Unverified50% confidenceFactExact time
The default eager eviction policy keeps only the raw text of the most recent three tool results, capped at 6000 tokens.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
UnverifiedThe script's only stopping condition was accumulating 50 accepted copy items; rejected attempts did not count toward this total, and there was no maximum attempt limit or timeout fallback.62% similarUnverifiedThe script's history-trimming function only bulk-removed old messages when the token count exceeded approximately 180,000; it did not stop the task, but merely kept requests below the context limit.60% similarUnverifiedIn the benchmark, each optimizer attempt is limited to 8 tool calls, at most 6 successful edits, and 2 benchmark feedback calls, with a 600-second guard, and modifications to weights, tokenizer, stopping rules, threads, and external libraries are prohibited.59% similarUnverifiedCompressing or truncating historical context leads to degraded task continuity, potentially creating a perverse incentive where saving tokens actually requires more tokens.57% similarUnverifiedThe Max mode mechanism differs from other tiers in that it removes the "think less" option, forcing the model to use the full thinking token budget.56% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1011771API
curl https://kongchang.com/api/v1/knowledge/claims/1011771MCP
get_claim(id=1011771)