Track 2 · Agentic Trading · Market Sentiment Agent
ARGUS researches; its sibling project trades. Qwen decides, a risk kernel that can only reduce stands between it and the venue, and Bitget's Agent Hub places every order on Bitget Demo. Everything below is read live from the agent's own published record, which its publisher rebuilds hourly from the hash-chained log — full record · source. This is run 2, the scored run; run 1, which came before it, is disclosed and not scored — its record.
Published record is 712 hours old: the hourly publish has not run since. The agent's own log keeps running, so the figures below lag it.
| Metric | Agent | A no-edge book over the same window |
|---|---|---|
| Sharpe (annualised) | -0.30 | median -0.5, p05 -12.3, p95 +11.1 |
| Max drawdown | -0.37% | median -0.45%, p05 -0.95%, worst -1.67% |
| Win rate | 29% | median 45.0%, p05 28.0%, p95 62.0% |
| Closed trades | 7 | median 23 |
| Return | -0.02% | median -0.04%, p05 -0.76%, p95 +0.94% |
| Fees paid (USDT) | 9.22 | — |
On 7 closed trades these figures say little yet: the next win takes the win rate to 37.5% and the next loss to 25.0%, from 28.6% now. The Sharpe is read from the whole book's equity, open positions included; the win rate from closed trades alone.
The right-hand column was committed in the genesis event before the first decision: what a book with no edge produces over the same window (validation/run2/strategy_replay.json (no-edge coin-flip book: five names a day at half a target, 50 seeds on each of 51 rolling 216 h windows of Bitget history, 2026-07-30 to 2026-09-26, ruled, sized and filled by the production kernel, planner and book)). The agent is judged against it, not against zero.
The venue refused 74 of the 95 orders sent: 72 protective exits on METAUSDT, 2026-09-28T09:16Z to 15:11Z — “Parameter METAUSDT_UMCBL does not exist”; 2 closes on MSTRUSDT, 2026-09-28T13:27Z to 13:38Z — “Parameter MSTRRWAUSDT_UMCBL does not exist”. Each name had a later order filled, so the refusals were the venue's state at the time rather than an order it could never take. The agent has no back-off on a refused exit: it sent the same one 72 times. Recorded as a known defect and not changed mid-run, so the scored record stays the one that ran.
Stay flat: coordinated social clusters are promotional, and the only extreme funding fact, AAPLUSDT.funding_z_live = 2.00272, lacks a second independent positioning confirmation. No thesis clears fees with confidence above even odds.
No instrument shows two independent positioning measures pointing the same way: the only funding z beyond reference.funding_z_extreme is AAPLUSDT.funding_z_live = 2.00272, while AAPLUSDT.retail_long_short_ratio was not measured.; The fired coordinated clusters for COINUSDT, CRCLUSDT, HOODUSDT, MSTRUSDT and GOOGLUSDT are repetitive promotional posts, not information, and crowd text alone cannot size a position.; BTCUSDT positioning is not overstretched: BTCUSDT.funding_z_live = -1.19114, BTCUSDT.retail_long_short_ratio = 0.9654 and BTCUSDT.oi_change_24h_pct = -2.52518 argue against acting on mood.crypto_fear_greed = 73 alone.; Venue risk is elevated for several US-session names: SP500USDT.demo_live_gap_bps = 55.5598 against SP500USDT.demo_live_gap_limit_bps = 266.5, NDX100USDT.demo_live_gap_bps = 60.3545 against NDX100USDT.demo_live_gap_limit_bps = 63.2, and METAUSDT.demo_live_gap_bps = 62.1445 against METAUSDT.demo_live_gap_limit_bps = 67.2.; With book.fees_today_bps = 0 and no confirmed edge, paying taker fees twice on a one-measure thesis would not be justified.
Stay flat: the wake-up clusters are coordinated promotion, not confirmed positioning, and no name shows two independent positioning measures beyond references. Venue gaps, especially METAUSDT.demo_live_gap_bps = 68.1778 against METAUSDT.demo_live_gap_limit_bps = 67.2, further block new exposure.
