SirStone/SirRoboGarage
1
0
Fork 0
Code Issues 28 Pull Requests Actions Projects Releases Wiki Activity

What does "better" mean for best-so-far weights? #89

New Issue
Closed
opened 2026-08-26 06:55:32 +02:00 by SirStone · 2 comments
SirStone commented 2026-08-26 06:55:32 +02:00
Owner
Copy Link
Copy Source

part-of: #88

Question

Define the in-bot comparable metric that decides whether a candidate champion replaces the stored best for an enemy. Candidates: GA internal champion fitness, observed battle/round performance, hybrid. Must be computable at save time, deterministic enough that "replace" beats "keep" correctly. Resolution fixes the comparison used by the write policy.

part-of: #88 ## Question Define the in-bot comparable metric that decides whether a candidate champion replaces the stored best for an enemy. Candidates: GA internal champion fitness, observed battle/round performance, hybrid. Must be computable at save time, deterministic enough that "replace" beats "keep" correctly. Resolution fixes the comparison used by the write policy.
SirStone added the wayfinder:grilling label 2026-08-26 06:55:32 +02:00
SirStone commented 2026-08-26 07:01:45 +02:00
Author
Owner
Copy Link
Copy Source

claimed by orchestrator session (ox-alpha), 2026-08-26

claimed by orchestrator session (ox-alpha), 2026-08-26
SirStone commented 2026-08-26 07:09:37 +02:00
Author
Owner
Copy Link
Copy Source

Resolution

The comparison metric for replacing the stored best per enemy is the GA's internal champion fitness:

  • Ground truth: GA training fitness crowns the best — contingent on virtual-gun simulation being correct (recorded assumption; vgun fidelity is load-bearing for metric trustworthiness, out of scope here).
  • Granularity: per generation.
  • Noise policy: single observation suffices — no sustained-evidence requirement. A lucky generation stealing the best slot is accepted; the GA retrains around it.
  • Cold start: no special case. The global champion exists as baseline from game start; the first per-enemy champion saves exactly like any later one.
## Resolution The comparison metric for replacing the stored best per enemy is the **GA's internal champion fitness**: - **Ground truth**: GA training fitness crowns the best — contingent on virtual-gun simulation being correct (recorded assumption; vgun fidelity is load-bearing for metric trustworthiness, out of scope here). - **Granularity**: per generation. - **Noise policy**: single observation suffices — no sustained-evidence requirement. A lucky generation stealing the best slot is accepted; the GA retrains around it. - **Cold start**: no special case. The global champion exists as baseline from game start; the first per-enemy champion saves exactly like any later one.
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
ready-for-agent
wayfinder:grilling
Wayfinder design question to grill
 wayfinder:map
Wayfinder map issue
 wayfinder:prototype
wayfinder:research
Wayfinder research question
 wayfinder:task
Wayfinder implementation task
No Label wayfinder:grilling
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
SirStone
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: SirStone/SirRoboGarage#89
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?