solpbc.org / solstone
personal memory agent
0
fork

Configure Feed

Select the types of activity you want to include in your feed.

fix: raise enrichment max_output_tokens from 8192 to 16384

Segments ≥360s (420-540s) were hitting the output token ceiling during
enrichment, causing IncompleteJSONError and losing all enrichment metadata
(topics, setting, emotions, corrections) for exactly the longest and most
valuable audio segments. The model (gemini-3-flash-preview) supports 65K+
output tokens; 8192 was unnecessarily conservative. Billing is per token
generated, so short segments are unaffected.

Fixes: D2 from field journal validation triage.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Jer Miller 487281bf 860d585d

+1 -1
+1 -1
observe/enrich.py
··· 109 109 contents=contents, 110 110 context="observe.enrich", 111 111 temperature=0.3, 112 - max_output_tokens=8192, 112 + max_output_tokens=16384, 113 113 thinking_budget=4096, 114 114 json_output=True, 115 115 )