Recognizing Shadow Forensics in Discovery

The most dangerous forensic testimony you'll face in 2026 may never be labeled as forensic testimony at all. A 2026 Cornell Law Review analysis calls 911 call analysis "shadow forensics": a technique that claims to detect a caller's guilt from word choice, phrasing, and cadence, yet sidesteps the admissibility gatekeeping that would gut any ordinary expert opinion.

The evasion mechanism is structural, and it's worth understanding precisely. The technique's methods are never disclosed, and its founders (chiefly former detective Tracy Harpster) are not proffered as experts. Instead, the witness on the stand is a 911 operator or patrol officer who attended a training. Because no "expert" appears, the technique "largely escapes Daubert or Frye evidentiary challenge," according to the Cornell analysis by Christopher Lau of the Wisconsin Innocence Project, who compares it to debunked fields like bitemark evidence and hair microscopy.

What does this look like in discovery? Watch for testimony from a non-scientist witness about how the caller "requested help," their phrasing, or indicators of "guilt" in the call. ProPublica documented the technique's use in more than 100 cases across 26 states, and reported that some prosecutors disguised it at trial, knowing it would not survive recognition as scientific evidence. One Orange County detective wrote to Harpster that training helped his district attorney spot "indicators of guilt" and offered "suggestions on how to introduce the 911 calls to the jury during trial."

Practical screening steps for every file:

• Ask in discovery whether any officer, operator, or investigator attended call-analysis training, and when.
• Demand the training materials and any written methodology. The Cornell article identifies undisclosed methods as the core of the evasion.
• Flag any "credibility" or "behavioral" opinion about the 911 call that traces to such training rather than to personal perception.
• Listen to the call yourself against a timestamped transcript, so you can test every claimed "indicator" against what the caller actually said and when.

That last step matters because your rebuttal should rest on the record, not intuition. A grounded workspace like Evidence Copilot produces diarized, word-level-timestamped transcripts of the 911 audio through its 911 call transcript analysis service, with citations to the exact moment, and the same workflow extends to related recordings through jail call transcription. When the State's witness claims the caller's "delay" or phrasing signals deception, you can put the actual sequence and timing in front of the court, as defense counsel did in contesting the disputed 17-minute delay in the Jessica Logan case, which the Illinois Supreme Court agreed to review after ProPublica's reporting.

The institutional record is now on your side. NACDL has condemned the practice, North Carolina's Office of Indigent Defense Services issued a statewide warning, and the University of Michigan's STPP policy brief found the technique lacks robust empirical support while reinforcing investigative confirmation bias. Build your motion around both the science and the pathway: name the shadow-forensics mechanism, demand the methodology, and force the technique into the sunlight it has spent two decades avoiding.

Building the Admissibility Challenge

The hardest part of challenging 911 call analysis isn't the science. It's that the prosecution never presents it as science at all.

A 2026 Cornell Law Review analysis calls the technique "shadow forensics" because it is engineered to slip past evidentiary scrutiny. The underlying methods stay undisclosed, the founders never take the stand as experts, and instead the testimony comes from 911 operators or police officers who attended a training. That structure is what lets the technique "largely escape Daubert or Frye evidentiary challenge." Your motion has to attack that evasion mechanism directly, going beyond the validity of the conclusions.

Start by demanding disclosure. If the witness's opinion about the caller's word choice, phrasing, or "how they requested help" traces to Harpster-style training, ask for the methodology, the training materials, and the empirical basis for each claimed indicator. The corroboration record is on your side: researchers who tried to validate the technique's claims have failed, and the University of Michigan's STPP policy brief concludes it lacks robust empirical support while reinforcing investigative confirmation bias.

Then attack the witness's qualifications. An officer who sat through a seminar is not a scientist, and the cross-examination writes itself: no peer-reviewed validation, no error rates, no disclosed method. When the prosecution resists disclosure, that resistance itself becomes the Daubert record.

Ground the motion in the institutional record, which has moved decisively against the technique:

• The NACDL, through its president Nellie King, has condemned the practice as junk science that belongs "in the junk pile of fake scientific theories."
• North Carolina's Office of Indigent Defense Services issued a statewide warning to defense attorneys about the technique.
• Fair and Just Prosecution urged member prosecutors to review past convictions that relied on it.
• The Colorado Bureau of Investigation stated the technique should never be "anything more than a potential investigative lead."

The scale of the problem justifies the aggression. ProPublica documented the technique's use in more than 100 cases across 26 states, and reported that some prosecutors disguised it at trial against "unwitting defendants" because they knew it would not survive admissibility review. One Orange County detective even wrote to Harpster praising suggestions on "how to introduce the 911 calls to the jury during trial." If your case shows signs of disguised use, a motion in limine targeting the behavioral opinions rather than the labeled technique alone may be your only real shot at exclusion.

Your own record matters too. Build the challenge on a diarized, timestamped transcript of the call itself, so every disputed "indicator" can be checked against the audio second by second. A workspace like Evidence Copilot supports that workflow through 911 call transcript analysis with word-level timestamps and citations to the exact moment, and the same review extends to related recordings through jail call transcription. When the expert's claimed indicator has no verifiable anchor in the call, say so in the motion, with the timestamp attached.

Grounding Your Review in Timestamped Evidence

The fight over 911 call analysis is really a fight over what the audio actually contains. If the prosecution's theory rests on what your client said, when they said it, and how they asked for help, your rebuttal has to live in the call itself, not in argument about intuition.

Start with a diarized, word-level timestamped transcript of the call. The Logan case shows why: the disputed 17-minute delay before the 911 call, and the roughly seven hours between the child being put to bed and the call, became the contested factual spine of the case, according to ProPublica's reporting. You cannot cross-examine a delay you haven't pinned to the second.

A timestamped transcript gives you three concrete things to do:

• Impeach the operator or officer on the actual sequence of statements, rather than on a paraphrase from a police report.
• Build a Daubert record showing what the caller factually said, against which the "indicators of guilt" testimony can be tested.
• Compare the call against bodycam and police reports for inconsistencies before the preliminary hearing, not after.

The University of Michigan's policy brief warns that the 911 COPS Scale is "routinely used to assess the credibility and even guilt of 911 callers," reinforcing confirmation bias and pulling attention from reliable evidence. A word-level transcript is how you pull that attention back.

Authenticity now belongs on your checklist too. Synthetic audio is cheap in 2026: policing-trend forecasters note that deepfakes are accessible enough that "a teenager can fabricate a crisis in a matter of minutes," and agencies are adopting AI tools with rapid implementation and limited oversight. Ask for chain of custody on the call file, the original recording format, and any processing the audio underwent before you received it.

This is where grounded tooling earns its keep. Evidence Copilot's 911 call transcript analysis produces diarized, word-level timestamps with citations to the exact moment in the audio, so every line of your cross ties back to a second you can play in court. The same workflow carries over to jail call transcription when the state's call evidence extends past the 911 recording.

The rule for 2026 is simple: every claim about what was said on that call should be checkable against a timestamp. If yours are and theirs aren't, you control the record.