fix(extraction): bump EXTRACTION_MAX_TOKENS 4096 → 8192

[moralespanitz]

Summary

Extraction LLM was truncating JSON output at ~14 KB during BEAM Sprint 2 CR mini-slice runs on dense 10-turn chunks. Bumping the max_tokens budget from 4096 → 8192 prevents the truncation.

Evidence

Server log during iter 7 (first attempt) before this fix:

[extractFacts] JSON parse failed (Unterminated string in JSON at position 14152 (line 339 column 8)); attempting repair
[extractFacts] JSON parse failed (Unterminated string in JSON at position 14172 ...)
[extractFacts] JSON parse failed (Unterminated string in JSON at position 14284 ...)
[extractFacts] JSON parse failed (Unterminated string in JSON at position 14301 ...)
[extractFacts] JSON parse failed (Unterminated string in JSON at position 14172 ...)
[extractFacts] JSON parse failed (Unterminated string in JSON at position 15084 ...)

Six truncations on one ingest pass. Conv-3 crashed.

After bumping to 8192: zero truncation across iter 7 v3 N=3 full-ingest reruns.

Risk

Marginal cost increase: Anthropic bills only for output tokens actually generated, and only the dense chunks that previously truncated will use more tokens. Most extractions stay well under 4096.

Companion changes (separate PRs)

• atomicmemory-benchmarks PR #8 lowers harness chunk size from 10 → 5 turn-pairs (defense-in-depth on the input side).

Test plan

• Server starts and runs with the change
• iter 7 v3 N=3 full-ingest reruns: no JSON parse failures
• npx tsc --noEmit clean
• npm test
• fallow --no-cache