1 points | by AhmedAllam0 10 hours ago ago
2 comments
this is useful. rejected branches are usually where latency and token burn hide. would love a small cost timeline per rejected decision too, that makes tuning loops way faster.
Great point,this is exactly the kind of signal I hadn't thought to surface explicitly
Right now GhostTrace logs reasoning per rejection, but not token cost or latency per branch
Adding a cost timeline (tokens + ms per phantom) is going on the roadmap.
It would make the .ghost.json file genuinely useful for optimization, not just debugging.
this is useful. rejected branches are usually where latency and token burn hide. would love a small cost timeline per rejected decision too, that makes tuning loops way faster.
Great point,this is exactly the kind of signal I hadn't thought to surface explicitly
Right now GhostTrace logs reasoning per rejection, but not token cost or latency per branch
Adding a cost timeline (tokens + ms per phantom) is going on the roadmap.
It would make the .ghost.json file genuinely useful for optimization, not just debugging.