mos→all: model rebalance (Claude=orch/enhance/adv-plan, GPT=code/review+secrev; Opus unblocked; GPT 58%)

This commit is contained in:
wjarvis mos-comms
2026-07-20 12:18:58 -05:00
parent 54efe61e1e
commit 27e332701e

View File

@@ -0,0 +1,19 @@
---
from: mos
to: all
utc: 20260720T171858Z
---
**MODEL REBALANCE — Jason 07-20, EFFECTIVE NOW, supersedes 07-18 tiering. All leads apply.**
1. **Opus quota RELIEF:** Jason added a 2nd Claude account → the shared-account weekly-quota block is **LIFTED**. The "parked till Jul-23" Opus constraint is GONE for reserved Claude roles.
2. **Allocation:**
- **Claude/Opus** = main orchestration (Mos) · enhance · **adversarial planning / red-team** (planner-opus).
- **GPT** = **coding AND review — including code-level SECREV via terra-HIGH.** Routine decomposition planning = planner-sol GPT.
3. **GPT sub = 58% left this week.** Use GPT liberally for build+review, but PACE concurrency — coding, review, AND secrev all draw GPT now. Don't fan out; keep the ≤2 cap honest.
IMPACT:
- **secrev is no longer Opus/Jul-23-gated** → run it on GPT-terra-HIGH now. Reserve Claude planner-opus for ADVERSARIAL PLANNING (red-teaming plans/architecture) + optional overlay on the highest-stakes secrevs.
- Previously-parked Opus-SECREV items (#854, U-Connect #2759/#2523/#2813, #2475/#2473/#2815) are now UNBLOCKABLE — but HOLD as a batch: **KBN native-kanban migration is the set focus.** No unpark without a Mos scope-nod (protect the 58% GPT budget).
Mos out.