Submit private API feedback

Submit a bug, feature request, documentation issue or model-quality report when the user asks, or when a NanoGPT problem is reproducible and blocks the task. Do not submit reports for transient errors, timeouts, rate limits or insufficient balance. The @nanogpt/mcp server exposes this as nanogpt_send_feedback. Requires a valid inference API key and Idempotency-Key. The key's account, or the team that pays for it, must have a balance or an active subscription; other accounts get 403 feedback_requires_funded_account and can use the website. Feedback requests do not incur usage charges. Browser cookies, management credentials, partner tokens and accountless x402 grant no access. All reports are private and are never published to the public feedback board. Support sees each report as submitted through an API key, not as a verified request from the account owner, so it cannot be used to request refunds, balance or account changes. Submission explicitly stores the supplied text; inference no-logging settings do not prevent feedback storage. Never send credentials or full prompts/responses. No logs are fetched or support-access permissions changed. Uncompressed JSON only, at most 16 KiB. Concurrent retries with the same key, Idempotency-Key and normalized fields return the original 201 receipt; changed fields return 409. Deleted or archived reports return 410 rather than being recreated. Quotas include retries: 5 submissions/minute and 20/day per account and per paying account (members of a team share their team's allowance), 60 requests/minute per key, 120/minute per IP, plus a shared daily limit on valid submissions across all accounts (429 feedback_capacity_reached). Aliases: /v1/feedback. Read the current status using GET; receipt acceptance does not promise a fix. Limit automatic retries to three total attempts per failed operation, honoring Retry-After when present and using backoff otherwise; use website support if failures persist.