The Pseudoscience in Your Discovery Dump

The phrase "911 call analysis" sounds clinical, almost forensic. In practice, it's a proprietary guilt-detection method that has quietly shaped thousands of investigations, and it's probably sitting in your discovery dump right now. Detective Tracy Harpster built the technique from a 2006 thesis examining 100 calls, identifying 16 guilt indicators that grew to 18 in a 2009 paper and 38 in his 2017 book, all organized under the COPS Scale (Considering Offender Probability in Statements). As of March 2024, Harpster had consulted on at least 2,500 homicide investigations, and the method has surfaced in at least 26 states, according to University of Cincinnati Law Review reporting. He has never testified as an expert witness in a criminal trial.

That gap matters. Because the method never faces Daubert or Frye gatekeeping in open court, prosecutors instead guide juries through the audio and transcript themselves, mixing Harpster's framework into lay testimony. One law review analysis describes the practice as sneaking the technique past expert-admissibility standards, which means the "analysis" reaches jurors without ever being tested.

The pipeline from caller to suspect runs through demeanor. Cornell research involving roughly 1,800 civilian and 300 law enforcement participants found that four out of five participants cited emotion or urgency as affecting their suspicion, and that poor communication, guardedness, and impression management raised it further, per Cornell's findings. Once suspicion attaches, confirmation bias tends to do the rest. The real-world results are sobering:

• A 2018 Washington case where a baby's death was reclassified as homicide partly because the mother's first word was "hi," read as suspicious politeness.
• Robert Roberson, sentenced to death row in Texas after reporting his daughter's death in a manner perceived as oddly unemotional. He has severe autism.
• A 2012 Fremont, Ohio defendant who pled to manslaughter and received 11 years after being shown Harpster's 911 call analysis handout.

Defense challenges do land. The Michigan murder conviction of 16-year-old Riley Spitler was overturned because 911 call analysis and other statement-analysis techniques should never have been admitted. At least six peer-reviewed studies have failed to replicate the police-training findings underpinning the method. When you're reviewing a 911 call transcript analysis in discovery, treat the analysis itself as a challengeable exhibit, not background noise.

Why Courts Keep Letting It In (and How Davis v. Washington Helps You Push Back)

If a 911 call comes into your discovery dump, the first question isn't whether the caller sounds guilty. It's whether the call is testimonial under Davis v. Washington, because that answer controls whether the state can play the recording without putting your client's accuser on the stand.

The 2006 Supreme Court ruling drew a "fundamental difference" between calls made during an ongoing emergency and statements to police after danger has passed. Calls during an ongoing emergency are non-testimonial and admissible even if the caller never testifies. Statements made after the danger passed are testimonial, and Justice Scalia's majority opinion was blunt about why: they "do precisely what a witness does on direct examination. They are inherently testimonial," because "no witness goes into court to proclaim an emergency and seek help" (NPR).

D.C. public defender Timothy O'Toole read the decision narrowly, in your favor: once the call extends beyond "name, rank, and serial number information," the court signaled it may not be admissible without live testimony subject to cross-examination. Attorney James Wisman predicted the fight would continue case by case, "was there an emergency, was this person just describing past events in a way that looks like they're just creating evidence for a trial?" That's your motion to limine in miniature. Map the call second by second: when did the emergency end, and when did the caller start narrating history for an audience?

Here's the workaround you're more likely to face. Analysts trained in "911 call analysis," the Harpster method used in at least 26 states, have never testified as experts in a criminal trial, even though its developer has consulted on at least 2,500 homicide investigations. Instead, prosecutors "guide juries through both the audio and transcript of the calls," what the University of Cincinnati Law Review describes as mixing the technique "into in-court statements and lay testimonies to sneak it past the standards for expert testimony."

That's where FRE 701 does your work for you. Lay opinion must be rationally based on the witness's perception, helpful to the trier of fact, and not based on scientific, technical, or specialized knowledge within FRE 702. The moment a detective walks the jury through "guilt indicators" like a suspicious first word or odd politeness, they're delivering expert conclusions under a lay label. Object on that basis, and force the state to either drop the analysis or defend it under Daubert or Frye, where the methodology's untested data, unknown error rates, and failure to replicate in at least six peer-reviewed studies become dispositive.

