@@ -65,6 +65,17 @@ screenshot. Write it to a caller-supplied path (`<SUITE>_REPORT_PATH`) and have
 disposable, seeds with SQL, exercises the real HTTP boundary, cleans up its fixtures, and writes
 the report.
 
+## Flaky tests
+
+A daily job (`.github/workflows/ci-flakes.yml`) files a `ci-flake` issue for every test or job
+that failed on a CI attempt and passed when the same commit was re-run, and reopens it if the
+flake recurs after it was closed. Fix the nondeterminism it names: pin the clock
+(`vi.useFakeTimers({ toFake: ['Date'] })` + `vi.setSystemTime`), give each test its own ids and
+rows, wait on the condition rather than a delay, and keep cold imports out of a test's timeout by
+importing statically. Never add a retry, a skip, or a longer timeout to make one pass; raise a
+budget only when its run history shows the budget itself is wrong, and record that evidence
+beside it.
+
 ## Unit test mechanics
 
 ### Structure
 
@@ -63,6 +63,17 @@ screenshot. Write it to a caller-supplied path (`<SUITE>_REPORT_PATH`) and have
 disposable, seeds with SQL, exercises the real HTTP boundary, cleans up its fixtures, and writes
 the report.
 
+## Flaky tests
+
+A daily job (`.github/workflows/ci-flakes.yml`) files a `ci-flake` issue for every test or job
+that failed on a CI attempt and passed when the same commit was re-run, and reopens it if the
+flake recurs after it was closed. Fix the nondeterminism it names: pin the clock
+(`vi.useFakeTimers({ toFake: ['Date'] })` + `vi.setSystemTime`), give each test its own ids and
+rows, wait on the condition rather than a delay, and keep cold imports out of a test's timeout by
+importing statically. Never add a retry, a skip, or a longer timeout to make one pass; raise a
+budget only when its run history shows the budget itself is wrong, and record that evidence
+beside it.
+
 ## Unit test mechanics
 
 ### Structure
 
@@ -8,9 +8,10 @@
 # The job provides DATABASE_URL, BETTER_AUTH_SECRET and ENCRYPTION_KEY; each group sets the rest of
 # its app's environment here. Reports and server logs land in $RUNNER_TEMP/e2e.
 #
-# Under `next dev` the first request cold-compiles the app under Turbopack, which takes 42-150s on
-# CI runners, so the readiness deadline only has to catch a hung boot: an exited server fails
-# immediately, and either way the server log tail lands in the job log.
+# Under `next dev` the first request cold-compiles the app under Turbopack, which takes 60-185s on
+# a healthy CI runner and has run past 300s on a degraded one. The readiness deadline, about three
+# times the healthy worst case, only has to catch a hung boot: an exited server fails immediately,
+# and either way the server log tail lands in the job log.
 #
 # Each app starts from an empty Turbopack dev cache: a cache written under other NEXT_PUBLIC_*
 # values, by a server that `next dev` SIGKILLs 100ms after SIGTERM, can panic Turbopack or wedge a
@@ -21,7 +22,7 @@ set -euo pipefail
 
 group=${1:?usage: http-e2e.sh <scim|cli|stop-after|desktop-inbox|project-files|mobile>}
 report_dir="$RUNNER_TEMP/e2e"
-ready_timeout_seconds=300
+ready_timeout_seconds=600
 mkdir -p "$report_dir"
 
 server_pid=''
 
@@ -0,0 +1,55 @@
+name: ci-flakes
+
+# Files one `ci-flake` issue per CI check that failed and then passed on a re-run of the same
+# commit, and reopens a closed one when its flake recurs (scripts/report-ci-flakes.ts). The
+# three-day lookback overlaps the daily schedule on purpose: the script skips occurrences an issue
+# already lists, so a missed or repeated run neither drops nor double-counts one.
+
+on:
+  schedule:
+    - cron: '23 7 * * *'
+  workflow_dispatch:
+    inputs:
+      days:
+        description: Look back this many days of CI runs (the script's default is 3).
+        required: false
+      dry_run:
+        description: Print the issues it would file or update without changing them.
+        type: boolean
+        required: false
+        default: false
+
+permissions:
+  actions: read
+  contents: read
+  issues: write
+
+concurrency:
+  group: ci-flakes
+  cancel-in-progress: false
+
+jobs:
+  report:
+    name: report
+    # GitHub-hosted: a pure API client with no dependencies, so it needs no install or cache.
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v6
+        with:
+          persist-credentials: false
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2
+        with:
+          bun-version: 1.4.2
+      - name: Report flaky checks
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
+          DAYS: ${{ inputs.days }}
+          DRY_RUN: ${{ inputs.dry_run }}
+        run: |
+          args=()
+          if [ -n "$DAYS" ]; then args+=(--days "$DAYS"); fi
+          if [ "$DRY_RUN" = true ]; then args+=(--dry-run); fi
+          bun scripts/report-ci-flakes.ts "${args[@]}"
@@ -37,7 +37,7 @@ Sim is the open-source AI workspace where teams build, deploy, and manage AI age
 
 ## Calling the Sim API
 
-- Base URL: \`${SIM_SITE_URL}\`. Every endpoint is under \`/api\`; there is no \`api.sim.ai\` host.
+- Sim Cloud base URL: \`${SIM_SITE_URL}\`. For self-hosted deployments, use your deployment's origin instead, including in the example below. Every endpoint is under \`/api\`; there is no \`api.sim.ai\` host.
 - Run a deployed workflow: \`POST ${SIM_SITE_URL}/api/v2/workflows/{workflowId}/execute\` with a JSON body like \`{ "input": { "fieldName": "value" } }\`.
 - Authenticate API keys with the \`X-API-Key\` header. \`Authorization: Bearer\` accepts only Sim OAuth access tokens, never API keys.
 - Responses wrap the payload in \`data\`. A synchronous run returns \`200 { "data": { "runId", "status", "output", ... } }\`; with \`"async": true\` in the body it returns \`202 { "data": { "runId", "statusUrl" } }\` and you poll \`statusUrl\`.
 
