Hermes

{"actions": [{"action_type": "web_research", "expected_output": "BTC.D phase classification, sector-bucket mapping for 28 Hermes symbols, DOT position sector-alignment check, and a proposed meta-allocation flag (overweight/underweight per sector) for future Hermes trading context.", "hypothesis": "BTC.D (Bitcoin dominance) is a leading indicator for which crypto sector phase is active (BTC → ETH → L1s → DeFi → Gaming → Memes). Hermes trades 28 symbols across L1, DeFi, L2, meme, and infrastructure sectors with zero sector-awareness. A BTC.D trend check could filter which sectors to overweight vs avoid, creating a genuinely new edge class orthogonal to GRU and non-GRU signals.", "risk": "Exploratory only — sector rotation is a slow signal with weekly/monthly frequency. Will not replace existing GRU or non-GRU entry logic. Could introduce a false sense of precision if BTC.D is noisy (sawtooth pattern in 50/50 markets). Kept isolated in experiments_hermes/ until validated against Hermes trade history.", "title": "Sector Rotation via BTC Dominance Tracking: Meta-Positioning Across Sectors", "workspace_path": "experiments_hermes/205-RESEARCH_btc_dominance_sector_rotation"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_btc_dominance_sector_rotation"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Create experiments_hermes/205-RESEARCH_btc_dominance_sector_rotation per the queued research_actions item above.", "title": "BTC Dominance Sector Rotation as Meta-Allocation Signal", "why_needed": "Hermes trades 28 symbols across multiple sectors but has zero sector-awareness. BTC.D is the primary 'altseason clock' and could provide a low-frequency meta-allocation overlay that filters which sectors to overweight vs avoid — orthogonal to all existing GRU/non-GRU signals."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "BTC.D 30-day history poll script (CoinMarketCap or CCXT), sector classification map for all 28 Hermes symbols (L1, DeFi, L2, infra, meme, payment, oracle, gaming), current phase determination (BTC season vs altcoin season vs neutral), and alignment check on current open DOT/USDT position.", "hypothesis": "BTC.D trend (above/below 50-day MA, rising/falling) predicts which sector of the 28-symbol Hermes universe will outperform. When BTC.D is falling, altcoin sectors (L1s, DeFi, L2s, memes, gaming) outperform BTC-correlated names. When BTC.D is rising, rotate to BTC-correlated infrastructure names. Hermes currently allocates without any sector-awareness, creating measurable opportunity cost.", "risk": "BTC.D is a macro indicator with slow cadence — may produce long periods of 'neutral' signal. Sector mapping is subjective (some symbols span multiple sectors). A single BTC.D poll doesn't create a tradeable signal by itself; it feeds into decision reasoning. Strictly paper-only queue item until validated against Hermes trade history.", "title": "Sector Rotation via BTC Dominance Tracking: Meta-Positioning Across Sectors", "workspace_path": "experiments_hermes/205-RESEARCH_btc_dominance_sector_rotation"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_btc_dominance_sector_rotation"], "needs": [], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "IDEA.md, RESULTS.md documenting Gate.io liquidation endpoint availability, 7-day sample collection, and 4h forward return analysis after >$1M long liquidation events", "hypothesis": "Large liquidation cascades ($1M+ per symbol per 4h) create measurable overshoot/reversal opportunities. When long liquidations cluster at support, forced selling flushes weak hands and creates a local bottom with positive forward EV at 4-8h horizons.", "risk": "This is entirely exploratory with no production code changes. Stays isolated in experiments_hermes/206-RESEARCH_liquidation_cascade_contrarian until validated.", "title": "Liquidation Cascade Clustering as Contrarian Entry Signal", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_contrarian"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_contrarian"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "1) Gate.io liquidation endpoint availability confirmed for 28-symbol universe. 2) 7-day baseline: per-symbol listing of >$1M cascade events with 4h/8h forward returns. 3) DOT/USDT entry at $0.79 checked for liquidation clustering. 4) Binary recommendation: is cascade_score >= threshold a deployable opportunity_scan field?", "hypothesis": "Heavy long liquidations (>$1M per symbol in 4h window) create measurable local bottoms via forced-weak-hand flush. These cascade events can be used as contrarian entry signals independent of GRU.", "risk": "Gate.io liquidation data may not have sufficient history for statistical validation. Small sample size for rare cascade events. DOT entry already made — the finding would be observational, not actionable for that position.", "title": "Liquidation Cascade Clustering as Contrarian Entry Signal — Gate.io data availability & 7-day baseline scan", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_contrarian"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_contrarian"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Gate.io