The 'Shadow Forensics' Problem: How 911 Call Analysis Slips Past Daubert

If a prosecutor can't get a junk science method past Daubert, the next-best move is to make sure nobody treats it as science at all. That's the playbook behind 911 call analysis, a technique that has quietly appeared in more than 100 cases across 26 states despite being widely discredited.

The method itself is simple to describe and impossible to validate. Its creator, Tracy Harpster, claims he can identify a guilty caller by "indicators" like low emotion, specific phrasing, or asking for police instead of an ambulance. The foundational study examined only 100 homicide calls, roughly two-thirds from Ohio departments, and at least eight subsequent studies have failed to replicate its findings. Departments pay up to $3,500 for an eight-hour seminar teaching officers to do exactly this.

Here's the part that matters for your motion practice. Courts have repeatedly rejected 911 call analysis when offered as expert testimony under Daubert and Frye. So prosecutors stopped offering it that way. Instead, they put a trained officer on the stand who testifies about "guilt indicators" based on "training and experience," which sounds like ordinary lay opinion under FRE 701. As Christopher Lau writes in the Cornell Law Review, the technique is "particularly pernicious because it operates in the shadows": the founders are never proffered as experts, the underlying methods are never disclosed, and the jury hears subjective interpretations of caller behavior without anyone formally challenging the science.

That's the threat. Not an expert admission you can litigate, but a lay-opinion veneer you have to strip off yourself.

• The officer's "training and experience" is the seminar, not a validated methodology.
• The "COPS Scale" behind the 38 guilt indicators is proprietary and not publicly available.
• The bias runs backward: suspicion is driven by the absence of expected emotion, not evidence of guilt.
• The consequences are real. Huwe Burton served over 20 years after being deemed "not emotional enough." Robert Roberson spent over 20 years on Texas death row because his "flat affect" read as guilt; it was autism.

Dr. Jessica Salerno's research found that even moderate emotion wasn't enough to lower officers' suspicion; callers had to sound highly emotional for perceptions to shift. The method penalizes people who don't perform grief or panic on cue, whether the reason is shock, trauma, culture, or neurodivergence.

Practically, this means you can't wait for the prosecution to flag the method in discovery. Review the police report and bodycam for Harpster-style language ("calm and collected," "no emotion," requests for police rather than an ambulance), then move to compel disclosure of any 911 call analysis training the testifying officer received. A timestamped, diarized transcript is the fastest way to check whether the report's characterization of the caller actually matches the audio, and it's the kind of work a grounded review workspace like Evidence Copilot is built for, with 911 call transcript analysis and jail call transcription cited to the exact second. When the officer testifies that the caller sounded unemotional, you want the audio in front of you, not just the report's summary of it.

Why the Science Fails: The 'Low Emotion' Bias and Eight Failed Replications

Every "guilt indicator" an officer might cite from a 911 call traces back to a single study of 100 homicide calls, and roughly two-thirds of those calls came from Ohio police departments. That's the entire empirical foundation of 911 call analysis, a method now used in more than 100 cases across 26 states. When you cross-examine a trained officer, that sample is your starting point.

Since that original study, at least eight scientific studies have tried to replicate the findings. They failed. The supposed indicators of guilt do not hold up under scrutiny, and even John Jarvis, the FBI co-author of the original study, has said he was uncomfortable with the method being used in real cases and that "there's no definitive answer as to whether this is useful."

The method itself is a moving target. It started as 16 factors in a 2006 thesis and grew into 38 indicators organized into the proprietary "COPS Scale" by 2017, according to legal scholarship analyzing its admissibility. The scale's metrics aren't publicly available, which makes meaningful challenge difficult if you don't know the terrain.

Here's what the psychological research actually shows: suspicion is driven by the absence of expected emotion, not by guilt. Research by Dr. Jessica Salerno found that police officers stayed suspicious of callers showing even moderate emotion; a caller had to sound highly emotional for perceptions to shift. The method penalizes people who don't perform grief on cue.