When you're prepping that objection, having the transcript and audio pinned to exact timestamps matters. Tools like Evidence Copilot produce diarized, word-level 911 call transcript analysis so every challenged segment in your motion cites the precise moment it came from, and similar jail call transcription review keeps the rest of your statement evidence grounded the same way. The framework is on your side. You just have to make the court apply it.

Daubert, Frye, and the Five Red Flags That Expose the Method

If the prosecution's case leans on how your client sounded on a 911 call rather than what they said, you're facing "911 call analysis," a methodology that has never passed scientific muster in any meaningful sense. Your job is to force it through the gatekeeping it has always managed to dodge.

Start with the Daubert five-factor test: whether the technique can be and has been tested, whether it has been published and peer reviewed, its known error rate, whether standards control its operation, and its acceptance in the relevant scientific community. Per the University of Cincinnati Law Review, 911 call analysis fails on nearly every factor: the underlying data has never been made public, error rates are unknown, and no operational standards exist. Frye states are an even easier win, since the standard requires only that a methodology be "widely accepted and used in a field," and this one isn't.

The Expert Institute's five red flags for unreliable expert methodologies map onto this technique almost line by line:

• Proprietary technique. Detective Tracy Harpster's COPS Scale grew from a 2006 thesis of 100 calls to 38 guilt indicators by 2017, with the raw data never released for independent verification.
• Failure to apply the scientific method. No independent testing, no replication. In fact, at least six peer-reviewed studies failed to replicate the underlying police training findings, according to Cornell research.
• Selective citation. The indicator counts kept shifting (16, then 18, then 38) with no published validation of any version.
• Statistical misuse. Claims drawn from tiny samples (100 calls in the thesis, 200 in the book) presented as predictive of guilt.
• Circular reasoning. "Guilty callers" were identified by outcomes, then their language was used to predict guilt.

Watch for the workaround. Prosecutors increasingly skip expert designation entirely and simply "guide juries through both the audio and transcript of the calls," sneaking the methodology past gatekeeping as lay commentary. That violates FRE 701, which bars lay opinions based on scientific, technical, or specialized knowledge. Object every time a witness starts interpreting "suspicious" word choices.

The demeanor trap deserves its own prep. Cornell's research on roughly 1,800 civilian participants found that four out of five cited emotion or urgency as affecting suspicion, yet people communicate in wildly different ways due to neurodivergence, disability, culture, and stress. The Robert Roberson case, where severe autism was read as "oddly unemotional" reporting of a child's death, shows the stakes. Investigate your client's caller characteristics early, and build the record that explains the demeanor before the state weaponizes it.

When you're working through the call audio and transcript against the police report, a grounded review workspace helps. Evidence Copilot produces timestamped, diarized transcripts of 911 audio and jail calls, with every AI answer cited to the exact moment, flagged for human review. The same workflow applies to related call evidence, including jail call transcription, so you can quote the state's own framing back at it with citations attached.

The method has already cracked under pressure. In Michigan, legal scholarship documents how 16-year-old Riley Spitler's murder conviction was overturned because 911 call analysis and similar statement-analysis techniques should never have been admitted. Build your motion the same way: five factors, five red flags, and a transcript that shows the "indicators" are nothing but ordinary human stress.

Building the Motion and the Cross-Exam

The motion to exclude 911 call analysis is where the Spitler reversal earns its weight. A 16-year-old's murder conviction was overturned in Michigan because the methodology should never have reached the jury. That precedent travels.

Start with a Daubert challenge built on five factors. The technique has never been independently tested. Harpster's raw data has never been made public. Error rates are unknown. No operational standards exist. At least six peer-reviewed studies failed to replicate the underlying police-training findings on which the COPS Scale rests. Each factor cuts the same direction: this is not science.

If your jurisdiction follows Frye, the argument is shorter. Statement-analysis indicators invented by a single detective and adopted in 26 states without independent validation do not meet "general acceptance" in any recognized scientific community. Harpster himself has never testified as an expert in a criminal trial, which tells you something about how the methodology holds up under cross.

