Case narratives for data packages, drafted from your logs and QC summaries.

Today a QA officer or senior project manager pulls the receipt checklist, the custody records, the run and preparation logs, the QC summary forms and every nonconformance or corrective action record for the batch or sample delivery group. Then they write up, method by method, what happened and which samples it touched, and pass the narrative to whoever signs the package.

In the proof of concept, AI reads copies of the same records and drafts the narrative in your template: receipt conditions, holding times, each method's QC exceptions, the qualifiers your report already carries, and the corrective actions your records describe. Every sentence names its source, and anything the records leave unexplained goes on a list for your reviewer. Nothing is signed, released or sent.

The receiver decides what it must say. EPA's Contract Laboratory Program sets it out in SFAM02.1, the Superfund analytical methods statement of work EPA posted in April 2026: a document labeled SDG Narrative, carrying the lab's name, the SOW, contract, case and SDG numbers and any modified analysis number, then detailed documentation of every QC, sample, shipment and analytical problem. It also asks for items a lab wouldn't think to include: the analytes run when only part of the target list was scheduled; each GC column by brand, internal diameter, length, coating and film thickness; whether interelement or background corrections were applied on ICP; the collision or reaction cell settings on ICP-MS; and any difference from preliminary results already sent. The draft starts from the receiver's own list, so a missing item shows up as a gap, not a silence.

Receipt problems are recorded, not resolved. The same SOW names receipt events the narrative must carry: samples arriving more than a day late but still at or below 10 °C; more than half but less than nine tenths of the required amount, run at reduced volume; volatile vials with headspace when enough clear vials remain; broken or partly filled containers; legible handwriting on the custody record or labels. A metals sample found at pH 2 or above is acidified, held at least 16 hours before preparation, and noted. Where EPA's sample support staff gave a resolution, the narrative records both the problem and the answer. EPA's national functional guidelines send validators to the lab's narrative for these things, and for example calculations and manual integrations, and count a bubble bigger than a pea, about 6 mm across, as headspace. The draft quotes your receipt checklist and custody record and never decides whether a sample was fit to run.

Two kinds of qualifier, two different people. In the CLP's electronic format the lab's own flags are U, J, B, E and D, with Y and Z left for the lab to define, and those definitions belong in the narrative. Validation qualifiers come later and from someone else: the national functional guidelines of November 2020 give validators J+ and J- for results that may be biased high or low, UJ for a non-detect whose limit is approximate, and R for rejected data. Texas guidance likewise leaves usability qualifiers to the data usability reviewer. So the draft lists each lab qualifier exactly as your report printed it, defines any lab-defined flag from your SOP, and never writes a validation qualifier.

States add their own forms. Texas's TRRP-13 guidance, revised May 2010, has every lab data package carry a laboratory review checklist, exception reports and reportable data items numbered R1 to R10. An exception report goes with each No or not-reviewed answer on the checklist, and with each analyte, matrix and method the lab isn't accredited for in Texas; it is tied to specific samples and names the QC parameter that failed and the samples affected. The release statement is signed by the lab manager or a designee, with printed name, title and date. The draft writes exception reports only for answers your staff already gave, and it never answers a checklist question or touches the release statement.

The same QC goes by several names. 40 CFR 136.7 lists twelve QC elements a lab's SOP must cover when a Part 136 method lacks its own, and it names several of them twice: a laboratory reagent blank is also a method blank, a laboratory fortified blank is a spiked blank or laboratory control sample, and a laboratory fortified matrix may stand in for a matrix spike. One of the twelve is corrective action with root cause analysis. The draft uses your template's terms throughout and copies each corrective action as your records state it, never supplying a cause they don't give. Your reviewer will check every exception against the QC forms, and every sample number, qualifier and corrective action against the narrative your lab signed.

If your reviewer says yes, new packages get a drafted narrative the same way for about 30 days.

Your QA officer edits and signs every narrative, and the lab manager's release of each package stays exactly where it is today.

The side-by-side goes on through the month. Any next job waits until this one is judged, and it is priced on its own.