Hey GB Team! I'm trying to setup the GB Edge Delivery with Vercel and I'm stuck on setting up the SDK Webhook. I've entered all of the values into the form but when I go to test it, I'm getting this error:

Failed to parse custom headers: "Vercel" not defined in [object Object] - 1:10

When I update the headers to manually input the Token for Authorization I get:

{"error":{"code":"forbidden","message":"Not authorized","invalidToken":true}}

When I run the exact same request via Postman, I don't get an invalidToken error

Hi how to create a new project

Hi! Is it possible to filter API requests? I mean, could I filter the

/experiments

endpoint request by

dateUpdated

Thanks!

Hi, We have

anonymous_id

(device_id) from web that can map to

customer_id

when users log in. Our existing join table handles 1:1 ID mappings across systems perfectly (

customer_id ↔ other_id_1 ↔ other_id_2)

. The problem: We now need to handle m:1 relationships (multiple

anonymous_id

→ one

customer_id

). I created a separate join table for

anonymous_id ↔ customer_id

, but metrics that depend on other IDs (e.g.,

customer_id ↔ other_id_1 ↔ other_id_2

) don't work with this approach. Question: Should we create one consolidated table with all ID combinations (

anonymous_id ↔ customer_id ↔ other_id_1 ↔ other_id_2)

? Will GrowthBook handle the necessary grouping/distincting across all these IDs correctly?

Hi guys, One doubt about the phases in an experiment. We are running an A/B test on Android, but we realised the experiment setup was incorrect and the balance was wrong. After fixing the configuration, we started a new phase with re-randomized data, and we set-up the correct date for checking the data result after the new phase. But we are still seeing the same balance issue in the experiment. PS: I have checked the code and the tracking, and it's working fine. Do you have any clue what could happen with the experiment and the balance?

Hi everyone 👋 I’m wondering if there’s a way to achieve a certain setup using only two experiments instead of three. We currently have one experiment that includes all new users, and another one that should include only 30% of new English-speaking users. Ideally, this could be handled with just these two experiments, but in GrowthBook we were only able to implement it using three experiments — two of which duplicate the same logic (one for the English locale and one for all other locales). This works technically, but it prevents us from viewing the combined results of these two “half-experiments” as a single result. Has anyone managed to configure something similar with only two experiments?

Hi team, I am new to GrowthBook. I am integrating my application with GrowthBook. My local Growth Book instance is set up. I have my own tracking of events (we don't use Amplitude or other standard CDP platforms). How do I initialize Growth Book so that I can embed all experiment lists in my tracking?

Hi in our backend, changing a value of the FF dosent reflect unless we are doing a new rebuild. streaming is on. cache is set to only 30 seconds.

Hi team, after migrating our data source from Trino to Snowflake, the automatic updates (Cron schedule / periodic refreshes) stopped working. Could you help us look into this?

Hello, is there any way using API to know if feature is marked as stale or not? We want to automate stale feature deletion/reporting. I`ve inspected API docs as well as browser network responses while opening stale feature, but could not find anything

Hey team, how have you been? I hope you're all doing well. I got involved in a project in which our client wants to run an experimentation program using GrowthBook. I have not much experience running the tool but I'm eager to learn more. After some back and forth, I have a few questions, that hopefully are not too dumb. 1. Write Permissions: GrowthBook doesn't write anything to the client's warehouse, right? I'm asking because it seems they created an empty table to be used as a "FACT TABLE" inside their Databricks environment. 2. Fact Table Definition: My understanding is that the Fact Table should actually be a SELECT query pointing to the client's existing raw events data (e.g., from Segment or Snowplow), not a new table created specifically for GrowthBook. Is that correct? 3. Databricks Permissions: Our current Access Token connection is "Connected," but the SQL Explorer shows "No tables found." Does this confirm that the token is missing

SHOW

and `USAGE`/`SELECT` permissions on the necessary Catalogs/Schemas, even if the basic connection works? Any guidance on these points would be hugely appreciated as we finalize the Databricks setup! Thanks a lot!

Hey all, just curious, but is there any way to remove or reduce the flickering when split testing a new page?

Hey everyone, I have two questions regarding the iOS SDK: 1. Does GrowthBook automatically cache feature payloads on disk? If yes, how long does the cache last by default, and is it possible to disable it or change the duration? 2. When using the SDK only for feature flags, does GrowthBook collect any user data (e.g., IP address or device identifiers) that I should declare in App Store Connect? Thanks!

