3
49%
main: 49%

Ran 17 Nov 2025 05:15PM UTC

Files 82

Run time 91min

Badge

Embed ▾

Committed 17 Nov 2025 05:14PM UTC coverage: 50.056% (+0.1%) from 49.916%

Job # aa2682570247c381c9c18b57f5416737347fe28e.3

Build Type

push

github

Committed by

web-flow

Commit Message

Fix reasoning_overlay pattern match for llmdb structure (#219)

Fix reasoning_overlay pattern match for llmdb structure

The reasoning_overlay() test helper was pattern matching on
`reasoning: true` but llmdb stores the capability as
`reasoning: %{enabled: true}`. This mismatch caused the function
to never match reasoning models, preventing it from applying the
required token budget increases.

Without the token bump, reasoning models like gemini-2.5-pro would
record fixtures with only 50 tokens, causing MAX_TOKENS finish
with no output since extended thinking consumes tokens before
generating any response text.

Changes:
- Updated pattern match from `reasoning: true` to
  `reasoning: %{enabled: true}` in test/support/helpers.ex:494
- Added test coverage demonstrating the bug and verifying fix

This allows reasoning models to correctly receive 4001+ token
budgets during fixture recording, preventing truncation errors.

Run Details

3582 of 7156 relevant lines covered (50.06%)

14.7 hits per line

agentjido / req_llm / aa2682570247c381c9c18b57f5416737347fe28e / 3
49%
main: 49%

README BADGES
x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job aa2682570247c381c9c18b57f5416737347fe28e.3

agentjido / req_llm / aa2682570247c381c9c18b57f5416737347fe28e / 3 49% main: 49%

README BADGES x

Markdown

Textile

RDoc

HTML

Rst

Source Files on job aa2682570247c381c9c18b57f5416737347fe28e.3

agentjido / req_llm / aa2682570247c381c9c18b57f5416737347fe28e / 3
49%
main: 49%

README BADGES
x