Repo sync

content/copilot/reference/copilot-billing/request-based-billing-legacy/model-multipliers-for-annual-plans.md

@@ -24,7 +24,7 @@ The models included with {% data variables.product.prodname_copilot_short %} pla
 
 Model multipliers and costs are subject to change.
 
-> [!NOTE] If you use {% data variables.copilot.copilot_auto_model_selection_short %} in {% data variables.copilot.copilot_chat_short %}, {% data variables.copilot.copilot_cli_short %}, or {% data variables.copilot.copilot_cloud_agent %}, you qualify for a 10% discount. For example, if a model has a multiplier of 1x you'll be billed at 0.9x instead.
+> [!NOTE] Users on legacy annual {% data variables.product.prodname_copilot_short %} plans will not receive access to new models and features. 
 
 ## Model multipliers
 
@@ -36,6 +36,8 @@ The following table shows the model multipliers per supported model.
 > * {% data variables.copilot.copilot_claude_sonnet_46 %}
 > * {% data variables.copilot.copilot_gpt_54_mini %}
 > * The multiplier for {% data variables.copilot.copilot_mai_code_1_flash %} is a promotional rate.
+> 
+> If you use {% data variables.copilot.copilot_auto_model_selection_short %} in {% data variables.copilot.copilot_chat_short %}, {% data variables.copilot.copilot_cli_short %}, or {% data variables.copilot.copilot_cloud_agent %}, you qualify for a 10% discount. For example, if a model has a multiplier of 1x you'll be billed at 0.9x instead.
 
 | Model | Multiplier |
 | --- | ---: |

data/reusables/dependabot/dependabot-updates-supported-versioning-tags.md

@@ -19,11 +19,20 @@ The `dependabot.yml` file doesn't control the versioning tags that you can use,
 | pipenv              | `pip`          | `a`, `b`, `rc`, `dev`, `post` | `requests@1.0.0a1`, `numpy@2.0.0b3`, `django@4.0rc1`, `black@1.0.0.dev5`, `pandas@2.0.5.post1` |
 | pip-compile         | `pip`          | `a`, `b`, `rc`, `dev`, `post` | `requests@1.0.0a1`, `numpy@2.0.0b3`, `django@4.0rc1`, `black@1.0.0.dev5`, `pandas@2.0.5.post1` |
 | poetry              | `pip`          | `a`, `b`, `rc`, `dev`, `post` | `requests@1.0.0a1`, `numpy@2.0.0b3`, `django@4.0rc1`, `black@1.0.0.dev5`, `pandas@2.0.5.post1` |
+| Gradle              | `gradle`       | `alpha`, `a`, `beta`, `b`, `milestone`, `m`, `rc`, `cr`, `snapshot`, `ga`, `final`, `release`, `sp` (case-insensitive) | `spring-boot-starter@3.0.0-RC1`, `kotlin-stdlib@2.0.0-beta`, `guava@33.0.0-SNAPSHOT`, `junit@5.10.0-M2`, `ktor@2.3.0-rc.1` |
+| Elm                 | `elm`          | None—strict `MAJOR.MINOR.PATCH` only (no pre-release versions) | `elm/core@1.0.0`, `elm/html@2.3.1`, `elm/json@10.0.0` |
+| Docker              | `docker`       | `alpha`, `beta`, `rc`, `dev`, `preview`, `pre`, `nightly`, `snapshot`, `canary`, `unstable` (heuristic detection) | `nginx@1.25.0-rc1`, `node@20.0.0-alpha.1`, `redis@7.0.0-nightly`, `alpine@3.18.0-dev`, `ubuntu@22.04-preview` |
+| git submodule       | `gitsubmodule` | None—pins to commit SHAs or git tags (no versioning scheme) | `my-lib@abc1234`, `shared-utils@v1.2.0` |
+| Go modules          | `gomod`        | `alpha`, `beta`, `rc` (SemVer prerelease after `-`) | `github.com/go-chi/chi@v5.0.0-rc1`, `google.golang.org/grpc@v1.60.0-beta.1`, `github.com/octo-org/octo-module@v0.17.0-alpha.1` |
 
 #### Ecosystem-specific versioning details
 
 The following details describe how {% data variables.product.prodname_dependabot %} interprets versioning for specific ecosystems.
 
