Court deadline calendaring with AI: extract the trigger, let docketing confirm the date

The most useful thing the workflow can extract is not the date. It is the structure: what event starts the period, how long the period is, what unit it uses, and which rule or order sets it. A draft entry that says "21 days after service of the motion, per paragraph 4 of the order" can be checked. A bare date cannot.

For each candidate entry, I would store the source document, the page, the quoted sentence, the trigger event, the period, and the proposed computed date as separate fields. When the trigger has not happened yet, the entry stays open rather than receiving a guessed date.

For federal civil matters, Federal Rule of Civil Procedure 6 sets the default method when a rule, local rule, court order or statute does not specify one. For a period stated in days, it excludes the day of the triggering event, counts every day including intermediate weekends and legal holidays, and includes the last day, with the period continuing to the next day that is not a Saturday, Sunday or legal holiday. Rule 6(d) adds three days after certain kinds of service, such as service by mail.

Those details are exactly why the computation should be visible. The draft entry should record which method was applied, whether additional days were added and why, and which holiday calendar was used. State courts, local rules and specific orders can differ, so the method is a property of the matter and the court, not a global setting.

I would not ask a language model to perform the arithmetic in free text. Extraction can be model-assisted; the computation should be deterministic code or the firm's existing rules-based docketing tool, so the same inputs always produce the same output and a reviewer can see the steps.

Every proposed entry should go to a named reviewer before it reaches the firm calendar. The review screen should show the quoted source sentence next to the draft, so the reviewer is checking the document, not the model's paraphrase of it.

The reviewer can accept, correct, or reject. Corrections should be recorded with the reason, because a pattern of corrections tells you where extraction or the rules configuration needs work. A rejected entry should record why, for example because the period was already calendared from an earlier order.

For high-consequence deadlines, many firms already require a second independent check. The workflow should support that practice rather than replace it. Two people confirming the same computed date from the same source is a stronger control than one person confirming a model.

Orders change. A new scheduling order may move some dates and leave others, and it may not say clearly which earlier deadlines it replaces. The workflow should compare a new order with existing entries for the same matter and flag every entry that might be affected.

I would not let the system silently move existing calendar entries. It should propose changes, link each proposal to the new source, and leave the earlier entry in place until a reviewer confirms. The history of what was calendared, when, and on what basis is part of the matter record.

Some entries require legal judgment: whether a period runs from service or from filing when the order is ambiguous, whether an extension applies, or whether a stipulation changed a date. The workflow should route those to the responsible attorney with the question stated plainly, not resolve them by picking the most likely reading.

Confidentiality applies throughout. Court orders in sealed or sensitive matters should stay within the firm's approved systems, and the extraction step should not send documents to tools the firm has not reviewed for confidentiality.

During a pilot, I would run the workflow alongside the existing process rather than in place of it. Compare its draft entries with what the docketing team calendared independently. Track missed entries, wrong trigger events, wrong periods, and dates that differed for a reason the reviewer had to explain.

The acceptance condition should be specific: for the pilot set of orders, every deadline the docketing team identified also appears as a draft entry with a correct source citation, and no entry reaches the calendar without a reviewer's confirmation. Speed is a secondary measure. A faster process that misses a deadline is not an improvement.

A sensible first release covers one court or matter type, one document type such as scheduling orders, and draft entries only. Notifications, client reporting and automatic reminders can follow once the extraction and review loop has proven reliable on real documents.

Bring a few redacted sample orders, your current docketing steps and the court rules you calendar against.