liquidation data availability confirmed, 7-day liquidation event dataset for top-10 symbols, 4h forward return comparison table showing whether liquidation clusters have positive EV", "hypothesis": "Symbols with heavy long liquidation volume (>$1M within 4h) near support levels produce positive 4h forward returns as weak hands are flushed and mean reversion occurs. Gate.io per-asset liquidation data can confirm this pattern.", "risk": "Data may not be available per-symbol through standard CCXT. Liquidation data may be aggregated by exchange vs per-symbol. Gate.io liquidation page shows aggregate data; per-symbol endpoint availability is unconfirmed. If unavailable, pivot to per-exchange aggregate data.", "title": "Liquidation Cascade Clustering as Contrarian Entry Signal", "workspace_path": "experiments_hermes/205-RESEARCH_liquidation_cascade_entry"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_liquidation_cascade_entry"], "needs": [{"need_type": "data", "priority": "medium", "suggested_next_step": "In experiment 205, first step: verify CCXT can fetch per-symbol liquidation data from Gate.io for DOT, BTC, ETH, and top-5 universe symbols.", "title": "Gate.io per-symbol liquidation data availability", "why_needed": "Liquidation cascade entry signal research depends on per-symbol liquidation volume data from Gate.io. If only aggregate data exists, the signal hypothesis needs to be reformulated."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "1) Confirmed liquidation endpoint availability per symbol. 2) 7-day liquidation event catalog for top-10 Hermes universe symbols. 3) Forward return comparison after $1M+ events vs baseline. 4) DOT/USDT entry ($0.79) checked for liquidation clustering preceding entry.", "hypothesis": "Gate.io provides per-asset liquidation data for Hermes universe symbols. Liquidation events >$1M within 4h create measurable overshoot that can predict forward mean-reversion entries.", "risk": "Stays in experiments_hermes/ until validated. No impact on live paper decisions. Gate.io liquidation data quality/coverage needs verification first.", "title": "Gate.io Liquidation Data Availability Scan", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_clustering"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_clustering"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "BTC.D 30-day chart, trend direction (rising/falling/flat), current altcoin season index score, sector allocation recommendation for Hermes 28-symbol universe.", "hypothesis": "BTC.D trend direction determines which sector to overweight. Current phase classification will inform whether DOT L1 position is well-allocated.", "risk": "Pure observation, no trading decisions. Risk: BTC.D data source must be free and reliable.", "title": "BTC Dominance Trend Snapshot for Sector Rotation Phase", "workspace_path": "experiments_hermes/205-RESEARCH_btc_dominance_sector_rotation"}, {"action_type": "code", "expected_output": "DOT OBI score at top-10 depth levels. If OBI > +0.30 (bid-heavy), confirms thesis. If OBI < -0.30 (ask-heavy), warns of sell pressure.", "hypothesis": "DOT/USDT order book at current levels shows whether bid or ask side dominates, providing microstructure confirmation for the open long position.", "risk": "One-shot CCXT call, zero risk.", "title": "DOT Order Book Imbalance Probe", "workspace_path": "experiments_hermes/206-RESEARCH_dot_obi_probe"}, {"action_type": "experiment", "expected_output": "Per-symbol funding rate snapshot, percentile distribution, extreme flags (>0.08% or <-0.05%), comparison to CoinSwitch heuristics.", "hypothesis": "Extreme funding rates (>0.08% or <-0.05% per 8h) on any of 28 universe symbols create contrarian entry opportunities that Hermes currently ignores.", "risk": "Data quality: Gate.io may not expose funding rates for all 28 symbols. Fallback: poll BTC and top-5 universe symbols only.", "title": "Funding Rate Contrarian Fade Data Scan (Gateway to topic:signal_backlog execution)", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_btc_dominance_sector_rotation", "experiments_hermes/206-RESEARCH_dot_obi_probe", "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade", "paper-trading/hermes-agent/ (context enrichment pipeline)", "paper-trading/hermes-agent/bin/ (enhance diagnostics or screening)"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open"}, {"need_topic": "sector_rotation", "need_type": "signal", "priority": "high", "status": "open"}, {"need_topic": "tooling_backlog", "need_type": "tooling", "priority": "medium", "status": "open"}], "proposals": [{"artifact_path": "paper-trading/hermes-agent/ (context enrichment pipeline)", "next_step": "Codex/human: Add fetch_funding_rate call to context builder, write btc_funding_rate and btc_funding_signal (bullish/bearish/neutral) fields.", "objective": "Provide live macro positioning flag for every trading decision. BTC funding rate is the single most reliable crowd-positioning read in crypto, and Hermes currently has zero funding-rate awareness.", "paper_slot": "Add to context enrichment pipeline: single line `btc_funding_rate = exchange.fetch_funding_rate('BTC/USDT:USDT')` in build_trading_context.py. No new slot needed.", "risk": "Gate.io may not expose funding rate via CCXT's fetch_funding_rate. Fallback: poll via REST endpoint directly or coingecko API. Zero trading risk — field is advisory only.", "rollback": "Revert the context enrichment addition.", "title": "Add BTC Funding Rate Field to Trading Context"}, {"artifact_path": "paper-trading/hermes-agent/bin/ (enhance diagnostics or screening)", "next_step": "Codex/human: Test Gate.io CCXT liquidation endpoint availability for Hermes symbols, then add 4h liquidation volume fields to per-symbol context.", "objective": "When market breadth shows risk_off (19/28 risk_off per trader_latest), add liquidation-volume check as a secondary filter: prefer symbols with elevated short liquidations as mean-reversion long candidates and elevated long liquidations as continuation short candidates.", "paper_slot": "Add to opportunity_scan in build_trading_context.py: per-symbol long/short liquidation volume over 4h window, flag extremes (>2σ from 14-day mean).", "risk": "Gate.io liquidation data may have rate limits or be capped. Fallback: use BTC liquidation as proxy for macro liquidation stress. No trading risk — field is advisory only.", "rollback": "Remove liquidation fields from context builder.", "title": "Short-Candidate Liquidation-Volume Filter for Non-GRU Universe"}]}

{"actions": [{"action_type": "data_scan", "expected_output": "Funding rates and OI snapshots for 28 universe symbols from Gate.io, compared to 0.03/0.05/0.10 bps thresholds with cascade_score = abs(OI_change_pct) * abs(funding_rate * 100). Log symbols exceeding cascade_score > 2.0.", "hypothesis": "After one-sided mass liquidations, markets bounce because (a) temporary > permanent price impact, (b) fire-sale discounts create mean reversion force, (c) Hawkes branching ratio n=0.6-7.0 means cascades self-excite for hours before decay. Tracking cumulative liquidation volume per symbol provides a regime-override signal — probability of reversal within 4-12h increases significantly post-cascade.", "risk": "Data scan is a point-in-time snapshot. Cascade detection requires 24h+ history. This scan only flags current cascade risk, not historical validation. No trading decisions will be based on this data alone — it queues the research pipeline, not a strategy.", "title": "Liquidation Cascade Bounce — Gate.io OI + Funding Rate One-Shot Scan", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_regime_override"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_regime_override"], "needs": [{"need_topic": "funding_rate_contrarian_fade", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "GRU has been dead for 42+ days. Funding rates + OI data are the highest-ROI non-GRU signal class available: freely pollable via CCXT, documented contrarian edge at extremes, Hawkes cascade framework provides concrete decay parameters, and integrates naturally into trading_context.json as per-symbol cascade_risk + funding_rate fields."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "JSON dump of funding rates for all 28 universe symbols from Gate.io. Each entry includes symbol, funding rate, timestamp, and next funding time. Compared against 0.03/0.05/0.10 bps thresholds and OI change data for cascade_score computation.", "hypothesis": "A simplified Hawkes self-excitation cascade score can be computed per symbol from Gate.io OI change and funding rate extremes. When OI drops >20% in 4h and funding rate exceeds |0.1%| (per 8h), cascade_risk=high. This signal can be integrated into trading_context.json as a per-symbol cascade_risk flag that disables new entries for 4-12h post-event.", "risk": "Pure data scan — no trading decisions. Risk is limited to API rate limits (CCXT retry built-in). No state modified.", "title": "Gate.io OI and Funding Rate One-Shot Scan for Cascade Detection Feasibility", "workspace_path": "experiments_hermes/207-RESEARCH_liquidation_cascade_hawkes_bounce"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/207-RESEARCH_liquidation_cascade_hawkes_bounce"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot scan: 7-day sector group returns for L1s (SOL, AVAX, NEAR, APT, DOT, INJ, ATOM, SUI), DeFi (AAVE, UNI, CRV, RUNE), L2s (ARB, OP, POL), memes (DOGE, PEPE), infrastructure (FIL, LTC, HBAR, TRX, XRP, ALGO, XTZ, FET). Compares L1 vs DeFi vs Meme returns. If DeFi outperformed L1s by >10%, tests whether that predicts mean-reversion within next 7 days. Output written to RESULTS.md.", "hypothesis": "Crypto capital rotates through predictable phases (BTC → L1s → DeFi → Gaming/AI → Memes → BTC). Tracking 7-day sector group returns from existing trading_context.json price data can detect rotation phases and inform sector-level position bias.", "risk": "Research-only — no live trading impact. Sector definitions are ad-hoc (28-symbol universe is not exhaustive). The EarnifyHub rotation sequence may not reproduce on Gate.io universe.", "title": "Sector Rotation Detection — DEFI/L1 Relative Strength One-Shot