 * **Bundler:** Bundler does not use a fixed set of prerelease tags. Any version segment containing a letter is treated as a prerelease (for example, `.alpha`, `.beta1`, `.rc2`). Hyphens in version strings are normalized to `.pre.` internally (for example, `1.0.0-beta` becomes `1.0.0.pre.beta`).
 * **Cargo:** Follows SemVer 2.0.0 prerelease conventions. Anything after `-` is a prerelease identifier (dot-separated, `[0-9A-Za-z-]`). Build metadata (`+...`) is allowed but ignored for version precedence.
+* **Gradle:** Aliases are also recognized: `pr`/`pre`/`preview`→`rc`, `eap`/`ea`→`alpha`. Additional prerelease qualifiers include `dev`, `experimental`, and `unstable`. Qualifiers are ordered by precedence: `alpha`/`a` < `beta`/`b` < `milestone`/`m` < `rc`/`cr` < `snapshot` < (empty/`ga`/`final`/`release`) < `sp`. Free-form identifiers not in this list are treated as stable.
 * **pip/pipenv/pip-compile/poetry (PEP 440):** The table lists canonical prerelease and postrelease suffixes per PEP 440. Aliases are also recognized and normalized to their canonical forms (`alpha`→`a`, `beta`→`b`, `c`/`pre`/`preview`→`rc`, `rev`/`r`→`post`). Epoch versions (`N!...`) and local versions (`+local`) are supported; local version segments are used only to break ties when the public version is identical.
+* **Elm:** The Elm package registry enforces strict SemVer (`MAJOR.MINOR.PATCH` integers only) and does not allow publishing pre-release versions. Dependabot compares versions numerically.
+* **Go modules:** Follows SemVer with a mandatory `v` prefix. Pseudo-versions (`v0.0.0-YYYYMMDDHHMMSS-commithash`) are used for unreleased commits and are always treated as pre-release. The `+incompatible` suffix marks modules at major version 2+ without a `go.mod` file and does not affect version ordering.
+* **git submodule:** Dependabot tracks the latest commit on the configured branch. There is no version comparison—updates always move the pinned SHA forward. If the submodule tracks a tag, Dependabot follows the tag's commit.

src/search/lib/ai-search-constants.ts

@@ -2,3 +2,9 @@
 // payloads are almost always pasted docs pages or unrelated content
 // and either time out or return no-answer. See github/cse-copilot#1214.
 export const MAX_QUERY_LENGTH = 500
+
+// cse-copilot returns HTTP 400 with this code in `detail.code` when Azure's
+// Responsible AI input content filter rejects a query. These are expected,
+// user-triggered rejections, so they are tracked separately and kept out of
+// the AI search error-rate monitor. See github/cse-copilot#1214.
+export const RAI_CONTENT_FILTER_CODE = 'RAI_INPUT_CONTENT_POLICY_BREACH_ERROR'

src/search/lib/ai-search-proxy.ts

@@ -6,7 +6,7 @@ import { getHmacWithEpoch } from '@/search/lib/helpers/get-cse-copilot-auth'
 import { getCSECopilotSource } from '@/search/lib/helpers/cse-copilot-docs-versions'
 import type { ExtendedRequest } from '@/types'
 import { handleExternalSearchAnalytics } from '@/search/lib/helpers/external-search-analytics'
-import { MAX_QUERY_LENGTH } from '@/search/lib/ai-search-constants'
+import { MAX_QUERY_LENGTH, RAI_CONTENT_FILTER_CODE } from '@/search/lib/ai-search-constants'
 
 const logger = createLogger(import.meta.url)
 
@@ -15,6 +15,27 @@ const logger = createLogger(import.meta.url)
 // established, but the connect + first-byte must complete within this window.
 const AI_SEARCH_TIMEOUT_MS = 9_000
 
+type ContentFilterCandidate = {
+  status: number
+  headers: { get: (name: string) => string | null }
+  json: () => Promise<unknown>
+}
+
+const isContentFilterRejection = async (response: ContentFilterCandidate): Promise<boolean> => {
+  if (response.status !== 400) {
+    return false
+  }
+  if (!response.headers.get('content-type')?.includes('application/json')) {
+    return false
+  }
+  try {
+    const body = (await response.json()) as { detail?: { code?: string } }
+    return body?.detail?.code === RAI_CONTENT_FILTER_CODE
+  } catch {
+    return false
+  }
+}
+
 export const aiSearchProxy = async (req: ExtendedRequest, res: Response) => {
   const { query, version } = req.body ?? {}
 
@@ -100,8 +121,13 @@ export const aiSearchProxy = async (req: ExtendedRequest, res: Response) => {
 
     if (!response.ok) {
       const errorMessage = `Upstream server responded with status code ${response.status}`
-      logger.error(errorMessage, { statusCode: response.status })
-      statsd.increment('ai-search.stream_response_error', 1, diagnosticTags)
+      if (await isContentFilterRejection(response)) {
+        logger.info(errorMessage, { statusCode: response.status })
+        statsd.increment('ai-search.content_filtered', 1, diagnosticTags)
+      } else {
+        logger.error(errorMessage, { statusCode: response.status })
+        statsd.increment('ai-search.stream_response_error', 1, diagnosticTags)
+      }
       res.status(response.status).json({
         errors: [{ message: errorMessage }],
         upstreamStatus: response.status,