All coordinated clusters are promotion: story.1, story.2, story.3 and story.4 show repeated near-identical posts across distinct sources, including WhatsApp recruitment and explicit signal calls; a coordinated story is not a tradable thesis by itself.; No instrument has two independent positioning measures pointing the same way beyond references; funding z-scores are all below reference.funding_z_extreme = 2 in magnitude, and the only retail long/short fact shown is BTCUSDT.retail_long_short_ratio = 0.9897, near reference.long_short_parity = 1.; Several triggered names have Demo venue risks near or beyond acceptable limits: METAUSDT.demo_live_gap_bps = 68.1778 exceeds METAUSDT.demo_live_gap_limit_bps = 67.2, so new exposure is barred; AAPLUSDT.demo_live_gap_bps = 78.1156 is close to AAPLUSDT.demo_live_gap_limit_bps = 82.3; NVDAUSDT.demo_live_gap_bps = 71.7226 and AMZNUSDT.demo_live_gap_bps = 72.3917 are also elevated.; BTCUSDT woke on a funding settlement but shows no crowded positioning: BTCUSDT.funding_z_live = -0.23489, BTCUSDT.funding_rate_live = 0.000055 and BTCUSDT.retail_long_short_ratio = 0.9897 give no fade or momentum edge after two taker fees.; Data coverage is hollow for several sentiment feeds, and one item was withheld as a prompt injection; absent facts are unknown, not neutral, so they cannot support opening risk.
Flat through the weekend freeze: BTCUSDT funding is crowded short while account ratios are long-skewed, and no US-session instrument can be opened until session.hours_to_us_reopen = 7.97075.
BTCUSDT 0.0 (proposed 0.00%, approved 0.00%)
Flat: US-session names are frozen until 2026-10-05T00:00:00Z, and BTCUSDT's funding, long-short and taker facts show only mild, unconfirmed long skew rather than a crowded trade worth paying taker fees to fade or follow.
US-session instruments are frozen until 2026-10-05T00:00:00Z, with session.hours_to_us_reopen = 15.8439, so they cannot be opened.; BTCUSDT shows only mild long skew: BTCUSDT.funding_z_live = 1.16208 is below reference.funding_z_extreme = 2, BTCUSDT.retail_long_short_ratio = 1.2252 and BTCUSDT.top_trader_long_short_ratio = 1.8275 are above reference.long_short_parity = 1 but not extreme, and BTCUSDT.taker_buy_sell_ratio = 0.9716 does not confirm aggressive buying.; BTCUSDT crowd text is thin and uncoordinated: BTCUSDT.social_mentions_24h = 24, BTCUSDT.social_velocity_per_hour = 1.28316, crowd.coordinated_clusters = 0, and the shown stories are mostly unrelated retail chatter rather than a coordinated BTC story.; Fee and venue constraints argue against marginal trades: market taker fees must be paid twice, book.fee_budget_left_today_bps = 7, and a thesis without two independent confirming positioning measures is not confidently above even odds after fees.
Flat. BTCUSDT woke us for a funding settlement but funding z is 0.0194581, far from the 2 extreme reference, retail skew is mild and taker flow is seller-heavy, while all US-session names remain weekend-frozen.
BTCUSDT funding and positioning are not extreme: BTCUSDT.funding_z_live = 0.0194581 versus reference.funding_z_extreme = 2, with only modest long skew in BTCUSDT.retail_long_short_ratio = 1.2188 and BTCUSDT.top_trader_long_short_ratio = 1.8515.; BTCUSDT flow is not confirming crowded longs: BTCUSDT.taker_buy_sell_ratio = 0.7478 is below reference.long_short_parity = 1, and BTCUSDT.oi_change_24h_pct = -0.26107 shows open interest slightly down over 24h.; Crowd text around BTCUSDT is promotional or non-informational rather than a coordinated informational story: story.3 is a signals-group post, story.5 is a scam question, and story.8 is a Coinbase referral post; crowd.coordinated_clusters = 0.; All US-session instruments are frozen for the weekend, with session.hours_to_us_reopen = 23.8547, so no US equity or index exposure can be opened now.; Venue-integrity facts argue against preparing US exposure at this moment: GOOGLUSDT.demo_index_move_bps_3h = 0, METAUSDT.demo_index_move_bps_3h = 0, and MSTRUSDT.demo_live_gap_bps = 202.385.