Chart Fabrication Is an Intake Problem, Not a Deposition Problem

MedLegal AI Editorial · April 18, 2026 · 9-minute read

Most plaintiff firms doing nursing-home work catch chart fabrication at deposition. A few catch it in trial prep. The best ones catch it at intake — before the case is accepted. The difference between those workflows is measured in paralegal hours per case and in the number of weak cases that never should have been taken.

The old workflow

For most of the last two decades the nursing-home chart-review process has gone like this:

1. Case comes in. Family retained the firm off a referral, or responded to a mailer.
2. Paralegal requests the facility chart. The chart arrives 2-8 weeks later as a thick stack of PDFs, sometimes with hand-redacted pages and occasional missing ranges.
3. Paralegal builds a chronology. 40-60 hours of work depending on record volume.
4. Attorney reviews the chronology, identifies the claim theory, drafts the complaint.
5. Case enters litigation. Records are requested again through discovery — this time in native format with metadata.
6. Differences between the two productions are noticed.
7. At deposition of the DON or the MDS coordinator, the attorney asks about the CNA flow sheet showing "q2h repositioning documented" on days when the heel ulcer went from stage I to stage III.
8. Sometimes the witness flinches. Sometimes the fabrication is conceded. Sometimes it's explained away.

This workflow is reasonable. It also buries the single most important piece of case-evaluation information — the strength of the fabrication evidence — until after the firm has already committed 60+ hours of work to the case. If the fabrication evidence turns out to be weak, the firm doesn't un-commit those hours. The case proceeds at reduced expected value.

What changed in 2024-2025

Three things. First, the medical-records LLMs that can actually handle 10,000-page records with scanned-quality pages matured. Second, HIPAA-posture-aware tooling (BAA-first, no training on uploaded records) reached production quality. Third, plaintiff-tuned pattern-detection models became available — not generic document-review AI, but models specifically tuned to flag the fabrication signatures plaintiff attorneys have been citing in motions for twenty years.

The consequence: the work that used to take a paralegal 40-60 hours of first-pass review can now be done in 30-90 minutes of compute, producing a structured list of anomalies the paralegal verifies rather than discovers from scratch.

That is a workflow reorder, not a workflow speedup. The paralegal's job changes from "read the chart and hope you notice the discrepancies" to "verify the 14 flagged discrepancies the tool surfaced." Different skill, different pacing, and critically — this can happen at intake instead of at discovery.

The intake-first workflow

Step 1: pre-retainer pattern scan

Family provides whatever records they have. Most families have death certificate, discharge summary, and maybe 50-200 pages of chart copies they obtained from the facility on their own. Even at this volume the pattern-detection can start. Same-minute flow-sheet entries. Impossible ADL improvement scores. MDS snapshots that don't match the nursing notes from the same week. These are visible in the fragments the family already has.

This takes 10-20 minutes of intake-coordinator time plus the compute run.

Step 2: conditional full-chart request

If the pattern scan raises zero or minimal flags, the firm has data to decline the case politely without sinking the paralegal hours. If the scan raises meaningful flags, the firm requests the full chart with a specific letter naming the anomalies already detected — signaling to the facility that the firm is paying attention, which sometimes changes the subsequent production.

Step 3: full-chart analysis before retainer

The full chart — once received — runs through the same pattern detection against the complete record. This is where MDS↔note↔Braden cross-checks go from indicative to definitive. The output is a one-page intake-briefing document: what was alleged, what the records show, where the fabrication evidence is strongest, and what additional discovery would lock it in.

A partner can read this in 15 minutes. The decision to sign the retainer is now based on the actual evidence, not on a family narrative and a thin chronology.

Step 4: discovery scoped precisely

When the case does proceed, discovery requests can be written tightly around the specific anomalies already identified — request for native-format EHR logs including edit metadata for the date range of the flagged entries, request for payroll and timekeeping records covering the CNAs who signed the flagged entries, request for MDS coordinator training materials and any internal audit memoranda.

Targeted discovery wins motions because targeted discovery reflects actual knowledge, not fishing.

The economics

A plaintiff nursing-home practice that takes 40 cases per year and does pre-retainer fabrication detection on each one is looking at 40 × (15-20 minutes intake + 30-90 minutes compute) = roughly 60-80 hours of combined human and compute time. The same practice doing the old workflow commits 40 × 40 = 1,600 paralegal-hours to chart review before case-selection decisions are made.

The bigger number is not the 1,520 hours saved. The bigger number is the decision to decline 8-12 cases per year that wouldn't have produced fabrication evidence strong enough to sustain punitive damages — cases that would have moved through litigation at reduced expected value, consuming partner time that could have been spent on the cases where the fabrication evidence was definitive.

This is an opportunity-cost argument, not a productivity argument. The firm's output isn't more cases; it's more strong cases, with fewer hours committed to weak ones.

What does not change

• You still want to meet the family. Pattern detection doesn't replace sitting across a table from someone who watched their parent die in a facility.
• You still want the expert review. An expert opinion on standard of care is not a records-parsing output.
• You still want the discovery workup. Native-format production, payroll cross-check, MDS coordinator depositions — all still matter. The difference is they're running against a case you already know how to win.
• You still want the complete chronology for trial. Intake pattern detection is not a chronology; it's a case-strength filter that sits upstream of the chronology.

The pushback, and why it is weaker than it sounds

"AI might miss something a good paralegal would catch."

Sometimes, yes. This is an argument for using AI as a first-pass filter and having the paralegal verify, which is the workflow described above. It is not an argument for skipping the first-pass filter. The alternative isn't "paralegal catches everything"; the alternative is "paralegal catches everything they have time to catch in 40 hours," which empirically is less than what a pattern-tuned model produces in 60 minutes.

"We don't want to decline cases at intake. The family deserves a chance."

Declining a case at intake with a clear "the records don't support the claim theory you described" conversation is a better outcome for the family than accepting the case, working it for 18 months, and settling for nuisance value. The family gets clarity sooner. The firm gets capacity to take the cases it can actually win.

"The tool will commoditize plaintiff work."

Pattern detection is a floor, not a ceiling. Firms that already do this work well by hand have their paralegal capacity freed up for the deeper analysis — the payroll cross-check, the MDS coordinator deposition strategy, the expert preparation. Firms that don't currently do this work well will, for the first time, have a baseline floor they couldn't hit before. Neither outcome commoditizes the partner's judgment, which is still what wins cases at trial.

How to start

If your firm isn't running pattern detection at intake today, the smallest viable version of the workflow is:

1. Pick two active cases and the family-provided records for each.
2. Run them through a records-analysis tool with chart-fabrication pattern detection. MedLegal AI is one option; there are others.
3. Compare the output to what the paralegal had already identified by hand.
4. Measure the false positive rate and the true-catch rate on items the paralegal hadn't flagged yet.
5. Decide whether the workflow reorder is worth it for your firm.

That's a 2-hour decision process. The cost of running the experiment is measured in dollars, not hours. The cost of not running it is a 2026 firm still operating on a 2005 intake workflow while competitors move upstream.