The Junk Science Hiding in Plain Discovery

The most dangerous expert in your next murder case may not appear on any expert disclosure list. He'll be a dispatcher or patrol officer with eight hours of training, testifying about whether a 911 call "sounded staged."

Here's the technique in brief: 911 call analysis claims to infer caller guilt from word choice, cadence, and phrasing. It was developed by Tracy Harpster, a retired Moraine, Ohio deputy police chief, from a master's thesis analyzing just 100 recorded calls (50 guilty, 50 not guilty) that yielded a 20-indicator checklist. The claims are remarkably specific: callers who say "bleeding" are innocent, callers who say "blood" are guilty, and 80% of callers who say they "need help" are innocent. Harpster charges up to $3,500 for his eight-hour class, typically paid with tax dollars, and the method has now been documented in more than 100 cases across 26 states.

The record on wrongful convictions is the part that should stop you cold. Russ Faria spent 3.5 years in prison after a Missouri prosecutor used a dispatch supervisor trained in Harpster's "analytical class" to testify his call sounded staged; he was acquitted at a second trial. Robert Roberson sits on Texas death row, in part because he was perceived as "oddly unemotional" reporting his infant daughter's death; he has since been diagnosed with severe autism. And Jessica Logan's conviction is now before the Illinois Supreme Court, after ProPublica's investigation prompted review of cases built on the technique.

Why haven't you heard of it? Because dozens of defense attorneys interviewed by ProPublica had never heard of it either. The method spreads without the usual fingerprints of expert testimony:

• The underlying methodology is never disclosed and the founders are never proffered as experts.
• 911 operators or ordinary officers testify about caller culpability after a single training.
• The technique thereby "largely escapes Daubert or Frye evidentiary challenge," as Christopher Lau's 2026 Cornell Law Review article puts it.
• Prosecutors have admitted getting "creative … without calling it 'science'" to avoid admissibility hearings entirely.

Lau calls this "shadow forensics": practices that shield a forensic technique from evidentiary oversight. It means your standard gatekeeping motion is structurally harder to bring, because there's no expert to exclude. North Carolina's Office of Indigent Defense Services has already issued a statewide warning telling attorneys to watch for it. You should assume the same pattern can appear in your jurisdiction, dressed as lay opinion from a witness who "just noticed" the caller lacked urgency.

The scientific foundation has collapsed independently of any courtroom fight. Five independent studies, including a 2020 FBI Behavioral Analysis Unit study, contradicted the method, and at least six peer-reviewed studies have failed to replicate the checklist's findings. No one has replicated Harpster's original results.

That's why every 911 call in discovery deserves a word-level, diarized transcript before anyone interprets it. The checklist operates on impressions of cadence and phrasing, so the actual words are the ground truth the state's theory has to survive. When we build 911 call transcript analysis at Evidence Copilot, the goal is putting the exact language, timestamped and attributed, in front of the trier of fact instead of a checklist gloss, and the same discipline applies to jail call transcription whenever the state's theory turns on what a defendant supposedly said.

Shadow Forensics: Why Daubert Won't Save You

Here's the trap: your Daubert motion is drafted, your Frye brief is polished, and then the state calls a patrol sergeant who took an eight-hour class. No expert disclosure names a methodology. No scientist takes the stand. The jury hears that your client's 911 call "sounded staged," and the gatekeeping rules never trigger.

Christopher Lau's 2026 Cornell Law Review article calls this "shadow forensics": practices police and prosecutors use to shield techniques like 911 call analysis from evidentiary oversight. The mechanism is simple. The founders of the technique are never proffered as experts, and the underlying methods are never disclosed. Instead, 911 operators or ordinary officers testify about a caller's culpability after attending a training, so the technique largely escapes Daubert or Frye challenge entirely.

Prosecutors have admitted as much. Missouri prosecutor Leah Askey wrote in a 2016 email to Tracy Harpster that the method was not "recognized" as science in her state and described "getting creative … without calling it 'science.'" In the Russ Faria case, she put a dispatch supervisor trained in Harpster's "analytical class" on the stand to testify that Faria's call sounded staged. Faria spent 3.5 years in prison before he was acquitted at a second trial.

