case/workspace
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
master
workspace/memory/2026-02-09.md

5.2 KiB
Raw Permalink Blame History

2026-02-09 — Monday

X Feed Analysis Day

Major session analyzing Polymarket/crypto tweets D J forwarded from X feed.

Tweets Investigated

  1. @browomo weather edge — Pilot METAR data for Polymarket weather bets

    • Wallet 0x594edB9112f...: Claimed +$27K, actual -$13,103 (387 losses, 51 wins)
    • Verdict: SCAM/engagement bait

  2. @ArchiveExplorer planktonXD — "Buy everything under 5 cents"

    • Wallet 0x4ffe49ba2a4c...: Claimed +$104K, actual -$9,517 (3090 losses, 1368 wins, 37% win rate)
    • Verdict: SCAM/engagement bait

  3. @krajekis BTC 15-min LONG — "+700% monthly, 1 trade/day at 9AM"

    • Backtested 25 days: 52% win rate (coin flip), strategy loses 76% of capital
    • Verdict: FABRICATED results

  4. @noisyb0y1 15-min arb — "$99K in a day"

    • Wallet 0xE594336603F4...: Real strategy (buying both sides), actual PnL ~$9K not $99K
    • Combined costs $0.70-$0.95 on some markets = genuine arb edge
    • Verdict: INFLATED numbers, but strategy has merit → bookmarked for scanner

  5. 5 more wallets — spawned sub-agent to research @IH2P, Super Bowl trader, NegRisk arb, Elon insider, $270→$244K bot

Pattern Identified

  • Fintwit Polymarket accounts follow identical template: big $ claim → wallet → product shill
  • Products being shilled: Clawdbots, Moltbook, Bullpen (affiliate/paid promos)
  • Real money is in selling engagement, not trading
  • Wallets are cherry-picked; PnL claims conflate position value with profit

Built Today

  • Crypto signals pipeline (projects/crypto-signals/)

    • Signal parser for Telegram JSON exports
    • Price fetcher (Binance US API)
    • Backtester with Polymarket fee awareness
    • Ready for D J's Telegram group export

  • Polymarket 15-min arb scanner — systemd timer every 2 min

    • Scans active Up or Down markets
    • Checks orderbooks for combined < $1.00
    • Paper trades and Telegram alerts
    • Finding: markets are tight at steady state, arb windows likely during volatility

  • Playwright installed — replaced flaky CDP websocket approach

    • connect_over_cdp('http://localhost:9222') works with existing Chrome session
    • X feed scraper working via Playwright

  • PowerInfer multi-GPU cron — weekly Monday check for feature support

Day-of-Week Stock Strategy

  • D J's friend suggested buy Friday/Monday, sell midweek
  • Backtested 5 years of SPY: Monday is strongest day (+0.114%, 62% win rate)
  • Buy Mon Open → Sell Wed Close: 56% win rate, +0.265% avg, $10K→$17.8K
  • But buy & hold still wins: $10K→$19K
  • Verdict: weak edge, needs additional filters to be useful

Infrastructure

  • Polymarket fee structure documented: 15-min markets have taker fees, max 1.56% at 50/50
  • Fee formula: shares × price × 0.25 × (price × (1-price))²
  • Binance international is geo-blocked; Binance US works

Pending

  • D J sending Telegram crypto signal group export (Option 2: JSON export)
  • Signal provider uses VWAP-based strategy
  • Sub-agent researching 5 more wallets

Evening Session — KIPP + Leverage Trading

Crypto Leverage Trading

  • Built short signal scanner (scripts/short_scanner.py) — RSI/VWAP/MACD/Bollinger scoring for 29 coins
  • Built leverage game engine (leverage_game.py) — $10K, 20x max leverage, longs/shorts/liquidations
  • Built auto-trader (scripts/leverage_trader.py) — connects scanners to game, TP/SL/trailing stops
  • "Leverage Challenge" game initialized, player "case"
  • Systemd timer every 15min
  • First scan: market quiet, no strong signals (highest short score 40 on FIL)

KIPP Setup

  • KIPP VM: 192.168.86.100 (Proxmox CT, 8 cores, 8GB RAM, Ubuntu 24.04)
  • Installed Node.js 22, OpenClaw 2026.2.9
  • Set up SSH key auth from Case's VM
  • Created KIPP identity: SOUL.md, AGENTS.md, MEMORY.md, USER.md
  • Initially tried GLM-4.7 Flash via llama.cpp — had issues, switched to Claude
  • Configured Telegram bot: @dzclaw_kipp_bot (token: 8535439246:AAE...)
  • Webchat accessible via https://kipp.host.letsgetnashty.com/
  • Gateway token: kipp-local-token-2026
  • Proxmox firewall was blocking LAN access — D J disabled it
  • Needed trustedProxies, allowedOrigins, dangerouslyDisableDeviceAuth for reverse proxy
  • D J ran openclaw configure to set up Anthropic auth
  • KIPP is LIVE on Claude + Telegram
  • Name confirmed: KIPP (not Kip)

KIPP Local LLM Attempt (Late Session)

  • Goal: Switch KIPP from Claude to local GLM-4 Flash 30B (q4) on llama.cpp at 192.168.86.40:8080/v1
  • llama.cpp server confirmed running and responding
  • Multiple openclaw.json config attempts failed — schema validation errors
  • Key schema findings for openclaw.json:
    • gateway.bind: must be enum (auto/lan/loopback/custom/tailnet)
    • agents.defaults.model: must be { primary: string, fallbacks?: string[] }
    • models.providers: array of { baseUrl, apiKey, api, models[] } for custom endpoints
    • auth.profiles: only allows provider/mode/email — NOT baseURL/apiKey
  • Gateway service installed but crashing on config validation
  • Status: BLOCKED — need to apply correct config schema, was mid-fix when session compacted

Infrastructure

  • SSH keys generated on Case's VM, copied to KIPP VM
  • loginctl enable-linger set for boot persistence
  • sshpass installed on Case's VM
  • llama.cpp server: 192.168.86.40:8080/v1 (GLM-4 Flash 30B q4)