That failure mode has names attached to it:

• Huwe Burton, convicted at 16 partly because he was deemed "not emotional enough," served over 20 years before exoneration.
• Robert Roberson, whose "flat affect" was treated as guilt on Texas death row for over 20 years, later explained by an autism spectrum disorder diagnosis.
• Shock, trauma, cultural norms around expressing distress, and neurodivergence can all make a caller sound "flat" to an officer trained to hear guilt.

Vincent Denault, a nonverbal communication expert, put it plainly: there is "no sign comparable to Pinocchio's nose in real life." No verbal cue reliably separates a guilty caller from an innocent one in shock.

Ask the officer how they accounted for trauma, culture, or neurodivergence before scoring your client's emotional display. They won't have an answer, because the training doesn't cover it. To prepare, you need the call itself transcribed and compared against the police narrative; if the report says "calm and collected" but the audio shows distress, that discrepancy is your cross. Evidence Copilot's 911 call transcript analysis produces timestamped, diarized transcripts for exactly that comparison, and jail call transcription lets you check post-arrest statements for consistency with the original call.

Spotting 911 Call Analysis Hiding in Your Discovery

Nobody is going to hand you a document labeled "911 Call Analysis was used in this case." That's the whole point of what Cornell Law Review author Christopher Lau calls "shadow forensics": the method operates in the shadows precisely because it can't survive a Daubert hearing. If you wait for the prosecution to flag it, you'll hear it for the first time at trial, dressed up as an officer's "training and experience."

So go hunting. The technique has surfaced in more than 100 cases across 26 states, according to ProPublica's reporting, and departments pay up to $3,500 for an eight-hour seminar teaching officers to read guilt into a caller's voice. The indicators bleed into police narratives even when nobody names the method. Here's what to search for in your reports, bodycam footage, and 911 audio:

• Affect language in the report: "calm," "flat," "unemotional," "coldly described the scene." These are the exact cues from the proprietary COPS Scale, which grew from 16 factors in a 2006 thesis to 38 indicators by 2017.
• Phrasing commentary: notes that the caller said "blood" instead of "bleeding," or described a victim as "this person" rather than a name.
• Request framing: observations that the caller "asked for police, not an ambulance," or failed to request medical help at all.
• Any officer or dispatcher opinion about what the caller "should have" sounded like, felt, or said.

When you find these markers, don't just note them. File a motion to compel disclosure of any 911 Call Analysis training the responding officers or the case detective received, including seminar materials, invoices, and attendee rosters. The underlying methods are never disclosed voluntarily, and the training record is what converts a hunch into a discoverable fact.

Then do the comparison work. Pull a timestamped, diarized transcript of the call and set it against the report narrative line by line. If the report claims the caller was "calm and collected" but the audio shows hesitation, crying, or a dispatcher repeatedly asking for clarification, you have impeachment material that predates any cross-examination. A grounded 911 call transcript analysis workflow makes that comparison fast, and Evidence Copilot's ingest signals surface report-versus-audio inconsistencies before anyone asks. The same discipline applies downstream: jail call transcription lets you check whether post-arrest statements actually track the initial call or whether the "inconsistency" only exists in the officer's retelling.

Remember the stakes behind the paperwork. Robert Roberson spent over 20 years on death row partly because a "flat affect" was read as guilt; his autism spectrum diagnosis came later. The Innocence Project documents at least eight failed replication studies, and the FBI co-author of the original study himself said there's "no definitive answer as to whether this is useful." Find it in discovery before the state makes it sound like common sense at trial.

Killing It in Court: Motions, Cross-Examination, and the Lay-Testimony End Run

The detective never took the stand as an expert. That's the whole problem. Because 911 call analysis keeps failing Daubert and Frye scrutiny when proffered as science, prosecutors sidestep the fight entirely and elicit the same conclusions through a trained officer's "training and experience" as lay opinion. Christopher Lau's Cornell Law Review piece calls this "shadow forensics," and it's the single most important dynamic to understand before you draft a motion.

Your move is the FRE 702 end run. Argue that testimony interpreting a caller's emotional state against a learned framework of guilt indicators is expert testimony in disguise, and demand the full Daubert analysis the prosecution is trying to avoid. Collin Derrig's UC Law Review article makes the case directly: the method fails on testability, peer review, and general acceptance, so the officer's seminar attendance can't launder it into admissible lay opinion. NACDL President Nellie King has put the organization squarely on record calling the technique part of the "junk pile of fake scientific theories," which gives you institutional backing on the general-acceptance factor.

