Records Classifier

Records Classifier reads your records and places each one in a Washington data category — Public, Sensitive, Confidential, or Confidential requiring special handling — with the passages and RCW citations behind the decision. It stops reading a record as soon as its category is settled, suggests a records retention series if you want one, and can work through a whole storage bucket and tag every file with its result.

What it's for

Knowing how sensitive a set of records is, quickly: before you move a file share to new storage, share data with a contractor, hand records to an AI tool, or decide what needs encryption or a data-sharing agreement. You drop in a few documents or a zip, or point it at a bucket, and get every record sorted by category with the evidence a person can check.

Navigating here

Records Classifier is an app Autessa turns on for your company; a company admin then installs it from the Marketplace (see Apps Enabled by Autessa). If your company already uses FirstPass, Records Classifier appears inside FirstPass as a Classify section and shares its rules and settings.

• Classify → Batches lists every set of records you have classified.
• Classify → Bucket connections lists the storage buckets it works through.
• Settings → Records Classifier holds the category table, when to stop reading, retention schedules, and API access.

Key concepts

• Record. One document, or an email together with its attachments. A record gets one category: the highest category of anything in it.
• Batch. A set of records classified together, from an upload or from a bucket connection.
• Data category. The Washington categories from WaTech's Data Classification Standard (SEC-08-01-S):

Category	Name	What it means
1	Public	Can be or has been released to the public.
2	Sensitive	Not protected by law, but for official use only. Still releasable when someone asks for it.
3	Confidential	Specifically protected from disclosure by law, such as personal information (RCW 42.56.590), employee information (RCW 42.56.250), or network security information (RCW 42.56.420).
4	Confidential, special handling	Protected by law and subject to especially strict handling rules, where disclosure could cause serious harm — for example health information, criminal justice information, federal tax information, or the location of a protected victim.

• Finding. A passage that matched one of your rules, with the rule's citation and the page it was on. Each rule has a category, so a finding is what places a record in a category.
• Handling flags. Labels such as PII, PHI, CJIS or FTI that tell you which handling rules apply to a record.

A category is a handling label, not a public records decision. A Category 2 record can still be requested and released, and a Category 3 record may still be partly releasable. Use FirstPass to respond to a public records request.

How to classify records

1. Open Classify → Batches and choose Classify records.
2. Name the batch and drop in documents, folders or zip files. Emails (.eml and .msg), PDFs, Office files, images and audio are supported.
3. Choose Start. Records appear in the batch as they are classified.

Reading a batch

The batch page shows how many records fall in each category, then a table of every record: its category, the main reasons, how many pages were read, the suggested retention series, and who reviewed it. Every column sorts, and you can filter by category, handling flag, records still needing review, or records where reading stopped early. Export CSV downloads the table as you see it.

Reviewing a record

Open a record to see it page by page with its classification beside it:

• Why lists each finding with its citation. Click one to jump to its page.
• Pages read says how far the classifier read. "Stopped at page 2 of 40 — a Category 3 finding was enough" means the rest of the record could not lower its category.
• Retention shows the suggested series, its retention and disposition, and other likely series.

Choose Accept (or press A) to agree, or set a different category with keys 1–4. A different category needs a short reason. Press J and K to move between records. Every acceptance and change is recorded in the audit trail.

When it stops reading

Once a record has a finding at your stop level, nothing later in it can lower its category, so the rest is not read. Set this in Settings → Records Classifier → Classification:

• At the first Category 3 or 4 finding (default) — fastest; tells you whether a record is sensitive.
• At the first Category 4 finding — reads on past Category 3 findings to catch special-handling information.
• Read every page — a full read of every record, with every finding listed.

A record that was read only in part always says which pages were not read. If the finding that stopped reading did not hold up on a closer check, the record is flagged for a person to review rather than reported as lower.

Choosing categories for your agency

Agencies don't all classify the same information the same way: some treat Social Security numbers or card numbers as Category 4. Settings → Records Classifier → Classification lists every rule with its category and handling flags; change a category and it applies to the next record classified. You can also turn individual rules off, including Category 2 detection (internal, official-use-only material).

Retention suggestions

Records Classifier can suggest the retention series each record belongs to, chosen only from the Washington schedules you turn on.

1. Open Settings → Records Classifier → Retention and turn suggestions on.
2. Choose your agency type: State agency (State Government General Records Retention Schedule plus your agency's own schedule) or Local government (CORE plus the sector schedules that apply to you).
3. Turn on the schedules that apply. The current published versions are included.
4. To add your agency-specific schedule, choose Import schedule and upload the DOCX or PDF exactly as the Washington State Archives published it.

You can turn individual series off, edit a series, or add your own. A suggestion is always just a suggestion: it is shown with its confidence and alternatives, and nothing is ever deleted or moved because of it. Records under a litigation hold or an open public records request must not be destroyed whatever their retention says.

Working through a storage bucket

Records Classifier can read records straight from your own Amazon S3 bucket and write the results back: each file is tagged with its category, and a results file is written for every batch. Your files are never moved, renamed or deleted.

Connect a bucket

1. Open Classify → Bucket connections and choose Add connection.
2. Enter the bucket, the folder to read (the source folder), the folder results go in (the output folder), your AWS account ID, and a KMS key if the bucket is encrypted with one. Choose One time to classify what is there now, or Watch for new files to classify files as they arrive.
3. The page shows a trust policy and a permissions policy to create a role named autessa-records-classifier in your account, with commands for bash and PowerShell. The role only trusts Autessa when the request carries your company's external ID, and it can only:
   • list and read files in the source folder,
   • read and write the tags on those files,
   • write files in the output folder.
4. Paste the role's ARN and choose Save and test connection. Each check shows whether it passed: assuming the role, listing the folder, reading tags, and writing to the output folder.

What gets written back

For each file in the source folder:

• Tags are added beside your own tags (yours are never removed): autessa:category (1–4), autessa:sensitive (true or false), autessa:retention-dan (when a series was suggested), autessa:reviewed (AI or HUMAN), and autessa:result (where its result file is). S3 allows 10 tags per file; if a file has no room, its tags are skipped and the results file says so.
• A result file in the output folder at the same path plus .classification.json, with the category, findings and citations, pages read, and retention suggestion. A zip file's result lists every record inside it.
• A batch results file at runs/<batch>/results.jsonl in the output folder, one line per file, written when the batch finishes.

When a person changes a record's category in the app, its tags, result file and batch results file are updated too.

Using the API

Everything above is also available over the REST API at /clients/records-classifier, authenticated with a client API key. Settings → Records Classifier → API shows ready-to-run examples.

Request	What it does
GET /bucket-access	The external ID and the role policies for your bucket and folders.
GET /sources, POST /sources	List or add bucket connections.
PUT /sources/<id>, DELETE /sources/<id>	Change or stop a connection (stopping deletes nothing).
POST /sources/<id>/test	Run the connection checks.
POST /sources/<id>/run	Classify the folder now; returns the batch ID.
POST /sources/<id>/pause, /resume	Pause or resume a connection.
GET /batches, GET /batches/<id>	Batches with their category counts.
GET /batches/<id>/records	Records with category, findings, pages read and retention. Filter with category and needsReview.
GET /batches/<id>/results	A short-lived link to the batch results file.
GET /records/<id>	One record.
POST /records/<id>/review	Accept a record, or set category (with a reason) or retentionDan.

Errors come back with a plain message and a stable errorCode. A connection can also call a webhook once a batch's results file is written.

Related

• Apps Enabled by Autessa
• Plate Review — another app that reads from your own bucket the same way