[BUG] Console muse-spark-1.3-contributor-free: 400 'reasoning encrypted_content was not issued to this caller' (zen/v1/responses)

[Vargas876]

Summary

OpenCode Desktop (Windows) with the free model muse-spark-1.3-contributor-free served by the Console provider (https://opencode.ai/zen/v1/responses, OpenAI Responses API) intermittently fails with HTTP 400: [invalid_request_error] reasoning \encrypted_content` was not issued to this caller`. The failure escalates over the life of a session and survives cleaning every reasoning blob from the message history, restarting the app, and upgrading the app (1.18.21 -> 1.18.31).

Steps to reproduce

1. In OpenCode Desktop, start a session with model opencode/muse-spark-1.3-contributor-free (Console provider).
2. Send a few messages (1 to several).
3. A later request fails with the error above (isRetryable: false).

The exact error recorded in the app event store:

{
  "name": "APIError",
  "data": {
    "message": "Error from provider (Console): Upstream request failed: [invalid_request_error] reasoning `encrypted_content` was not issued to this caller",
    "statusCode": 400,
    "isRetryable": false,
    "responseHeaders": {
      "cf-placement": "remote-ORD",
      "cf-ray": "a3acf5b85a095c86-MIA",
      "date": "Mon, 14 Sep 2026 05:23:34 GMT",
      "server": "cloudflare"
    },
    "responseBody": "{\"model\":\"muse-spark-1.3-contributor-free\",\"error\":{\"param\":null,\"type\":\"invalid_request_error\",\"message\":\"Error from provider (Console): Upstream request failed: [invalid_request_error] reasoning `encrypted_content` was not issued to this caller\"}}",
    "metadata": { "url": "https://opencode.ai/zen/v1/responses" }
  }
}

What I already tried (none fixed it)

• Upgraded the desktop app from 1.18.21 to 1.18.31 (fully applied).
• Model options with store: false + include: ["reasoning.encrypted_content"] (makes the blob replay explicit -> same error).
• Model options with reasoningEffort: "none" + reasoningSummary: "auto" (blobs still emitted by the upstream).
• Removed all 160 reasoningEncryptedContent blobs from opencode.db; new sessions regenerate them and the error returns.
• New session sometimes starts clean, then the error returns as history grows.

Evidence

• Reasoning parts stored in opencode.db all carry an itemId (rs_...) plus reasoningEncryptedContent, content: null, summary: [] -- the upstream only hands out the opaque encrypted_content, consistent with Zen gateway does not forward reasoning/thinking stream for Muse Spark 1.2 #43584 for Muse Spark through the zen gateway.
• The app event store has 55 instances of this exact 400 error. cf-placement: remote-ORD is constant in all of them, but the Cloudflare edge (cf-ray) varies across MIA, ATL, PHL, PIT, CLT, TPA, TLH.
• With store defaulting to true, the 1.18.31 engine emits item_reference for reasoning (it does not resend the plaintext blob), yet the gateway still reports that the encrypted_content "was not issued to this caller".

Suspected root cause

The zen gateway issues encrypted_content bound to a specific caller/session when the item is created, but a later request that references that item is handled by a different backend/caller context (routing, worker, or upstream binding), so the signature check fails. This matches the class of bug in:

• Model in @ai-sdk/openai sdk fails on the 2nd request #5104 (Responses API reasoning items must hit the same server each time; fixed there by removing id+encrypted_content replay under store: false).
• [Bug] @ai-sdk/azure Responses API: encrypted content verification fails after 3-4 tool-calling turns (store: false) #29879 (GitHub-adjacent report: cross-region LB routing broke the encrypted-content chain; upstream fixed it with session pinning, X-Target-Id).

Here the failure happens even when opencode uses item_reference, so the fix likely belongs on the zen gateway side: pin the caller/session to the issuing backend, or stop issuing encrypted_content for models that only provide the opaque blob (fall back to summaries).

Environment

• OS: Windows 11, 64-bit
• App: OpenCode Desktop 1.18.31
• Provider: Console (opencode.ai zen), model muse-spark-1.3-contributor-free
• Endpoint: https://opencode.ai/zen/v1/responses

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• [Bug] muse-spark-1.3-contributor-free: reasoning encrypted_content was not issued to this caller when switching models in same session #48805: Same error (reasoning encrypted_content was not issued to this caller) with muse-spark-1.3-contributor-free, triggered by model switching in the same session
• Muse Spark 1.3: resuming a session after idle fails with reasoning encrypted_content not issued to this caller #48915: Same error with Muse Spark 1.3 when resuming a session after idle
• Getting Error from provider (Console): Upstream request failed: [invalid_request_error] reasoning encrypted_content was not issued to this caller #48973: Same exact error message from the Console provider
• Network interruption corrupts conversation state: reasoning encrypted_content was not issued to this caller #49083: Same error triggered by network interruption corrupting conversation state
• [Bug] Muse Spark 1.3 Free (Xhigh): reasoning encrypted_content error on follow-up prompts with images #48947: Same error with Muse Spark 1.3 Free on follow-up prompts with images
• Zen gateway does not forward reasoning/thinking stream for Muse Spark 1.2 #43584: Related — Zen gateway reasoning/thinking issues with Muse Spark family

If your case is meaningfully different from all of the above (e.g. a distinct trigger or repro scenario not covered there), please comment on one of those issues or clarify here. Otherwise, consider adding a +1 or additional context to the most relevant existing issue.

[Vargas876]

Duplicate of #48915 (same underlying bug: caller-bound \encrypted_content\ replayed across load-balanced zen egress invalidates the blob, bricking the session). Our evidence (mid-session trigger without idle, desktop Windows, constant \cf-placement: remote-ORD\ with rotating \cf-ray\ edges, persistence across 1.18.21->1.18.31 and DB blob cleanup) has been merged into #48915.

There is an open fix PR: #48918 (recover stale encrypted reasoning — drop
easoningEncryptedContent/itemId and retry). Closing this as duplicate to keep the tracker consolidated.