engine room / answer trace

wrun_01M1VSXW4K6B282Y7WW5DM9HP6

Answered in 3.46s, first token at 2.81s, 1,388 tokens, $0.00024, 1 verified citation.

Visitor identity is a one-way hash; message content never enters traces.

Timeline

Queue

enqueue + dispatch

Guardrail

allow

Dispatch

step handoff

Retrieval

hybrid

Dispatch

step handoff

Generation

glm-5.3-flash

25ms
627ms
278ms
366ms
765ms
1.40s
0first token 2.81s3.46s

The gray spans are durable-workflow dispatch: each stage is an independently retried, checkpointed step, and the handoffs are what that crash-safety costs.

What happened

  1. Admitted at the edge

    A 21-character message from caller 3e48ef9d… passed schema validation, the per-caller rate limit, and both token budgets (charged an estimate of 1,030 tokens, settled later against actual usage).

  2. Became a durable workflow

    The endpoint never calls a model. At 16:48:20.119 the request was enqueued as run wrun_01M1VSXW4K6B2…; tokens stream into the run’s persistent log, which is why a page reload can reattach mid-answer.

  3. Cleared the guardrail

    gpt-oss-safeguard classified the message against policy 6bb17bbe0908 in 627ms: allow.

    “User asks about seeing the résumé, a normal topic related to Bharathi.”

  4. Pulled its evidence

    Hybrid search (dense embeddings fused with BM25) returned 6 passages in 366ms; only these entered the model’s context.

    1. faq#00.0328
    2. profile/experience#20.0323
    3. profile/about#30.0321
    4. profile/awards#20.0320
    5. profile/journey#20.0320
    6. profile/about#00.0305

  5. Generated

    zai/glm-5.3-flash streamed the answer, first token 2.81s in, spending 1,304 tokens in / 84 out ($0.00024).

  6. Verified and settled

    Every citation was checked against the concept registry and resolves. Token budgets were settled against actual usage and this trace was written.

    faq ×2

Record

Status
finished (stop)
Admission
admitted
Guardrail reason
User asks about seeing the résumé, a normal topic related to Bharathi.
Tokens in / out
1,304 / 84
Cost
$0.00024
Model
zai/glm-5.3-flash
Prompt version
45b79cdd7ce5
Policy version
6bb17bbe0908
Input size
21 chars
Caller (hashed)
3e48ef9d22e096da
Request id
a3a4f8c9-e60c-49cb-b3b6-f1de55b90c8b
Enqueued
2026-09-06T16:48:20.119Z
Finished
2026-09-06T16:48:23.578Z