> For the complete documentation index, see [llms.txt](https://doc.wearepatchworks.com/product-documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.wearepatchworks.com/product-documentation/release-information/core-release-notes/04-11-26-release-notes-core.md).

# 04/11/26 Release Notes (core)

> **Release date:** 4th November 2026

***

## New Features

### **New Shape: Cross Reference**

Process flows can now read and write cross-reference lookups directly, using a new [Cross Reference](https://doc.wearepatchworks.com/product-documentation/process-flows/building-process-flows/process-flow-shapes/standard-shapes/cross-reference-shape.md) shape.

Previously, cross-reference lookups could only be used within mapping transformations, and keeping them up to date meant editing values by hand or building cache-based workarounds. The new shape lets a flow maintain a lookup as data passes through - for example, after creating a product in NetSuite, write the returned internal ID against the SKU, ready for any other flow to use. Records in a cross-reference lookup do not expire and are available to every flow.

* **Upsert** - match on a key field and insert the record if it's absent, or update it if it's present. The payload passes through unchanged, and the run log summarises what happened (e.g. `4 inserted, 0 updated, 0 unchanged, 3 skipped`).
* **Lookup** - match on a key field and write the matched value to a target field in the payload. Choose the lookup direction (left to right or right to left), a match type (exact, starts with, contains) and an output type (string or integer). Keys with no match use the lookup's default value, if one is set.
* **JSON & XML support** - target fields with wildcard paths for arrays (`products.*.sku`) in JSON, or XPath (including namespaces and attribute targets) in XML. Flow variables can be used in paths.
* **Safe to manage** - a lookup can't be deleted while a flow references it, and renaming a lookup won't break the flows using it. The shape is fully supported in flow duplication, export/import and blueprints, with lookups included as blueprint dependencies.

***

### **New Shape: Data Transformation**

A new [Data Transformation](https://doc.wearepatchworks.com/product-documentation/process-flows/building-process-flows/process-flow-shapes/standard-shapes/data-transform-shape.md) shape handles common payload transformations - converting between formats, or running a find/replace - without writing or deploying a custom script.

Pick from a library of predefined transformations, maintained centrally by Patchworks. Each transformation shows what it accepts and produces, and you can test it against a sample payload before saving. To run several transformations in sequence, add a shape for each - they run in order, with the output of one becoming the input of the next.

Transformations are versioned: when a newer version is released, flows already using the previous version keep running unchanged - older versions simply emit a warning in the run log and can no longer be selected for new use. Bespoke transformations can also be added privately for a specific company - contact support if there's a script you reuse often enough that it should be on the list.

***

### **Flow Health Check**

You can now run a [health check](https://doc.wearepatchworks.com/product-documentation/process-flows/building-process-flows/flow-health-check.md) on any flow version, straight from the flow canvas.

Previously, problems such as a step pointing at a deleted script, a map transform with nothing selected, or a route whose first step was removed only came to light when the flow failed at runtime, or when a blueprint publish was attempted. A new health check button on the flow canvas checks the version you're currently viewing - draft or deployed - on demand.

* **Clear results** - each problem is reported with a message, its location and a suggested fix, grouped and filterable by severity, with a *go to step* option that takes you straight to the affected shape.
* **Informational only** - the check never changes your flow and doesn't block deployment; a version with no problems simply reports that the check passed.
* **Re-run any time** - fix a problem and run the check again to confirm the current state of the version.

***

### **Review Patchworks - In-App Reviews**

You can now rate Patchworks from inside the dashboard. A new [**Review Patchworks**](https://doc.wearepatchworks.com/product-documentation/getting-started/reviewing-patchworks.md) option in the sidebar (and an occasional prompt at milestones such as your first flow going live, with *remind me later* and *don't ask again* options).

***

## Improvements

### **Mapping: handle null & empty values before transforms**

Map can now remove or skip null and empty values earlier in the mapping flow, so transforms like Date Format are less likely to fail when source data is missing or blank.

* **Remove if null / Remove if empty** (source fields) - drop a mapping row entirely before any transforms run. For example, a `cancelled_date` field that's sometimes null can be excluded from the output instead of breaking a date transform.
* **Skip if null / Skip if empty** (transforms) - bypass the transform and pass the value through unchanged.

Existing maps are unchanged until these options are turned on. The new behaviour applies to Map JSON mapping; XML and legacy mapping are not affected. One thing to note: skipping a transform in the middle of a transform chain may behave unexpectedly if a later transform depends on its output. Labels for the new toggles are English-only for now, with other languages to follow.