Scan", "workspace_path": "experiments_hermes/206-RESEARCH_sector_rotation_defi_l1"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_sector_rotation_defi_l1"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "7-day sector group return table for L1, DeFi, L2, Meme, Infrastructure groups; DEFI/L1 ratio timeseries; rotation phase flags per Hermes 28-symbol universe", "hypothesis": "Crypto capital rotates through predictable phases: BTC dominance peaks first, then L1s, then DeFi/revenue protocols, then Gaming/AI, then memes, then back to BTC. Tracking DEFI_INDEX / L1_INDEX ratio via existing price data flags rotation phases and informs sector-level position bias.", "risk": "No live sector rotation signal yet; this is a data-scan-only queue item. Cannot affect decisions until validated. Sector boundaries are approximate (some tokens straddle categories).", "title": "Sector Rotation Detection via DEFI/L1 Relative Strength Index", "workspace_path": "experiments_hermes/207-RESEARCH_sector_rotation_detection"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/207-RESEARCH_sector_rotation_detection"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "experiments_hermes/206-RESEARCH_liquidation_cascade_bounce/RESULTS.md with per-symbol OI change, funding rate, cascade_score, and recommendation for which symbols need cascade_risk flag integration.", "hypothesis": "Per-symbol cascade_score = abs(OI_change_24h_pct) * abs(funding_rate * 100). If any symbol scores > 2.0, cascade_risk flag should be raised. At extremes, cascade bounce edge produces 60-70% WR 4-12h post-event.", "risk": "Gate.io OI data may not be available for all 28 symbols (only perpetual futures have OI). One-shot scan may miss cascade events outside the sample window. This is isolated as a data_scan — no trading changes until validated.", "title": "Gate.io OI + Funding Rate Cascade Scan — One-Shot for Top 10 Hermes Symbols", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_bounce"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_bounce"], "needs": [{"need_type": "data", "priority": "high", "suggested_next_step": "During the one-shot scan, log which symbols have available OI and funding rate endpoints. Gate.io perpetual futures (FIL-PERP, LTC-PERP, etc.) should expose fundingRate. OI may require fetchOpenInterest endpoint.", "title": "Gate.io OI + Funding Rate Data Availability for Cascade Detection", "why_needed": "Detecting liquidation cascades requires per-symbol OI change and funding rate data. If Gate.io does not expose OI for thin alt symbols, cascade detection is limited to BTC/ETH/SOL only. Must verify before developing cascade_risk integration."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "RESULTS.md in experiments_hermes/206-RESEARCH_sector_rotation_detection/ showing: (1) current 7-day sector returns, (2) rotation phase flag, (3) whether DeFi > L1 by >10% predicts mean-reversion in next 7 days, (4) integration recommendation for trading_context.json", "hypothesis": "Crypto capital rotates through predictable phases (BTC dominance → L1s → DeFi → Gaming/AI → Memes → back to BTC). Computing sector-group relative strength from existing price data will reveal rotation phases and inform sector-level position bias.", "risk": "Low — pure analysis of existing data, no trading actions. Sector labels are heuristic and may not perfectly capture token narratives. However, the cost of false rotation signals is limited (advisory flag only).", "title": "Sector Rotation Detection — 7-Day Sector Group Returns from Trading Context", "workspace_path": "experiments_hermes/206-RESEARCH_sector_rotation_detection"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_sector_rotation_detection"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "experiments_hermes/205-RESEARCH_liquidation_cascade_regime_override/RESULTS.md with per-symbol cascade_score table, detection rate, and comparison to 24h OI/funding baselines", "hypothesis": "When Gate.io OI drops >20% in 4h AND funding rate exceeds ±0.1%, a cascade liquidation is in progress. Post-cascade, mean-reversion bounce probability within 4-12h increases significantly (Hawkes branching ratio n=0.6-7.0). Hermes should flag cascade_risk=high and disable new entries until cascade decays.", "risk": "Gate.io OI data may not be deep or reliable enough for low-volume alts (FIL, AAVE). OI changes may also reflect non-cascade events (funding rate settlement, whale repositioning). Pure research scan — no trading impact. Requires next tick to populate the experiment directory.", "title": "Gate.io Liquidation Cascade Signature Scan — OI Drop + Funding Rate Spike", "workspace_path": "experiments_hermes/205-RESEARCH_liquidation_cascade_regime_override"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_liquidation_cascade_regime_override"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "sector_rotation_results.md with 7-day rolling sector return comparisons (L1 vs DeFi vs Meme vs Infra vs L2), rotation phase timestamps, and correlation analysis between sector outperformance and subsequent 7-day mean-reversion", "hypothesis": "Crypto capital rotates through predictable phases: BTC dominance first, then L1s, then DeFi/revenue protocols, then memes. Tracking DEFI_INDEX / L1_INDEX ratio in trading_context.json would flag rotation phases and inform sector-level position bias. When DeFi outperforms L1s by >10% over 7 days, mean-reversion to L1s follows within the next 7 days.", "risk": "Sector labels are heuristic (some symbols like FIL/TRX don't fit neatly). 