@@ -16,7 +16,7 @@ export function Tabs({ className, ...props }: TabsProps) {
     <FumadocsTabs
       {...props}
       className={cn(
-        'min-w-0 scroll-mt-[calc(var(--fd-docs-row-3,0px)+1rem)] border-[var(--border)] [&>[role=tablist]>button]:font-normal! max-sm:[&>[role=tablist]>button]:min-h-11 [&>[role=tablist]]:min-w-0 [&>[role=tablist]]:[scrollbar-width:thin]! max-sm:[&>[role=tablist]]:min-h-14',
+        'min-w-0 scroll-mt-[calc(var(--fd-docs-row-3,0px)+1rem)] border-[var(--border)] [&>[role=tablist]>button]:font-normal! max-sm:[&>[role=tablist]>button]:min-h-11 [&>[role=tablist]]:min-w-0 [&>[role=tablist]]:[scrollbar-width:thin] max-sm:[&>[role=tablist]]:min-h-14',
         className
       )}
     />
 
@@ -109,6 +109,7 @@ const SELECT_TITLES = new Set([
   'Operation',
   'Method',
   'Unit',
+  'Event',
   'Event type',
   'Validation',
   'Account',
 
@@ -198,7 +198,7 @@ Apply one precise edit to an existing text file without rewriting it. Use search
 | `anchor` | string | No | For insert_after, the complete line after which content is inserted. |
 | `startAnchor` | string | No | For delete_between, the complete first line to delete. The start anchor is removed. |
 | `endAnchor` | string | No | For delete_between, the complete ending boundary line. The end anchor remains in the file. |
-| `occurrence` | number | No | For anchored edits, which matching anchor occurrence to use, starting at 1. Defaults to 1. |
+| `occurrence` | number | No | For anchored edits, which start or insert anchor match to use, starting at 1. Defaults to 1. The end anchor is the first match after it. |
 | `expectedRevision` | string | No | Refuse the edit unless the file still holds the content this revision names, as returned by Get File or an earlier write. Use it so an edit computed from what you read cannot overwrite someone else’s change. |
 
 #### Output
 
@@ -200,7 +200,7 @@ Images smaller than **2 KiB**, SVG, HEIC, and AVIF files are not indexed through
 
 New knowledge-base Confluence connections enable image reading by default. Connections created before image support keep it off until you choose **Read with AI**; their next sync then indexes images.
 
-The default admission limit is **500 new or changed images per sync**. Remaining images wait for another sync. Self-hosted operators can configure `KB_IMAGE_EXTRACTIONS_PER_SYNC`. Unchanged images reuse their extraction, including during a full resync; transient or malformed responses can cause additional model requests.
+The default admission limit is **500 new or changed images per sync**. Remaining images wait for another sync. With **Member accounts**, each run that continues a sync can read up to 500. Self-hosted operators can configure `KB_IMAGE_EXTRACTIONS_PER_SYNC`. Unchanged images reuse their extraction, including during a full resync; transient or malformed responses can cause additional model requests.
 
 Image extraction follows [model billing](/platform/costs): BYOK usage is billed by the provider, while platform-key usage follows the deployment's Sim billing policy. Choose **Don't index** to stop image indexing; the next sync removes indexed images. Switching back processes them again and incurs the applicable model usage.
 
 
@@ -124,7 +124,7 @@ For Ollama on the Docker host, use `OLLAMA_URL=http://host.docker.internal:11434
 | Variable | Description |
 |----------|-------------|
 | `KB_IMAGE_EXTRACTION_MODEL` | Gemini model that reads Confluence image attachments into searchable text. Defaults to `gemini-3.8-flash`; a value that is not a Gemini 3 or later model in the catalog falls back to the default. Needs `GEMINI_API_KEY` (or a workspace Gemini key); without one, image attachments are left out of syncs |
-| `KB_IMAGE_EXTRACTIONS_PER_SYNC` | New or changed images one connector sync sends for reading, at least `1`. Defaults to `500`; the rest wait for the next sync, which reports as partial |
+| `KB_IMAGE_EXTRACTIONS_PER_SYNC` | New or changed images one connector sync run sends for reading, at least `1`. Defaults to `500`; the rest wait for the next sync, which reports as partial. With member accounts, each continuation run can read up to this many |
 | `KB_CONFIG_VISION_REQUESTS_PER_MINUTE` | Image reads per minute per Gemini key, shared by every sync on the deployment. Defaults to `300`; lower it to match a key with a smaller Gemini quota |
 | `KB_EMBEDDING_MODEL` | Embedding model for new knowledge bases. Defaults to `text-embedding-3-small`; use `ollama/<model>` for a model on your own Ollama. An unrecognised hosted model id falls back to the default, but an `ollama/` id is taken at face value — if that model is not on the server, knowledge-base creation fails rather than falling back |
 | `EMBEDDING_OUTPUT_DIMS` | Stored vector width: `384`, `768`, `1024`, `1536`, or `3072`. Hosted models default to `1536` when supported, otherwise their largest storable width. Invalid hosted widths fall back with a warning; Ollama can infer its actual width when unset |