. Our own estimate is 3 to 6 weeks, and most of that time goes on reading and rechecking. Teams work in house, through a contract lab or consultant, or with software for parts of the job.

The six people behind a NAMs package

Three ways teams build a NAMs case today

From one liver chip donor to a signed request for a 3-month study: 144 monkeys and two signatures

In shortDr. Lena Ortiz, a program toxicologist, opens a grid of questions against methods. One cell, a single-donor liver chip run, holds a group of 48 monkeys. A three-donor run arrives, three agents read and regrade it, and the count of monkeys that could be spared moves from 32 to 80. She and the regulatory lead sign. Every count stays "could be spared" until FDA answers.

1

Every safety question against every method, on one grid

The page opens on the grid. Veltrimab's safety questions run down the side: liver injury, cytokine release, immune function, anti-drug antibodies, off-target binding, exposure margin and embryo-fetal development. New approach methods run across, with human data as a supporting line. Each cell holds a grade and its evidence, and the top right counts the plan's 144 monkeys one dot at a time: 32 could be spared, 48 are held by the liver question, and 64 by embryo-fetal development.

The rule is printed on the screen. "Covered" means one method graded Strong and a second line at least Moderate. That rule belongs to the team, and it can be changed in settings.

2

One cell, and the sentence that holds 48 monkeys

Lena opens the weak cell. The liver chip result was clean, but report LC-0118, page 4, says only one hepatocyte donor was tested, and the method's qualification plan asks for at least three. So the cell is Weak, and the whole liver question stays partly covered. The panel states what the cell decides: 48 monkeys held, 0 could be spared.

3

A three-donor report arrives, and three agents read it

On October 6, Dana Okafor's lab sends LC-0131. It waits under "Evidence to map", already matched to that cell. Lena clicks "Add to the matrix". The assay reader checks donors and controls, the evidence mapper regrades the cell and the critic checks the new citations.

The report itself is highlighted. Three donors, no change in the liver measures, and each chip caught the reference compound, which shows it could have detected injury.

4

The cell turns Strong, and the count moves from 32 to 80

With the Moderate human cell and Phase 1 lines beside it, the liver question is now covered under the team's rule. Forty-eight more monkeys move to "could be spared". Embryo-fetal development still holds 64, because the placenta chip has not been run.

5

The package weighs the case, with one concern left open

Package PKG-0412 asks to shorten the chronic monkey study from 6 to 3 months. It shows supports on one side of a balance and open concerns on the other. The single-donor concern is struck through, and one stays open: anti-drug antibodies in 2 of 10 high-dose animals, with a mitigation for Lena to accept or reject.

The score of 88 of 100 and "8 of 8 expectations met" are the team's own instrument. They help people see what is thin, and FDA does not score packages this way.

6

Two people sign, each for a recorded meaning

Omar Haddad, the nonclinical regulatory lead, has signed the meeting questions. Before Lena signs the science, the checks are on the screen: every expectation has evidence, every sentence is cited, no overstated wording, model credibility assessed. She signs with her password, with the meaning "Approved for submission to FDA". Priya Ranvelle, the 3Rs officer, is told. An agent never signs and never files.

The evidence matrix, cited justification, seven-step model check and agency questions, screen by screen

In shortBehind the balance sit the documents a reviewer reads: the evidence matrix, the justification with a citation on every sentence, model credibility assessments and the "Does the Agency agree" questions. Other screens show a human-specific target, a package where shortening was ruled out, the portfolio across nine programs, the evidence library, the dashboard and the rules people can change.

The evidence matrix and the justification it feeds

A model's seven steps, and what happens when it stands alone

The "Does the Agency agree" questions that end the package

A human-specific target, and a package where shortening was ruled out

Teriquamab binds a target only humans have, so no monkey helps. Its plan holds a transgenic-mouse study of 120 mice. Kidney injury is covered by a three-donor glomerulus chip and kidney organoids, so 84 mice could be spared. The human exposure question still holds 36, because its model's first verification missed on one antibody. NWB-6120 shows the opposite result. Its 1-month study had a finding, so shortening was ruled out and the package closed as not a candidate. A tool that always said yes would be useless.

The portfolio, the library, the dashboard and the rules

Seven evidence agents, the toxicologist and regulatory lead who decide, and what goes in and comes out

In shortSeven agents read, grade, count, draft and check. They never sign, never file and never change a grading rule. A toxicologist decides the science and a regulatory lead decides the questions, and nothing goes to a regulator without both. The product provides controls that support your compliance with your own record and signature duties. Your team remains responsible for that compliance.

Who decides what on a NAMs package

The people in the Veltrimab sample

From study reports and models to a signed package and agency questions

Connecting to the systems your nonclinical program already runs

The solution reads from your GLP study archive and SEND datasets, your in vitro lab system, contract lab portals and literature sources, and it watches FDA and ICH publications. Access is by role. Each agent run, grade change and signature goes on the audit trail, and reports stay where they are, read in place. Information security is SOC 2 Type II audited and managed to ISO 27001. GLP, 21 CFR Part 11 and similar obligations stay with your organization: the controls here support your compliance with them.

What a thin NAMs package and a slow one cost a nonclinical program

In shortNobody publishes the cost of a NAMs package or of one that fails. The figures that exist are about monkeys and calendar days: up to $50,000 per animal as FDA states it, 75 days to a Type C meeting, and about $40,000 for a day of direct Phase II or III trial cost. The money sits in studies you book because the case was not ready. Find your own numbers before you believe anyone else's.

Where your NAMs costs hide, and how to find your number in a week

What a NAMs calculator will not count

Your return on faster evidence packages: toxicologist days released, plus animal costs if FDA agrees

In shortTwo levers matter: the time a team stops spending on reading and rechecking, and the animal cost of studies avoided when an agency agrees. The first you can measure in a pilot. The second depends on someone else's decision, so keep it apart and set it low. If the case works only because of the second lever, say so to your finance team.

Your return on four NAMs packages a year

At these defaults, team time alone does not cover the tool's cost. The animal line carries the case, and it depends on an agency's answer. A team with more packages, or a real measured saving, reads this differently. So try your own numbers, and try the share agreed at zero.

How ready your nonclinical team is for NAMs automation: plans, reports, rules and signatures

In shortEight questions, each scored 0 to 2, and one for reflection. The score says where to start. A low score points to baseline work, and it is no judgment on your science. Nothing you choose leaves this page.

Twelve questions for any NAMs evidence software vendor, us included

In shortThese test what matters in this work: whether every grade traces to a passage, whether a draft is ever described as law, whether the tool says no, and who signs. Ask us too. Ask to see each answer on your own files.

Testing NAMs software in ten weeks: one finished package, planted errors, one live ask

In shortStart with a package you already finished, because you know where it was thin. Plant a few errors and see what comes back. Then run one live ask beside your usual process. Measure days, gaps found, citations that hold and false alarms. Your toxicologists decide whether it earned its place.

What to measure on a NAMs pilot, and what a pass looks like

NAMs questions answered: FDA Modernization Act 3.0, the 6-month monkey study, ISTAND and EMA

In shortEach answer stands on its own, with status as of November 3, 2026.