7-day window is arbitrary — may need sensitivity testing.", "title": "Sector Rotation Detection via DEFI/L1 Relative Strength Index", "workspace_path": "experiments_hermes/206-RESEARCH_sector_rotation_detection"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_sector_rotation_detection"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Funding rate table with per-symbol rate, percentile vs history, and matching PineScriptForge entry conditions", "hypothesis": "Gate.io funding rate distribution across Hermes universe will reveal which symbols have extreme (>0.1% or <-0.05%) funding that the contrarian fade strategy could target.", "risk": "CCXT rate limit on Gate.io may truncate scan; one-shot is isolated, no side effects", "title": "Funding Rate One-Shot Scan for 28 Universe Symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"}, {"action_type": "data_scan", "expected_output": "Per-symbol cascade_score table with flags (pass, borderline, absent); recommend threshold for cascade_risk flag", "hypothesis": "At least 2-3 of FIL, AAVE, LTC, HBAR, or FET will show cascade signatures (OI drop >20% OR funding rate > ±0.1%) in the past 24h, validating the cascade_risk flag concept.", "risk": "Gate.io may not expose OI for all symbols; fallback to funding rate only", "title": "OI Change + Funding Rate Cascade Risk Scan for Top 10 Symbols", "workspace_path": "experiments_hermes/206-RESEARCH_cascade_risk_flag"}, {"action_type": "code", "expected_output": "JSON file with sector group returns and rotation phase classification", "hypothesis": "Computing DEFI_INDEX / L1_INDEX ratio from existing price data will reveal rotation phases that Hermes can use for sector-level position bias.", "risk": "Sector group assignments are heuristic; no live trading impact", "title": "Sector Rotation Index Computation", "workspace_path": "experiments_hermes/207-RESEARCH_sector_rotation_index"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade", "experiments_hermes/206-RESEARCH_cascade_risk_flag", "experiments_hermes/207-RESEARCH_sector_rotation_index"], "needs": [{"need_type": "data_quality", "priority": "high"}, {"need_type": "diagnostic", "priority": "medium"}], "proposals": [{"artifact_path": "experiments_hermes/206-RESEARCH_cascade_risk_flag", "next_step": "After Exp 206 confirms thresholds, propose 1-line addition to context_enrichment.py and 1-field addition to trading_context schema.", "objective": "Provide Hermes with awareness of liquidation cascade states — disables entries on symbols experiencing cascades, enables bounce-anticipation for symbols post-cascade.", "paper_slot": "No new slot. Add a 'cascade_risk' computed field to context_enrichment.py analogous to 'anomaly' fields. Values: none, watch (OI+funding threshold borderline), active (both thresholds breached).", "risk": "False positives in illiquid altcoins where OI is naturally low. Use 24h rolling baseline, not absolute thresholds.", "rollback": "Remove the cascade_risk field from context_enrichment.py — 1 line revert.", "title": "Add cascade_risk field to trading_context.json per-symbol blocks"}, {"artifact_path": "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade", "next_step": "After Exp 205 confirms rate thresholds for Gate.io, propose adding funding_rate fetch to the trading context builder.", "objective": "Provide Hermes with per-symbol funding rate and percentile flag, enabling the contrarian fade strategy as a non-GRU signal layer.", "paper_slot": "No new slot. Add 'funding_rate' and 'funding_percentile' fields to existing symbol blocks in context_enrichment.py.", "risk": "Funding rates update every 8h; stale rates could mislead. Include timestamp and freshness indicator.", "rollback": "Remove the funding_rate/funding_percentile fields from context_enrichment.py.", "title": "Funding Rate Field in trading_context.json per-symbol blocks"}]}

{"actions": [{"action_type": "data_scan", "expected_output": "Sector assignment table for all 28 symbols; 90-day weekly sector momentum time series; correlation of top-quartile sector momentum vs individual symbol h=12 forward return; recommendation on whether to add sector_momentum advisory to trading_context.json", "hypothesis": "Assigning sector tags (L1, L2, DeFi, Meme, Storage, Oracle, Payment, Utility/Privacy) to each of 28 Hermes universe symbols and computing weekly sector-level avg_4h_ret and avg_RSI can inform which altcoins to overweight in non-GRU entry decisions. Top-quartile sector momentum predicts individual symbol outperformance at h=12.", "risk": "Sector tags are subjective (many tokens span multiple sectors). Sector rotation signals are well-known and may decay once deployed. This is research-only — no risk of capital loss until deployed as a trading_context field.", "title": "Sector Rotation Monitoring — tag sectors + compute sector momentum for all 28 Hermes symbols", "workspace_path": "experiments_hermes/207-RESEARCH_sector_rotation_monitoring"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/207-RESEARCH_sector_rotation_monitoring"], "needs": [], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "RESULTS.md with per-symbol cascade frequency, forward EV at h=4/8/12, WR, and recommendation on whether to deploy as a trading_context advisory signal or entry rule", "hypothesis": "Liquidation cascades create statistically reliable over-reactions on 15m bars. A 3-bar pattern (range >2x ATR, close <25th percentile, next 1-2 bars recover >50% of range) identifies cascade wicks with positive forward EV at h=4,8,12 for LONG entries, compared to non-cascade bars.", "risk": "OHLCV-only cascade detection may produce false positives (large range bars from news events that don't mean-revert). EV may be negative for some symbols. Still worth testing since it uses existing data with zero API cost.", "title": "Liquidation Cascade Recovery Capture — OHLCV-only detection on 28 Hermes symbols", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Per-symbol CSV with cascade bar count, mean forward return at h=4/8/12, WR for cascade entries vs non-cascade, and a recommendation on whether to add a cascade_flag to trading_context.json", "hypothesis": "Cascade-like bars (range > 2x ATR, close < 25th percentile, next 1-2 bars recover > 50% of cascade bar range) have positive forward EV at h=4,8,12 for LONG entries compared to non-cascade bars.", "risk": "Cascade patterns may be rare at 15m resolution even with elevated frequency. If < 20 samples per symbol, aggregate across all symbols for pooled analysis. Not deployed until validated against 90-day OHLCV.", "title": "Liquidation Cascade Recovery Capture — scan 90 days of OHLCV for cascade patterns across all 28 Hermes symbols", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "experiments_hermes/209-RESEARCH_liquidation_cascade_capture/RESULTS.md with per-symbol cascade frequency table, forward EV at each horizon, win-rate breakdown, and pipeline integration recommendation", "hypothesis": "Cascade-like bars (range > 2x ATR(14), close < 25th percentile of bar range, next 1-2 bars recover > 50% of cascade bar range) occur in Hermes' 28-symbol universe and produce positive forward EV at h=4,8,12 for LONG entries vs non-cascade bars. LiveVolatile data (Feb 2026) shows 70-85% of cascade drops recover within 2-6 hours for BTC, but altcoin recovery rates and optimal hold horizons may differ.", "risk": "Remains isolated in experiments_hermes until validated and deployed via formal proposal. Pattern detection thresholds (2x ATR, 25th percentile close, 50% recovery) are initial defaults and may require calibration. Altcoin cascades may behave differently from BTC — smaller sample sizes per symbol.", "title": "Liquidation Cascade Recovery Capture — OHLCV Pattern Scan for 28 Hermes Symbols", "workspace_path": "experiments_hermes/209-RESEARCH_liquidation_cascade_capture"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/209-RESEARCH_liquidation_cascade_capture"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "IDEA.md, RESULTS.md with per-symbol cascade count, forward WR at h=4/8/12, comparison vs non-cascade mean-reversion WR", "hypothesis": "Cascade-like bars (15m range >2x ATR, close <25th percentile, next 1-2 bars close above mid of cascade bar) have positive forward EV at h=4,8,12 for LONG entries, and this EV exceeds non-cascade mean-reversion baselines.", "risk": "Purely OHLCV pattern-based — no real liquidation data. Cascade detection is a proxy. May produce false positives. Isolated in experiments_hermes until validated.", "title": "Liquidation Cascade Recovery Capture — OHLCV pattern scan across 28 symbols", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "CSV of cascade bar events per symbol, forward returns at h=4,8,12, win rate, avg return, Sharpe. Comparison vs non-cascade control bars. RESULTS.md summary.", "hypothesis": "Liquidation cascade bars (3-bar pattern: range >2x ATR, close <25th percentile, next 1-2 bars recover >50% of range) produce positive forward EV at h=4,8,12 for LONG entries. If 70-85% of cascade events recover within 2-6h (per LiveVolatile), the detectable pattern should be profitable in Hermes' 15m universe.", "risk": "Risk 1: cascade detection may produce too few events (<10 per symbol) for statistical significance. Risk 2: GRU-dead period means no GRU signal to filter cascade entries — may need to pair with existing market breadth or VWAP band. Risk 3: isolated to 15m bars; cascade may resolve faster than 15m resolution captures.", "title": "OHLCV Cascade Bar Detection — 90-day scan across 28 Hermes symbols", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Per-symbol stats: cascade bar count, cascade WR at h=4/8/12, mean return, % of bars that cascade. Aggregate across symbols: pooled cascade WR, mean return, Sharpe. If positive EV, a deployment proposal to add cascade_flag to trading_context.", "hypothesis": "Cascade-like bars (3-bar pattern: range >2x ATR, close <25th percentile, next 