Hi everyone 👋 I have a question regarding image uploads in experiment variations. We use a custom S3-compatible storage, which is causing upload failures on our instance. While investigating, I found a similar issue opened in GitHub about GrowthBook only supporting official AWS S3 endpoints. Just to confirm my understanding: • Does GrowthBook currently support only AWS S3, and not custom S3-compatible endpoints? • Are you still using AWS SDK v2, or has SDK v3 support already been introduced? • If SDK v3 has been added, does that enable specifying a custom S3 endpoint for uploads? • And finally, is support for custom S3 endpoints planned on the roadmap? Thanks in advance, just trying to understand whether this is currently supported natively or not.

Hi all 👋 I'm trying to host GrowthBook on Azure using Container Apps, but I keep getting the error: `Error connecting to the GrowthBook API at

http://{custom_domain}:3100

. Has anyone successfully hosted the project on Azure and has some pointers? 😅 I feel like Ingress might be the culprit, not allowing me to open up more than 1 http-port. (It only supports multiple TCP ports, afaik)

I'm trying to work out the best way of structuring my fact tables to work with the new product analytics dashboards. I generally tend to be interested in proportion metrics - e.g. conversion rates. Up until now I've had fact tables where one row represents each conversion event. When used in an experiment the metric value is then expressed as a % of the experiment audience, which is great. Is there a way I can reuse this metric to create a conversion rate in a product analytics dashboard? I can't work out how to do it as there doesn't seem to be a way to express what the total audience is. For example, I don't think I can specify the metric to show as a proportion of all site visitors - or other fact table values. So is there a way I can monitor an overall site conversion, without changing my existing fact table?

Background: I'm experiencing an issue with the Flags Explorer setup in Next.js where the

getProviderData()

function from

@flags-sdk/growthbook

doesn't seem to be mapping flag options from experiment variations tied to that flag in growthbook. There's a discrepancy between two different documentation approaches for next.js + vercel flags explorer: 1. Next.js and Vercel Feature Flags Tutorial - manual implementation of

getFlagApiData()

that explicitly loops through the GrowthBook payload to extract options from rules, variations, and force values. (older) 2. Next.js SDK (Vercel Flags) Documentation - Uses the built-in

getProviderData()

function from

@flags-sdk/growthbook

package, which should handle this automatically. (new) Issue: When using the SDK's

getProviderData()

function as shown in the newer docs, the flags appear in the toolbar with the default as their only option. Question: Is this a known limitation of

getProviderData()

, or did I miss something in the docs?

Hello growthbook👋 we want to start testing a new feature. We want to proceed very cautiously and initially only include 10% of the user base (with a 50:50 split). If we see after 1-2 days that there are no problems, we want to roll it out to 30% of the user base, and so on. What is the best approach here? For example, can we use a namespace and adjust it during the experiment? Looking forward to your feedback. Thank you in advance 🙏

Quick question; Is the Growthbook Proxy a requirement for streaming to work? Without a proxy, I can't seem to get it working. (I've not tested with a proxy)

Hey Growthbook Team, We were getting a lot of initialization failure for our growthbook setup in flutter, with the error

failed to lookup host

, and wanted to implement a retry mechanism when initialization fails. GBSDKBuilder provides a failure callback but seems like there is no init retry mechanism build around failure handling, our init logic is written in a singleton pattern so creation new object if init fails doesn't seem viable either, so wanted to understand how can we tackle this problem are there any helpful inbuild methods whcih gb provides for init retry?

We’ve been using growthbook for a few minutes with the client side SDK, but we’re looking to swap over to lambda edge. I’ve been trying to get it working but not having much luck. I’ve confirmed a basic lambda function is configured w/ our cloudfront, but once I add the growthbook specific handler, I run into errors. When I deploy the lambda version that uses growthbook, I’m seeing logs with this error:

