Investigate p99 latency regression

design-system · claude-opus-4-8 · 104m · 0m ago · Liam Patel

Part of: Form component consolidation

Session tokens

6.1M tokens

1.2M raw tokens

Avoidable tokens (est.)

728K tokens

12% of session

Quality

craft check score

Outcome

In progress

9 cycles · 1 corrections · 104m wall-clock

PromptGenerateVerifyCorrect

8maccepted

12maccepted

14mrejected

12maccepted

14maccepted

correction

9maccepted

accepted w/o verify

10maccepted

14maccepted

11mrejected

8 verify events · 1 accepts without verify — chips are pinned to the cycles that triggered them.

Grounded answers with mandatory cycle citations

Answers are PROXY — generated over the measured transcript. This query is visible to the engineer in their access log.

What drove the score — pinned to its cycles

Verification discipline86
1 of 7 accepts had no verify event between diff and accept
Plan-mode discipline45
No plan-mode use in this session
Correction load86
1 of 9 cycles were corrections of earlier output
Prompt clarity76
Context, constraints, and a done-condition raise first-pass acceptance

Caught in this session

low
Prod access
Session touched production database credentials
Flagged in-session by your risk rules