1-2 bars recover >50% of range) have positive mean-reversion EV at h=4,8,12 compared to non-cascade bars. From LiveVolatile research: 70-85% of cascade drops recover within 2-6 hours.", "risk": "GRU-level signal quality not expected. Cascade detection is a lightweight pattern — may not reach deployable WR. Stays isolated in experiments_hermes/ until validated.", "title": "Liquidation Cascade Recovery Capture — 90-day OHLCV scan across 28 Hermes symbols", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery/RESULTS.md with: (1) cascade bar frequency per symbol over 90 days; (2) forward EV at h=2,4,8,12 for LONG entries on cascade bars vs non-cascade baseline; (3) best-performing symbols by cascade EV; (4) if positive, specification for adding a 'cascade_flag' and 'cascade_recovery_score' field to trading_context.", "hypothesis": "Liquidation cascade bars (range > 2x ATR, close near low, next 1-2 bars recover >50% of range) produce statistically reliable mean-reversion entries across the 28-symbol Hermes universe, with 60%+ WR at h=4-8.", "risk": "Pure OHLCV pattern detection cannot distinguish genuine liquidation cascades from normal volatility clusters. The 3-bar pattern (big range + close low + quick recovery) may capture non-cascade mean-reversion setups, which is still useful but dilutes the specific cascade hypothesis. Backtest overfitting on a small number of extreme events is the main risk — need minimum 20 cascade bars per symbol for statistical significance.", "title": "Liquidation Cascade Recovery Capture — OHLCV cascade bar detection and forward EV scan", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Per-symbol count of cascade bars, forward EV at h=4/8/12 vs non-cascade bars, win rate delta", "hypothesis": "Cascade-like bars (range >2x ATR, close at extreme, fast follow-through recovery) have positive forward EV as LONG entries at h=4,8,12.", "risk": "Pure OHLCV proxy for liquidations — may miss cascades that resolve outside bar boundaries. Stays isolated in experiments_hermes/, no runtime changes.", "title": "Liquidation Cascade Recovery — OHLCV scan on 90d Hermes universe", "workspace_path": "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery"}, {"action_type": "data_scan", "expected_output": "Per-sector momentum time series, rank correlation between sector momentum and individual symbol 12h forward return", "hypothesis": "Sector-level momentum (avg 4h return within sector) predicts individual symbol short-term outperformance.", "risk": "Sector labels are subjective; some symbols span multiple categories. Stays isolated — no runtime changes until validated.", "title": "Sector Rotation Tagging and Momentum Scan", "workspace_path": "experiments_hermes/207-RESEARCH_sector_rotation_momentum"}, {"action_type": "data_scan", "expected_output": "Per-symbol volatility regime count, past Hermes trade PnL distribution by regime (low/normal/high)", "hypothesis": "High-vol regime (ATR ratio >1.3) correlates with worse Hermes trade outcomes; position sizing should adapt.", "risk": "Simple ATR ratio is a coarse volatility proxy — may not capture structural regime shifts. Isolated research only.", "title": "Volatility Regime from OHLCV ATR Ratio", "workspace_path": "experiments_hermes/208-RESEARCH_volatility_regime_sizing"}, {"action_type": "data_scan", "expected_output": "Per-symbol current funding rate, distribution stats, extreme flags (>0.05%, negative)", "hypothesis": "Already documented in topic:signal_backlog. One-shot CCXT funding rate query for all 28 universe symbols from Gate.io.", "risk": "Stays isolated — no runtime changes. APY-calibrated, not bps-per-8h; may need unit conversion.", "title": "Funding Rate Contrarian Fade — First Data Scan (Exp 205 acceleration)", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_liquidation_cascade_recovery", "experiments_hermes/207-RESEARCH_sector_rotation_momentum", "experiments_hermes/208-RESEARCH_volatility_regime_sizing", "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "GRU dead 43+ days, funding rates are highest-ROI non-GRU signal class available. Need to execute Exp 205 scan to validate before deployment."}, {"need_topic": "liquidation_cascade_recovery", "need_type": "research", "priority": "medium", "status": "open", "why_needed": "If OHLCV proxy for liquidation cascades has positive EV, it becomes a new non-GRU signal class for mean-reversion entries — directly usable while GRU is dead."}, {"need_topic": "sector_rotation_monitoring", "need_type": "diagnostic", "priority": "medium", "status": "open", "why_needed": "A sector tag + sector momentum signal would improve non-GRU entry decisions by filtering which altcoin sectors have tailwinds."}, {"need_topic": "volatility_regime_sizing", "need_type": "diagnostic", "priority": "medium", "status": "open", "why_needed": "Lightweight ATR ratio volatility state could inform position sizing without new data sources. No Hermes trades have been analyzed for vol-regime PnL impact."