2025-11-18T15:22:00.113Z	a762f8a1-1df6-4c0b-98aa-686ec37e6ff8	ERROR	Unhandled Promise Rejection 	
{
    "errorType": "Runtime.UnhandledPromiseRejection",
    "errorMessage": "TypeError: Invalid URL",
    "reason": {
        "errorType": "TypeError",
        "errorMessage": "Invalid URL",
        "code": "ERR_INVALID_URL",
        "input": "undefined",
        "stack": [
            "TypeError: Invalid URL",
            "    at new URL (node:internal/url:825:25)",
            "    at Ms (/var/task/index.js:31:69858)",
            "    at Object.w3 [as proxyRequest] (/var/task/index.js:31:75389)",
            "    at /var/task/index.js:31:65455",
            "    at Generator.next (<anonymous>)",
            "    at s (/var/task/index.js:31:64093)"
        ]
    },
    "promise": {},
    "stack": [
        "Runtime.UnhandledPromiseRejection: TypeError: Invalid URL",
        "    at process.<anonymous> (file:///var/runtime/index.mjs:1448:17)",
        "    at process.emit (node:events:519:28)",
        "    at emitUnhandledRejection (node:internal/process/promises:252:13)",
        "    at throwUnhandledRejectionsMode (node:internal/process/promises:388:19)",
        "    at processPromiseRejections (node:internal/process/promises:475:17)",
        "    at process.processTicksAndRejections (node:internal/process/task_queues:106:32)"
    ]
}

Nothing is growthbook specific in those logs, but I’m guessing it’s coming from something growthbook is doing. A couple of questions: 1. PROXY_TARGET — I’m pointing this to our S3 bucket that serves the website, which is

http

, not

https

— is that okay? If I use https, that’d point back to a cloudfront distro URL and I think it’d go into some kind of infinite redirect loop 2. Is there a way to test this w/o having to redeploy my cloudfront distro and wait for the website to break

I'm thinking about query cost/performance optimization with the pre-compute dimensions setting. What are the tradeoffs I'm making by selecting pre-compute??

Hi, I’m wondering why I’m getting this “Invalid experiment” error? I’ve set up a 50/50 experiment and am trying to associate it

Hey there, I am looking to implement a user retention metric after viewing an experiment. I'm using:

Metric: Retention

Fact table: Events, Row filter: event_name = 'page_view'

However the lowest time I am able to set is: Event must be at least

1 Minutes

after experiment exposure Which effectively is

timestamp >= (exposure_timestamp + '1 minutes')

However, I wish I could use:

timestamp > (exposure_timestamp)

Note: not equal Is this possible with a fact metric?

I have a question regarding the growthbook setup in nuxt 2. To make it work, I had to fetch features through url, as the setup in documentation did not work. Is it expected for nuxt2? Do I still get the cashing features of growthbook even if I dont follow the documentation for vue?

Hi. How fact table optimization feature works? I want to know details, and impact on user side.

----------------- Hi, I'm using the GrowthBook experiment platform with a fact table–based data source (BigQuery). When I define N metrics on top of the fact table, GrowthBook is able to compute all metrics with a single underlying query, which is very efficient. However, when I view Experiment → Results and apply a segment, GrowthBook stops using the single fact-table query and instead executes *N separate queries*—one for each metric. This makes the query execution significantly less efficient and increases BigQuery cost. Is this expected behavior? And is there any way to preserve the single-query optimization when segmentation is applied?

Hi all! We have a mobile app (Android and iOS). Share your experience running an A/B test on subscription pricing. What challenges did you encounter?

Hello, we are currently running on version 3.3.0 and would like to upgrade to the newest 4.2.0, is there any issues with doing that or is there any special considerations we should be aware of?

Hi, What is the difference between these two metric definitions? 1. Ratio metric with

Numerator = SUM(profit per order)

,

Denominator = Unique Users

2. Mean metric with

per user aggregation = SUM(profit per order)

. When defining the Mean metric (2), the Growthbook interface says:

The final metric value will be the average per-user value for all users in the experiment. Any user without a matching row will have a value of 0 and will still contribute to this average.

So I don't understand the difference between a denominator of "Unique Users" in 1 and "average per-user value" in 2. I would expect these two metrics to be identical. However, on the experiment overview page I can see they have the same numerator value (1.7m), but the Ratio (1) has 5k Unique Users for the Denominator, while Mean (2) has a denominator of 83k. We're doing an A/A test (no difference between the two treatments) and are seeing Growthbook report a statistically significant Relative uplift for the Ratio Metric, but not for the Mean metric. I would like to understand the details of what is happening behind the scenes a bit better.