Planning: Add provider-aware selection, health signals, and HydraFusion positioning #18

New issue

Closed

opened 2026-09-05 09:29:43 +00:00 by creatidy · 0 comments

creatidy commented

2026-09-05 09:29:43 +00:00

(Migrated from github.com)

Record the next planning refinements before Task 3/M2 implementation:

assess GitHub Project HydraFusion as current prior art and clarify Scarcity Router positioning;
make bounded orchestration a product/service invariant, not only repository-agent governance;
add configurable provider blackout/schedule policy, initially for avoiding Z.ai during user-defined peak hours;
represent OpenAI banked Codex resets as replenishment capacity without silently consuming them;
plan provider/service health inputs: OpenAI status, direct local Ollama health, and provider-native Z.ai signals; explicitly reject status.hellozai.com because it is Zai Payments, not Z.ai/GLM;
assess optional reuse/integration boundary with creatidy/ollama-monitoring without merging projects by default;
plan Artificial Analysis as a cached, provenance-bearing capability/performance evidence source, not runtime capacity or selector truth;
do not persist API keys in repository docs or fixtures.

This is documentation/planning only. Do not implement collectors, selection logic, orchestration runtime, or Task 3 in this issue.

Record the next planning refinements before Task 3/M2 implementation: - assess GitHub Project HydraFusion as current prior art and clarify Scarcity Router positioning; - make bounded orchestration a product/service invariant, not only repository-agent governance; - add configurable provider blackout/schedule policy, initially for avoiding Z.ai during user-defined peak hours; - represent OpenAI banked Codex resets as replenishment capacity without silently consuming them; - plan provider/service health inputs: OpenAI status, direct local Ollama health, and provider-native Z.ai signals; explicitly reject status.hellozai.com because it is Zai Payments, not Z.ai/GLM; - assess optional reuse/integration boundary with creatidy/ollama-monitoring without merging projects by default; - plan Artificial Analysis as a cached, provenance-bearing capability/performance evidence source, not runtime capacity or selector truth; - do not persist API keys in repository docs or fixtures. This is documentation/planning only. Do not implement collectors, selection logic, orchestration runtime, or Task 3 in this issue.