src/search/tests/ai-search-proxy.ts

@@ -0,0 +1,148 @@
+import { describe, test, expect, vi, beforeEach } from 'vitest'
+
+import statsd from '@/observability/lib/statsd'
+import { fetchStream } from '@/frame/lib/fetch-utils'
+import { aiSearchProxy } from '@/search/lib/ai-search-proxy'
+import { RAI_CONTENT_FILTER_CODE } from '@/search/lib/ai-search-constants'
+import type { ExtendedRequest } from '@/types'
+
+vi.mock('@/observability/lib/statsd', () => ({
+  default: { increment: vi.fn(), gauge: vi.fn() },
+}))
+
+vi.mock('@/frame/lib/fetch-utils', () => ({
+  fetchStream: vi.fn(),
+}))
+
+vi.mock('@/search/lib/helpers/get-cse-copilot-auth', () => ({
+  getHmacWithEpoch: () => 'test-auth',
+}))
+
+vi.mock('@/search/lib/helpers/cse-copilot-docs-versions', () => ({
+  getCSECopilotSource: () => 'docs',
+}))
+
+vi.mock('@/search/lib/helpers/external-search-analytics', () => ({
+  handleExternalSearchAnalytics: async () => null,
+}))
+
+const incrementedMetrics = () =>
+  (statsd.increment as ReturnType<typeof vi.fn>).mock.calls.map((call) => call[0])
+
+function buildResponse() {
+  const res = {
+    statusCode: 0,
+    body: null as unknown,
+    status(code: number) {
+      this.statusCode = code
+      return this
+    },
+    json(payload: unknown) {
+      this.body = payload
+      return this
+    },
+    setHeader: vi.fn(),
+    flushHeaders: vi.fn(),
+    write: vi.fn(),
+    end: vi.fn(),
+    headersSent: false,
+  }
+  return res
+}
+
+function buildRequest(): ExtendedRequest {
+  return {
+    body: { query: 'a disallowed query', version: 'dotcom' },
+    language: 'en',
+  } as unknown as ExtendedRequest
+}
+
+function mockUpstream(
+  status: number,
+  jsonBody: unknown | (() => never),
+  contentType = 'application/json',
+) {
+  const json = vi.fn(async () => {
+    if (typeof jsonBody === 'function') {
+      return (jsonBody as () => never)()
+    }
+    return jsonBody
+  })
+  ;(fetchStream as ReturnType<typeof vi.fn>).mockResolvedValue({
+    ok: status < 400,
+    status,
+    headers: {
+      get: (name: string) => (name.toLowerCase() === 'content-type' ? contentType : null),
+    },
+    json,
+  })
+  return json
+}
+
+describe('aiSearchProxy upstream 400 handling', () => {
+  beforeEach(() => {
+    vi.clearAllMocks()
+  })
+
+  test('RAI content filter 400 increments content_filtered, not stream_response_error', async () => {
+    mockUpstream(400, {
+      code: 400,
+      message: 'Responsible AI input content policy breach',
+      detail: { code: RAI_CONTENT_FILTER_CODE },
+    })
+    const res = buildResponse()
+
+    await aiSearchProxy(buildRequest(), res as never)
+
+    const metrics = incrementedMetrics()
+    expect(metrics).toContain('ai-search.content_filtered')
+    expect(metrics).not.toContain('ai-search.stream_response_error')
+    expect(metrics).toContain('ai-search.call')
+    expect(res.statusCode).toBe(400)
+  })
+
+  test('non-RAI 400 increments stream_response_error', async () => {
+    mockUpstream(400, { code: 400, message: 'some other bad request' })
+    const res = buildResponse()
+
+    await aiSearchProxy(buildRequest(), res as never)
+
+    const metrics = incrementedMetrics()
+    expect(metrics).toContain('ai-search.stream_response_error')
+    expect(metrics).not.toContain('ai-search.content_filtered')
+    expect(res.statusCode).toBe(400)
+  })
+
+  test('malformed 400 body increments stream_response_error', async () => {
+    mockUpstream(400, () => {
+      throw new Error('invalid json')
+    })
+    const res = buildResponse()
+
+    await aiSearchProxy(buildRequest(), res as never)
+
+    const metrics = incrementedMetrics()
+    expect(metrics).toContain('ai-search.stream_response_error')
+    expect(metrics).not.toContain('ai-search.content_filtered')
+    expect(res.statusCode).toBe(400)
+  })
+
+  test('non-JSON 400 body increments stream_response_error without parsing', async () => {
+    const json = mockUpstream(
+      400,
+      () => {
+        throw new Error('should not be parsed')
+      },
+      'text/html',
+    )
+    const res = buildResponse()
+
+    await aiSearchProxy(buildRequest(), res as never)
+
+    const metrics = incrementedMetrics()
+    expect(json).not.toHaveBeenCalled()
+    expect(metrics).toContain('ai-search.stream_response_error')
+    expect(metrics).not.toContain('ai-search.content_filtered')
+    expect(res.statusCode).toBe(400)
+  })
+})