Attack the methodology's opacity, too. The original Harpster study rested on just 100 homicide calls, roughly two-thirds from Ohio departments, and the framework has since ballooned into the proprietary 38-indicator "COPS Scale" that nobody outside the training circuit can independently examine. At least eight subsequent studies have failed to replicate the findings. Even John Jarvis, the FBI co-author of the original study, has said there's "no definitive answer as to whether this is useful."

On cross, hammer the "low emotion" bias. Dr. Jessica Salerno's research at Cornell shows suspicion is triggered by the absence of expected emotion, and that officers needed callers to sound highly emotional before suspicion dropped. Build your questions around everything that produces a flat affect besides guilt:

• Acute shock and dissociative trauma responses in the minutes after discovering a body
• Autism spectrum disorder: Robert Roberson spent over 20 years on death row before his "flat affect" was explained by an autism diagnosis
• Cultural norms that suppress public displays of grief or panic
• The Huwe Burton case, where a 16-year-old was deemed "not emotional enough" and wrongfully served over 20 years

Ask the officer which of these factors the COPS Scale accounts for, and watch them admit it doesn't. Vincent Denault's line that there's "no sign comparable to Pinocchio's nose in real life" is worth quoting in briefing.

Practically, you can't cross what you haven't heard. A timestamped transcript of the call, compared line by line against the police narrative and bodycam, is where impeachment lives. Tools like Evidence Copilot's 911 call transcript analysis surface those report-versus-audio discrepancies, and pairing it with jail call transcription lets you test post-arrest statements against the original call.

Finally, watch your jurisdiction. North Carolina's Office of Indigent Defense Services has formally warned defenders to watch for 911 call analysis in their cases, and the method has shown up in over 100 cases across 26 states. If your state is on that list, a motion in limine raising the FRE 702 argument before trial beats objecting mid-testimony every time.

A Working Workflow: Reviewing the Call, the Report, and the Bodycam Together

The fastest way to catch "shadow forensics" is to line up three things side by side: the 911 audio, the police report, and the bodycam. When you review them as one record instead of separate exhibits, the gaps show themselves.

Start with the transcript. Get a diarized transcript with word-level timestamps for the 911 call, then listen while you read. Rename speakers once and confirm who said what, because dispatch audio is messy and the report will smooth it over. If the report says the caller was "calm and collected," check the actual seconds. Hesitation, a shaky voice, or a rushed answer at 2:14 is your impeachment, and it only works if you can point to the exact moment.

Next, run the report against the call. You're looking for the guilt indicators the method teaches: "lack of emotion," phrasing like saying "blood" instead of "bleeding," or a request for police instead of an ambulance. As Christopher Lau notes in the Cornell Law Review, these indicators often surface in police narratives before anyone discloses that an officer attended a Harpster-style seminar. If you find them, that's your motion to compel disclosure of the training.

Then bring in the bodycam and jail calls. Sync the bodycam to the 911 timeline so you can see what officers did versus what the report says they did. Review jail calls for post-arrest statements and compare them against the initial call; jail call transcription with timestamps makes that comparison possible at case scale. The Innocence Project's research on the "low emotion" bias gives you the cross-exam frame: ask the officer how they accounted for shock, trauma, culture, or neurodivergence before deciding the caller sounded guilty.

A grounded workflow looks like this:

• Ingest everything into one case: 911 audio, bodycam, jail calls, and the report PDF.
• Review diarized transcripts with word-level timestamps; confirm speaker labels yourself.
• Check each report claim against the call and footage, quoting both sides.
• Draft cross-exam questions where every question ties to a quoted source moment, then review every AI output yourself before it goes in a memo.

That last point matters. Tools like 911 call transcript analysis can flag a report-vs-call discrepancy before you ask, but the attorney decides whether it means anything. Evidence Copilot marks every AI output as pending human review for exactly this reason: at least eight replication studies have shown how badly intuition performs on 911 calls, and your review should be the opposite of that. Every claim you make at trial should trace back to a specific second of audio or a specific page of the report.