Your agent can call ax1om · scoring is live over MCP, free tier included.Start freeAlready have an account?

Get a demoStart free
Changelog

Changelog

What shipped, and what it changes.

Subscribe by RSS

Bug fixModelsSalesforceHubSpot

Criteria values match literally

A criteria value used in a contains filter was handed to the matcher as a regular expression. For most values that made no difference. For a value with a ., +, ( or | in it, an email domain or a product SKU say, the pattern matched rows it should not have, or none, and a stray open bracket could fail the run. It was also an open door for an expensive pattern.

Every criteria match site now matches the literal text, 17 of them across datasets, training, rescoring and the CRM assemblies, and a guard test fails the build if a new site forgets.

If a model’s criteria contain one of those characters, its next training run or scheduled rescore may select a slightly different set of rows: the set the criteria always meant.

Permalink

ImprovementDashboardModels

CSV imports are de-identified before anything is stored, or refused

An upload used to be read twice: once at admission with Python’s csv module, and once more with pandas for de-identification. When pandas refused a file the first reader had accepted, the de-identifier fell back to column names alone, and a column of email addresses under an innocuous header could land in storage as it arrived.

There is one parser now. It reads the file once, every cell kept as text and empty cells preserved. The column analysis and the de-identification both work from that same frame, and the identifiers are stripped before the import row is written. A file the parser refuses is a 400 that names the row and carries no cell value, and nothing is stored, on the direct upload and on upload by reference alike.

Two smaller changes ride along. The per-column metadata used to keep up to 50 sample values per column; for any column the detector flags as an identifier, the samples are now empty and the column carries its category instead. And duplicate header names are suffixed the way pandas does (Email, Email.1) rather than kept verbatim. Metadata already stored for earlier imports was rewritten the same way the night this shipped.

PermalinkLearn more

ImprovementAPIBillingDashboard

Scoring keys and billing are owner and admin actions

Before this release, any member of an org could mint a scoring key, see every key’s prefix and caps, revoke one, open checkout, and change the payment method in the customer portal. The role check existed on one verb, renaming a key, and nowhere else.

Every one of those actions now requires the human owner or admin role. The API refuses with 403 and the code admin_required. Agent credentials are refused on the key verbs too, whatever the role of the person who granted them: a credential is a control surface, and a key that can mint keys would outlive its own revocation.

Members see one line in the console where the key list used to be: API keys are managed by your org’s admins. A member who needs to score over the API gets a key from an admin, and scoring with that key is unchanged.

PermalinkLearn more

2026