The FBI's role makes this harder to spot, not easier. A 2008 study co-authored by Harpster and FBI agent Susan Adams was published as an "exploratory analysis," but the FBI sent a version to departments in a bulletin with a readership of roughly 200,000, without the exploratory label. The FBI later published dissent in another bulletin, but that reversal has gone largely unnoticed. Even Judge John Bailey, a former Pennsylvania sergeant, said learning of it made him "scratch my head" given the technique's origins at the FBI Academy.

So what do you do when the gatekeeping rules don't apply? Attack the testimony you actually get:

• Demand the specific training, materials, and studies behind the officer's opinion. Harpster doesn't allow outsiders into his classes or access to his data, which is itself the point.
• Cross on the named indicators one by one. "Bleeding" versus "blood," an isolated "please," a "Huh?", each claim comes from a thesis built on roughly 100 calls, and at least six peer-reviewed studies have failed to replicate the findings.
• Force the witness to concede the confounders the checklist ignores: anxiety, education, culture, speech impediments, mental health, neurodivergence.
• Build the record even if you lose. Fair and Just Prosecution and the Innocence Project are pushing post-conviction review, and the Illinois Supreme Court has agreed to review Jessica Logan's conviction.

North Carolina's Office of Indigent Defense Services has already issued a statewide warning to attorneys to watch for this technique. Treat that as your intake checklist. When a detective, dispatcher, or operator is endorsed to interpret a call, get a word-level, diarized transcript first so the actual language, not a checklist gloss, is what the trier of fact sees. That's the discipline we built 911 call transcript analysis and jail call transcription around at Evidence Copilot: timestamped words with citations, so every claimed indicator can be checked against what was actually said.

The technique survives because it hides. Your job is to name it in the record.

What the Research Actually Shows About Callers and Suspicion

The empirical case against 911 call analysis has never been stronger, and defenders who understand the research can dismantle demeanor testimony before it reaches the jury. A four-study Cornell series published in Law and Human Behavior, drawing on roughly 1,800 civilian and 300 law-enforcement participants, found that demeanor-based suspicion is what turns callers into suspects, with emotion cited by 4 of 5 participants as the top driver. Once suspicion attaches, confirmation bias does the rest.

That research maps five behavioral dimensions officers and laypeople actually use: emotion, urgency, cognitive load, information management, and impression management. Conveying strong emotion and urgency reduced suspicion; sounding guarded or overly composed raised it. Law-enforcement participants showed gender bias against male callers, while civilian participants showed none. You can build cross-examination directly on those dimensions.

The confounding variables are where the checklist collapses. Cornell psychologist Jessica Salerno warns that neurodivergence, culture, disability, and acute stress produce wide variation in trauma response, and there is no one-size-fits-all demeanor for tragedy. Robert Roberson, on Texas death row, was perceived as "oddly unemotional" reporting his infant daughter's death; he has severe autism.

• At least six peer-reviewed studies failed to replicate the checklist training's findings.
• Five independent studies, including two FBI studies, contradicted Harpster's method.
• The evidence base rests on roughly 100 calls in the original thesis and about 200 total.
• The 2020 FBI Behavioral Analysis Unit found some indicators ran in the opposite direction of earlier findings.

Before you argue about what a caller sounded like, get the words themselves. A timestamped, speaker-diarized 911 call transcript analysis puts the exact language in front of the trier of fact instead of a checklist gloss, and every AI output stays marked for human review, because the tool never draws the legal conclusion for you.

Your Litigation Playbook: Challenging the Call, Not the Expert

The core problem with 911 call analysis is that there's rarely an expert to challenge. As Christopher Lau's 2026 Cornell Law Review article explains, the founders are never proffered and the methods are never disclosed; instead, dispatchers and patrol officers testify about caller culpability after an eight-hour training. The technique largely escapes Daubert or Frye review by design. So your playbook has to attack the testimony you actually get, not the expert you wish the state had disclosed.

