Changelog

Every update to Primentra, 30 releases so far. Search it to find when something changed.

New

17
  • Lookalikes finds near-duplicates per entity on the columns you choose, even when they are typed differently. It is off until an administrator switches it on
  • A review list per entity shows each group of lookalikes with its score and the reason per column
  • Merge a group into one record and pick the value to keep for every column. References to the merged records move to the record you keep
  • Undo a merge with Unmerge. The old codes of merged records keep working
  • A new Merge right per role and entity decides who may merge and unmerge
  • Test these rules runs the rules on the current records and saves nothing
  • Start from a template for companies, people, addresses or products
  • Auto-merge merges the groups above a score you set. It is off until you switch it on, and on an entity that needs approval it only proposes the merge
  • When you type a new record in the grid, a hint shows the records it looks like
  • The REST API can merge and unmerge, up to 1,000 records in one call
  • AI assist gives a second opinion on a group and can draft a rule, with your own AI provider and key. It is off by default and only advises
  • AI assist takes your own instructions per entity, on top of the fixed prompts, for the second opinion and the rule suggestion. Templates for companies, people, addresses and products fill them in
  • The Retention tab has a card for the AI call log. Calls older than 90 days, or the days you set, are deleted. The calls of today always stay
  • The Log Viewer has an AI calls tab: every AI assist call, 50 at a time, with filters on purpose, outcome, entity and date. The AI assist tab links to it
  • The Model field of AI assist lists the models of your provider. You can still type a name, such as an Azure deployment
  • Named schedules: make a schedule once under Settings, Schedules, and pick it by name for staging and for Lookalikes. The page shows each schedule in words, its next run and who uses it. Primentra never deletes a schedule in use
  • Lookalikes runs a full check on a named schedule, with its own switch. The switch is off until you turn it on

Improved

15
  • Staging in Scheduled mode picks a named schedule. The upgrade gives each staging schedule a named schedule with the same values, and the next run stays the same moment
  • The Settings menu on the left has four groups: Model, Operations, Platform and Migration. My Approvals, My Submissions and Preview As moved to the gear menu, and the gear shows how many changes wait for your approval
  • AI assist shows a saved API key as its first 6 and last 4 characters, for example sk-ant••••••••••••x7Qa. A key under 40 characters shows only dots. The key itself never reaches the browser
  • The online demo has AI assist with made-up answers: a second opinion, a rule suggestion and the AI calls tab. No key and no record leaves your browser
  • The gear menu has three groups in the order you use them: your daily work (My Approvals, My Submissions), administration, and your account. A group with nothing for you is not shown
  • When you dismiss the Set up Primentra card, it stays hidden in every browser you sign in on. To bring it back, click Show the Set up Primentra card again on the help page Setting Up Primentra
  • The Lookalikes settings are clearer. In the run status, the pairs, groups and not-checked records open the review queue and the Not checked list, and Settings, Schedules opens the Schedules page. Minimum shows n/a for Exact and Normalized, and the range of each number shows on screen. A warning shows at once when no column is set under Compare records that share. The next scheduled run uses the date format of the rest of Primentra, and a note at Save settings says a change takes effect at the next run
  • The Order column left the Lookalikes columns table. It never changed a result. Each column keeps its stored order
  • The Lookalikes settings are easier to read. Template cards name the fields they could not find, and say a draft replaces the rules on the screen. The AI instructions moved below the four numbered steps and the template texts use the words of the screen. Which value wins explains with an example that an empty value never wins. Most records sharing a value and Bulk-run threshold have their own heading, Matching limits. The columns table has no sort arrows, because you edit it and do not search it
  • The explanation at the top of the Lookalikes settings, Hide or Show, is kept on your account and follows you to another browser. The info icon next to a numbered step opens the explanation at that step
  • The Lookalikes settings use the entity's own records. Under each column, a real pair shows the similarity the setting gives. The explanation of Compare has a playground for two texts. A histogram shows the scores of the last run, with the count per range as you move the thresholds. A line says about how many pairs a run would compare for your Compare records that share choice, and names the values a run skips with their record counts, the same values Test these rules names. The explanation shows a real pair as its worked example
  • Test these rules shows a Simulation: how many groups auto-merge would merge by itself with these rules, and why the others wait for a person. Nothing is merged
  • In the Lookalikes settings, Cancel asks Discard changes? before it puts back the saved settings. The two-text playground starts empty with an example in gray. On an Extra proof row Minimum is gray, under the note that it changes no score, and a Minimum that Primentra fills in lights up for a moment. Which value wins has no sort arrows, so the rows stay where you edit them. The warning for a rule with no Compare records that share column says why Save is refused, and a click on Save in that state brings the warning into view
  • The explanation on the Lookalikes tab has Learn more links to a new help page, How Lookalikes compares values. It shows step by step how a value is cleaned, how Jaro-Winkler and the edit distance work, how the score is made and which records meet. The link opens a new tab, so your changes on the tab stay. Under Schedule and run logs, the Schedule field now lines up with the other fields
  • The question before you leave a page with unsaved changes names where the changes are, for example the Lookalikes tab, and its buttons are Keep editing and Discard & leave

Fixed

6
  • An Excel import reads dates up to 31-12-9999. A date after October 2173, such as 01-01-3000, reached the server as a number, and the row was refused
  • An Excel export writes a column with only a time as a time. Excel showed 10:30 as 02.01.1900 10:30:00. Files from the old export still import
  • Auto-merge no longer merges two records that a Must be close column refuses when a third record links them, for example with that column empty. The group stays in the queue and says why, the merge screen warns you before you merge, and the REST merge answer names the column and the two records
  • When rows fail in a staging batch, Process Batch says how many failed and that Batch History shows why. POST /api/staging/:entityId/process says how many in a message, and which call gives the reason per row
  • After a reload, the Retention tab could show an empty box or an old number. It now shows the days in the database, and a box you leave blank shows its default after Save
  • A warning in errorlog.txt now keeps its reason, for example why the Lookalikes worker did not start. Before, the line came without it