Then attack the workaround. Prosecutors often avoid expert-witness gatekeeping by guiding jurors through the audio and transcript themselves. That lay-witness presentation violates FRE 701's prohibition on opinions based on specialized knowledge reserved for FRE 702. Object when the prosecutor pauses the recording to explain why a pause, a word choice, or a flat tone is "suspicious." That testimony is expert opinion dressed as common sense.

For cross-examination, build questions around the five red flags for unreliable methodology: proprietary techniques, failure to apply the scientific method, selective citation, misuse of statistical language, and vague or circular reasoning. Ask the witness to define "passive resistance." Ask for the error rate. Ask which peer-reviewed journals have validated the 38 indicators. The answers expose the house of cards.

The demeanor evidence cuts deepest when you investigate the caller. Cornell research confirms that neurodivergence, PTSD, intellectual disability, and language barriers all shape how distress sounds on a recorded line. Robert Roberson sat on death row in Texas because severe autism read as "oddly unemotional" when he reported his daughter's death. That is the template. Subpoena school records, mental-health history, language assessments, and prior 911 calls from the same caller. A caller who sounds "guarded" on this call may sound guarded on every call.

Once you have the transcript diarized with timestamps and the caller's history mapped, the cross-exam writes itself. When the prosecutor plays the recording and points to a two-second silence, you can respond with the medical record that explains it. 911 call transcript analysis that surfaces these moments quickly gives you the time to build the human context that dismantles the pseudoscience.

For the parallel jail-call side of the case, the same discipline applies. Jail call transcription with speaker diarization lets you track who said what and when, so you can flag any inconsistent statements before the state does and turn them into mitigation rather than impeachment.

Getting a Usable 911 Call Transcript Into Your Case File

The raw MP3 from the 911 center is useless in a hearing if you can't point the judge to the exact second the caller says the thing that matters. Everything downstream, from a Davis/Hammon admissibility argument to impeachment of an officer's report, depends on having a timestamped, speaker-labeled transcript in your case file.

Start with the audio itself. Request the original recording from the PSAP, not just the version embedded in the discovery dump, and upload it alongside everything else in the case. A workspace like Evidence Copilot produces diarized transcripts with word-level timestamps, so when the dispatcher's questions drift from "name, rank, and serial number" territory into narrative-gathering, you can cite the precise moment that shift happens. That matters because the Supreme Court drew its line in Davis v. Washington between calls made during an ongoing emergency and statements made after the danger has passed, and the "was this an emergency?" question gets litigated call by call, minute by minute.

Once you have the transcript, cross-check it against everything else in discovery:

• Compare the caller's statements to the responding officer's police report. If the report paraphrases the call in a way that adds or drops detail, that's an impeachment gap with a timestamp attached.
• Sync the call against bodycam footage of the arrival. What the caller said should match what officers saw; discrepancies cut both ways.
• Flag any portion where the caller is describing past events rather than a present emergency, because that's the testimonial-side argument.
• Note the caller's demeanor markers only as context. Cornell research on how 911 callers become suspects found roughly 1,800 civilian participants and about 300 law enforcement participants, with 4 out of 5 citing emotion or urgency as affecting suspicion, which is exactly the kind of demeanor-based inference you want documented and challenged, not adopted (Cornell Chronicle).

This is also where you catch the quiet version of 911 call analysis showing up in your case. Prosecutors in at least 26 states have "guided juries through both the audio and transcript of the calls" instead of calling an analyst as an expert, sneaking the Harpster methodology past Daubert gatekeeping, as documented in a University of Cincinnati Law Review analysis. If the State's narrative of the call tracks the COPS Scale indicators, you want your own transcript ready to object under FRE 701, which bars lay opinions based on specialized knowledge.

Evidence Copilot's 911 call transcript analysis workflow is built for this: auto-signals compare the call audio against earlier evidence in the case, surfacing candidate inconsistencies between the report and the call before you've thought to ask. The same applies to post-arrest communications; jail call transcription gives you the same timestamped, citable record for calls that often get handed over as barely-indexed audio files.

Every AI output stays marked as pending human review, and the originals stay immutable, so what ends up in your exhibit pack is a transcript you can defend line by line.