fix(ops): groq-key restart loop, disk cleanup, context sync - #235

Draft

JoTalbot wants to merge 33 commits into

mainfrom

agent/20260814-quant-backfill-ppo

Commits on Aug 14, 2026

feat(quant): scale-free ML direction model v2 + honest OOS eval script
JoTalbot
committed
8d668f0
docs(coordination): record paper entry unblock (ML model v2)
JoTalbot
committed
cd0ee30
docs(coordination): record paper entry unblock in context
JoTalbot
committed
03ac5a3
docs(coordination): session paper-fix deployment verification
JoTalbot
committed
aa8832f
fix(quant): RL bridge onehot/features/clamp + drop dead tickers
JoTalbot
committed
05bd3fd
docs(coordination): close RL bridge and dead ticker items
JoTalbot
committed
96d716d
feat(quant): backfill history, retrain ML/PPO, fix signal source selection
JoTalbot
committed
9501cf2
docs(coordination): record backfill + ML/PPO retrain stage
JoTalbot
committed
35fff86
feat(quant): orderbook research expansion + docs sync
JoTalbot
committed
bda4d3b
docs(coordination): record orderbook research expansion
JoTalbot
committed
ed6689b
chore(quant): mm-watcher auto-runs market-making sim at 1000 snapshots/pair
JoTalbot
committed
0e303a9
fix(quant): bitstamp has no SOL/USDT - per-exchange symbol filter; interval 5s
JoTalbot
committed
e622845
docs(coordination): orderbook watcher record
JoTalbot
committed
ac8dbb2
feat(quant): inventory-aware MM simulator v2 + WATCH signals in morning brief
JoTalbot
committed
a4b1c6f
docs(coordination): MM v2 research + morning brief WATCH
JoTalbot
committed
d306aac
fix(ops): health check watches current models (ppo_v9, catboost v2)
JoTalbot
committed
3be4cd2
docs(coordination): disk cleanup + health check model paths
JoTalbot
committed
3e857ff
docs(coordination): final MM run on 14.5k snapshots - negative verdict
JoTalbot
committed
dce3ca0
feat(quant): WATCH backtest, ML drift monitor, feature experiment, retrain timers
JoTalbot
committed
3171c6b
docs(coordination): record WATCH verification + ML automation stage
JoTalbot
committed
839c66a
chore(ops): sync orderbook unit to deployed state (6 exchanges, 5s interval)
JoTalbot
committed
ee84cae
fix(quant): honest PPO OOS validation - clamp like inference bridge
JoTalbot
committed
174c395
fix(quant): apply action clamp to legacy train script validation too
JoTalbot
committed
db27bdb
docs(coordination): record honest PPO OOS finding + horizon experiment
JoTalbot
committed
ef0ee81
docs(coordination): record 10k backfill + retrain no-improvement result
JoTalbot
committed
4b0a1e6
feat(quant): monthly backtest of deployed Directional v2 rules
JoTalbot
committed
4f5f269
feat(quant): monthly backtest --months window parameter
JoTalbot
committed
d70286b
feat(quant): strategy research - found positive-OOS trend-following LS
JoTalbot
committed
7c88f3a
docs(coordination): record profitable strategy research result
JoTalbot
committed
11de569
feat(quant): corrected engine - compounding + next-bar timing; MA-LS earns
JoTalbot
committed
9f95fbb
docs(coordination): record earning algorithm MA-LS finding
JoTalbot
committed
ed052f3

Commits on Aug 15, 2026

docs(coordination): read-only project analysis session 2026-08-15
Arena.ai Agent Mode
committed
fd38614
fix(ops): stop aios-groq-key restart loop, disk cleanup, context sync
Arena.ai Agent Mode
committed
907b094

Navigation Menu

fix(ops): groq-key restart loop, disk cleanup, context sync - #235

fix(ops): groq-key restart loop, disk cleanup, context sync#235
JoTalbot wants to merge 33 commits into
mainfrom
agent/20260814-quant-backfill-ppo

Commits on Aug 14, 2026

feat(quant): scale-free ML direction model v2 + honest OOS eval script

docs(coordination): record paper entry unblock (ML model v2)

docs(coordination): record paper entry unblock in context

docs(coordination): session paper-fix deployment verification

fix(quant): RL bridge onehot/features/clamp + drop dead tickers

docs(coordination): close RL bridge and dead ticker items

feat(quant): backfill history, retrain ML/PPO, fix signal source selection

docs(coordination): record backfill + ML/PPO retrain stage

feat(quant): orderbook research expansion + docs sync

docs(coordination): record orderbook research expansion

chore(quant): mm-watcher auto-runs market-making sim at 1000 snapshots/pair

fix(quant): bitstamp has no SOL/USDT - per-exchange symbol filter; interval 5s

docs(coordination): orderbook watcher record

feat(quant): inventory-aware MM simulator v2 + WATCH signals in morning brief

docs(coordination): MM v2 research + morning brief WATCH

fix(ops): health check watches current models (ppo_v9, catboost v2)

docs(coordination): disk cleanup + health check model paths

docs(coordination): final MM run on 14.5k snapshots - negative verdict

feat(quant): WATCH backtest, ML drift monitor, feature experiment, retrain timers

docs(coordination): record WATCH verification + ML automation stage

chore(ops): sync orderbook unit to deployed state (6 exchanges, 5s interval)

fix(quant): honest PPO OOS validation - clamp like inference bridge

fix(quant): apply action clamp to legacy train script validation too

docs(coordination): record honest PPO OOS finding + horizon experiment

docs(coordination): record 10k backfill + retrain no-improvement result

feat(quant): monthly backtest of deployed Directional v2 rules

feat(quant): monthly backtest --months window parameter

feat(quant): strategy research - found positive-OOS trend-following LS

docs(coordination): record profitable strategy research result

feat(quant): corrected engine - compounding + next-bar timing; MA-LS earns

docs(coordination): record earning algorithm MA-LS finding

Commits on Aug 15, 2026

docs(coordination): read-only project analysis session 2026-08-15

fix(ops): stop aios-groq-key restart loop, disk cleanup, context sync