Start at screening. When the state endorses a dispatcher, operator, or detective to interpret a call, treat that endorsement as a red flag and demand the specific training, materials, and studies behind the opinion. Harpster charges up to $3,500 for his eight-hour class, doesn't allow outsiders into his classes, and doesn't release his underlying data, so a witness who can't produce the methodology is telling you something. Put the refusal on the record. The gap itself becomes your argument: if the opinion rests on secret data, it isn't the kind of lay opinion the rules contemplate.

Then cross-examine the checklist one indicator at a time. The claims are specific and falsifiable, which makes them vulnerable:

• "Bleeding" supposedly signals innocence while "blood" signals guilt, and an isolated "please" or a "Huh?" counts as a guilt indicator.
• The evidence base is a couple of tiny studies of roughly 100 calls each, drawn from a 100-call master's thesis.
• At least six peer-reviewed studies have failed to replicate the checklist's findings.
• Five independent studies, including a 2020 FBI Behavioral Analysis Unit study and a separate FBI missing-child study, contradicted the method, with some indicators running in the opposite direction.

For each indicator the witness relies on, force the same three answers: name the study, state the sample size, and identify the replication. When the answer is "the class," you've won the point in front of the jury.

Build your rebuttal theme around confounding variables the checklist ignores: anxiety, education, cultural factors, speech impediments, and mental health. Cornell's research series, covering roughly 1,800 civilian and 300 law-enforcement participants, found emotion was the most-cited driver of suspicion, named by four of five participants. The Robert Roberson case is your model fact pattern: a caller perceived as "oddly unemotional" who has since been diagnosed with severe autism.

Finally, build the record even if you lose the motion. Prosecutors have admitted "getting creative ... without calling it 'science'" to dodge admissibility hearings, and the Russ Faria case shows the pattern: a dispatch supervisor trained in the analytical class testified the call "sounded staged," and Faria spent 3.5 years wrongfully imprisoned before acquittal. Fair and Just Prosecution and the Innocence Project are pushing post-conviction review, and the Illinois Supreme Court has agreed to review Jessica Logan's conviction. A clean record today is the appellate claim tomorrow.

Groundwork matters here. Getting a word-level, diarized transcript of the call before you interpret anything is step one, and the same discipline applies to jail call transcription when the state's theory turns on what a defendant supposedly said. For teams working through 911 call analysis in discovery, a grounded 911 call transcript analysis workflow keeps the actual words, not a checklist gloss, in front of the trier of fact.

Start With the Words: Getting a Defensible 911 Call Transcript

Every 911 call that ends up in discovery is a piece of evidence that can convict or clear your client, and the first move is making sure the transcript reflects the actual audio rather than someone's interpretation of it. Word-level timestamps and speaker diarization are the ground-truth step that lets you annotate the call against the state's claimed indicators instead of accepting a checklist gloss. Without that foundation, you're cross-examining a summary, not the call itself.

The Harpster checklist operates on impressions of cadence, grammar, and phrasing, with specific claims like "bleeding" signaling innocence and "blood" signaling guilt, or 80% of callers who say "need help" being innocent. Those are falsifiable propositions. If your transcript doesn't preserve the exact word, the pause before it, and who said it, you can't test the proposition. A 911 call transcript analysis with timestamps and diarization puts the actual language in front of the trier of fact, not a gloss. The same discipline applies to jail call transcription when the state's theory depends on what a defendant supposedly said and when.

Pull the audio from discovery first, not the transcript the state hands you. Then build a working transcript with word-level timestamps so you can mark every claimed indicator against the audio: the "Huh?" that the original thesis flagged as a guilt signal, an isolated "please," comments about the victim's eyes, the moment the caller acknowledges death. Each marker becomes a potential cross-examination point.

The replication record matters here. At least six peer-reviewed studies have failed to replicate the checklist findings, and five independent studies, including two FBI studies, contradicted the method. The underlying evidence base is roughly 200 calls across a couple of small studies. When a witness testifies that a call "sounded staged," you want the transcript open beside the audio so you can ask the witness to point to the exact second that supports the opinion. If they can't, the opinion doesn't survive scrutiny.

Annotate the transcript against the state's claimed indicators before you do anything else. That's the discipline that turns a 911 call from a prosecutor's narrative prop into evidence you can actually use.