***

### **AI Studio agent updates**

* **Upgraded models** - every agent now runs on the latest generation of models, each on a tier suited to its job. Output limits are larger, so long scripts and answers are no longer cut off, and flow documentation generation now reports a failure instead of loading forever.
* **Assistant suggestions** - the Assistant now steers you to the right agent for the job. Ask it to document or explain a flow and it offers the Documentation Agent; tell it you're migrating from another platform and it offers the Migration Agent - alongside the existing Implementation Agent suggestions.
* **Migration Agent map building** - when auditing an export from another platform, the Migration Agent now builds the field mappings it finds into Map shapes on the proposed flow, rather than leaving mapping as a manual follow-up.
* **Flow Builder builds Cross-Reference steps** - ask the flow builder to "save the product ID against the SKU" and it adds and configures a Cross Reference shape (Upsert or Lookup), reusing an existing lookup or creating one as needed.
* **Safer AI endpoint updates** - when AI-proposed connector endpoint updates touch an endpoint that's in use, you're warned which flows are affected and asked to confirm before anything is applied - both from the connector's endpoints tab and from the flow builder chat.
* **Health Monitor payload access** - when diagnosing a failed run, the Health Monitor can now retrieve the payloads saved against that run. It summarises what's there and asks before opening payload contents - so a large run with many payloads won't be dumped into the conversation. Payloads remain subject to the standard retention period.
* **Clearer message limits** - agent messages are capped at 5,000 characters, with a clear message suggesting you attach larger content as a file instead of the send silently failing.

***

### **Redesigned Scripts page**

The Scripts page has been rebuilt around cards, with **My scripts** and **Marketplace** tabs. Each script card shows its language, status, type, version count and when it was last updated, with quick actions to open the editor, edit with AI, manage versions, deploy or see where it's used. You can search, filter by status, type and language, and sort - and your filters are kept in the URL, so a filtered view can be shared or bookmarked. Marketplace templates can be previewed and installed from the same page.

***

### **Faster Flows page**

The Flows page now loads significantly faster for accounts with large numbers of flows. Flows appear progressively as they load, return visits show your list instantly from a cache while refreshing quietly in the background, and changes made elsewhere - a colleague renaming a flow, or the AI flow builder creating one - appear in your open list within a few seconds, without a reload.

***

### **Manual runs remember your last payload**

The *initialise flow with payload* dialog now opens with the payload you last ran for that flow, so you don't have to paste it in every time. Payloads are remembered per user, per company and per flow, kept across flow versions, and stored in your browser only - they're forgotten after 30 days, cleared on logout, and payloads over 100KB aren't remembered.

***

### **Run Logs: see who triggered a run**

The Run Logs page now shows who triggered each run - manual runs display the user's email under the trigger type - and a new **Triggered by me** filter shows just your own runs. The page also works better on narrower screens, and run detail waterfalls no longer stretch the page sideways.

***

### **Caches page shows all flow run caches**

The caches management page now lists every flow run cache, regardless of when it was last used. Previously only recently active caches were shown, making older ones impossible to see or delete. Note that a flow run cache's stored contents still expire on their usual TTL, so an older cache's detail page may show as empty.

***

### **Endpoint changes in Platform Logs**

Changes to endpoints are now properly captured in Platform Logs: creating, renaming, editing and deleting endpoint variables (with old and new values), header changes, schema updates and pagination settings. Secret variable values and authorisation headers are masked, and secrets are never written to the logs.

***

### **Flow Runs API: metrics and stable paging**

The flow runs list API can now include `operations_sent`, `operations_received`, `cpu_time`, `payload_size` and `metrics` per run via the `include` parameter, and paging is now stable - runs sharing the same start time no longer appear twice or go missing across pages.

***

### **Versions page: "used in" for non-deployed versions**

The *used in* indicator on the flow versions page now appears for non-deployed versions too, not just the deployed one.

***

## Bug Fixes

* **Try Catch retries no longer repeat successful requests** - when one payload in a batch failed inside a Try Catch, the retry re-sent every payload from the beginning, including those that had already succeeded - which could create duplicate records (e.g. duplicate orders) in the destination system. Retries now re-send only the failed payload; anything already sent successfully is sent once.
* **Home page success/failure filters** - clicking the successful or failed flow runs box on the home page now correctly filters the run logs list, instead of showing all runs.
* **Marketplace and connectors pagination** - moving to page 2 of the Blueprint marketplace, connector templates or connectors list no longer mislabels it as page 1 and strands you there; paging, search and filters now behave correctly on all three lists.
* **AI chat input height** - pasting a very long message into an AI conversation no longer grows the input beyond the window, hiding the send button.
* **IBM MQ connection test** - testing an IBM MQ connector no longer sends a blank queue manager and channel. A failed test no longer saves the invalid credentials, and a saved CA certificate now shows a "certificate saved" hint instead of appearing empty.
* **Flow import: map cache lookups** - importing a flow whose Map uses a cache lookup now wires the transform to the installed cache, instead of leaving it blank and the map broken until fixed by hand.
* **Script pagination: duplicated Content-Type** - pagination scripts that return the previous request's headers no longer cause a doubled `Content-Type` header on subsequent pages.
* **Database endpoints: "Open endpoint" link** - choosing *Open endpoint in new tab* on a connector step using a database connector now opens the correct query page instead of an invalid endpoint URL. Other connector types are unaffected.

***

*For questions about this release, please contact the Patchworks support team or refer to the* [*product documentation*](https://doc.wearepatchworks.com/product-documentation)*.*
