"Defensible disposal" is one of those phrases that gets used frequently in e-discovery circles and defined precisely almost never. In our experience, the ambiguity isn't accidental. Disposal decisions carry legal exposure, and vague language gives everyone plausible deniability. But when a TAR workflow is involved, the non-review population, those documents the predictive coding model scored too low to warrant attorney eyes, demands a clear documented standard. Not vague. Not aspirational. Actual records that hold up.
This article lays out what that documentation looks like, why platform-agnostic approaches matter, and what recall rate evidence you need before any disposal decision is defensible.
What the Non-Review Population Actually Is
In a TAR workflow, the collection is typically split into two populations after the active learning phase concludes. The review population is the set of documents above the confidence threshold. These get attorney eyes. The non-review population is everything below. These documents receive a disposition, either explicit or implicit, based on the model's assessment that they are unlikely to be responsive.
Here's the thing most firms don't say clearly: the non-review population is not the same as the non-responsive population. It's the population the model predicts is non-responsive, with a certain degree of error. That error rate is what recall validation is designed to quantify, and it's what disposal documentation has to address head-on.
A TAR workflow that produces an 88 percent recall rate, which is the threshold many practitioners treat as adequate for matters where completeness is a priority, means roughly 12 percent of responsive documents remain in the non-review population. Whether that's acceptable depends on the matter. But the decision has to be documented as a decision, not buried in process.
The Documentation Standard: What Audit-Ready Records Look Like
Audit-ready disposal documentation isn't complicated, but it has specific components that can't be skipped. In our tracking across matters on Relativity, DISCO, Everlaw, and Reveal, the firms whose TAR protocols hold up in court have records that include all of the following:
- QC sampling methodology: 95 percent confidence, plus or minus 2 percent margin of error is the standard. That requires a sample of approximately 2,401 documents from the non-review population. The sampling method (random, stratified, systematic) and seed must be documented.
- Sample results: The actual coding of each sampled document, with the reviewer's identity and coding timestamp. Not a summary. The record.
- Recall rate calculation: The estimated recall rate derived from sample results, with the statistical formula used and the resulting confidence interval.
- Protocol decision: The explicit statement that, based on the QC results, the non-review population was determined to be acceptable for disposal under the agreed TAR protocol, signed by the supervising attorney.
- Production record: Which documents from the review population were produced, and when.
Five components. Every matter. No exceptions. If any one of these is missing when opposing counsel challenges the protocol, you're in a significantly weaker position than the firm that has all five.
Platform-Agnostic Means Documentation That Travels
One underappreciated aspect of disposal documentation is that it needs to work regardless of which platform the TAR workflow ran on. This matters more than it sounds.
Real talk: matters move. Firms change platforms between phases of discovery. Platforms sunset features, get acquired, or change how they export reports. A disposal documentation package that lives entirely within a platform's native audit log is fragile. When you need to produce it three years later for a sanctions motion, you need it to exist outside the platform in a format any attorney can read.
Platform-agnostic documentation means exporting the QC sampling records, recall calculations, and protocol decision memos as standalone files, PDF or structured CSV, and storing them with the matter record. Not leaving them in the platform. The QC report from Relativity's Analytics module is useful internally. What you need for a court exhibit is a document you control.
Discovarc generates disposal documentation packages designed for this purpose: structured records that capture the full QC sampling methodology, coded results, and recall calculation in formats that export cleanly to PDF for the matter file. The underlying data stays in the platform. The documentation package doesn't.
Recall Rate Documentation: What Numbers You Need and Why
The recall rate tells you what percentage of responsive documents in the total collection ended up in the review population. A 90 percent recall means 9 out of every 10 responsive documents were caught. One in ten was not.
What the documentation needs to show is not just the number, but how you got there. The calculation should be reproducible from the documented sample. That means recording: the size of the non-review population, the sample size drawn, the number of responsive documents found in the sample, and the extrapolation methodology used to estimate overall recall.
Fact: a 95 percent confidence interval with plus-or-minus 2 percent margin of error on a sample of 2,401 documents means that if the sample shows 1.5 percent responsiveness in the non-review population, you can state with 95 percent confidence that the true responsiveness rate is between 1 percent and 2 percent. That's a range. Document it as a range, not a point estimate. Point estimates invite challenges that ranges survive.
Courts that have evaluated TAR protocols, beginning with Da Silva Moore and running through more recent matters, have consistently asked for this kind of statistical rigor. Absent it, the recall documentation doesn't do the work you need it to do.
Practical note: recall documentation is most defensible when it's generated before disposal, not reconstructed after. The QC sample should be drawn, coded, and calculated as a formal step in the review workflow, with results signed off by supervising counsel before any disposal action is taken.
Disposition Documentation for Production
Disposal documentation doesn't end with the non-review population. The production population needs its own disposition record. What was produced, to whom, on what date, in what format, and under what bates numbering scheme are all elements that belong in the matter record.
Production disposition documentation also needs to capture whether any withholdings were made from the review population on privilege, relevance, or other grounds, and the basis for those withholdings. A privilege log is the formal mechanism, but the connection between the TAR protocol, the review population, and the privilege log needs to be explicit in the disposal package, not implicit.
In our experience, the production disposition record is the piece firms most often treat as a downstream concern, something for the end of the matter. That's backwards. Building it as the production is assembled, not reconstructing it afterward, is where the defensibility lives.
Why This Matters More Than Platform Choice
Litigation support firms sometimes spend considerable time debating which platform, Relativity, DISCO, Everlaw, Reveal, produces the best TAR results. That's a reasonable conversation. But in our data, the disposal documentation gap is not primarily a platform problem. It's a process problem.
We've seen matters on every major platform with excellent QC sampling records and clear recall documentation, and matters on every major platform with essentially nothing on file. The platform doesn't write the protocol memo. The platform doesn't sign off on the disposal decision. People do. And what distinguishes defensible disposal from indefensible disposal is whether those people treated documentation as a first-class deliverable rather than an afterthought.
If your current TAR workflow produces a recall rate report but doesn't generate the QC sampling records, recall calculation, and protocol decision memo as exportable artifacts, the documentation gap is yours to fill. The good news is that it isn't technically complex. It's a process commitment. Make it once, standardize it across matters, and the defensibility follows automatically.