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Sector mapping table, 7d sector return heatmap, historical ENTER decision bias analysis by sector, draft sector_rank field spec for symbol snippets", "hypothesis": "Hermes' 28-symbol universe naturally segments into 6+ sectors (L1, DeFi, AI, Storage, Meme, L2, RWA). Sector-leading symbols systematically outperform sector-lagging ones by >5% per rotation leg. A computed sector_rank field in symbol snippets would let Hermes bias capital toward in-flow sectors without introducing new data sources.", "risk": "Pure data analysis, no trading decisions. Stays in experiments_hermes/. Sector definitions may have overlap (e.g., POL is L1+RWA). Some sectors may have only 1-2 symbols, making ranking noisy.", "title": "Cross-Sector Relative Strength Rotation — Hermes 28-Symbol Universe Mapping and 7d Return Analysis", "workspace_path": "experiments_hermes/207-RESEARCH_sector_rotation_hermes_universe"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/207-RESEARCH_sector_rotation_hermes_universe"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "A sector mapping table for all 28 Hermes symbols, 7d sector return comparison, identification of currently-leading vs lagging sectors, and a proposed 'sector_rank' field format for trading_context.json symbol snippets.", "hypothesis": "Capital rotates between crypto sectors (DeFi, L1, Meme, AI, RWA, Storage, L2, Privacy) on multi-day to weekly timeframes. Sector-leading symbols outperform sector-lagging symbols by 5-10% per rotation leg. Mapping Hermes' 28 symbols to sectors and tracking 7d relative sector performance would enable a 'sector_rank' per-symbol field that identifies which cohort is currently rotating in.", "risk": "Stays isolated because sector rotation is a medium-frequency signal (multi-day), not a tick-by-tick entry trigger. If deployed without validation, it could bias Hermes toward sector-chasing. The experiment first validates whether the signal exists in the existing data before any deployment proposal.", "title": "Cross-Sector Relative Strength Rotation within 28-Symbol Universe", "workspace_path": "experiments_hermes/206-RESEARCH_sector_rotation_relative_strength"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_sector_rotation_relative_strength"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Sector map for all 28 Hermes symbols across 7-9 sectors. 7d sector cohort returns table with separation metrics. Comparison of top-sector vs bottom-sector forward performance in Hermes' decision window. Feasibility assessment for adding a 'sector_rank' field to symbol snippets in trading_context.json.", "hypothesis": "Capital rotates between crypto sectors (DeFi, L1, Meme, AI, RWA, Storage, Privacy) on multi-day to weekly timeframes. Symbols in the leading sector cohort outperform lagging sector cohorts by 5-10% per rotation leg. Adding a 'sector_rank' field to symbol snippets would improve Hermes' symbol selection by biasing toward in-flow sectors.", "risk": "Pure research, no trading changes. Sector labels are approximate (many tokens span multiple sectors/have ambiguous classification). Even if cohort separation exists, causal attribution to 'rotation' vs random walk is not proven by a single scan. Opportunities are multi-day not intra-tick, so the field would tilt directional bias rather than trigger instant entries.", "title": "Cross-Sector Relative Strength Rotation within 28-Symbol Universe", "workspace_path": "experiments_hermes/206-RESEARCH_cross_sector_rotation"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_cross_sector_rotation"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Per-symbol funding_rate snapshot with percentile flags vs 0.03/0.05/0.10 bps thresholds; positive vs negative counts; extreme values identified; price-momentum divergence scan", "hypothesis": "Extreme funding rates (>+0.05% or <-0.05% per 8h) are reliable contrarian signals independent of GRU. When funding is extreme positive and price weakens, it flags overleveraged longs about to liquidate. When funding is extreme negative, short-squeeze risk is elevated.", "risk": "Funding rate data is a snapshot, not a time series. No backtest to validate the contrarian fade hypothesis on Hermes-specific symbols. Exchange-specific differences (Gate.io vs Binance). Stays isolated until validated against Hermes paper trade timestamps.", "title": "Funding Rate Contrarian Fade as Non-GRU Signal Layer — queued for later validation", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"], "needs": [], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "Sector mapping table for 28 symbols, sector 7d return ranking from trading_context.json, backtest of sector-top-3 vs sector-bottom-3 forward return separation, sector_rank field spec for symbol snippets", "hypothesis": "Capital rotates between crypto sectors (L1, L2, DeFi, Meme, AI, Storage, RWA) on multi-day to weekly timeframes. Symbols in leading sectors outperform lagging sectors by 5-10% per rotation leg. Mapping the 28 Hermes symbols to sectors and tracking 7d relative performance would enable sector-aware capital allocation.", "risk": "Sector mapping is subjective (some symbols span multiple sectors). 7d return may be noisy. Correlated sector moves may not provide enough separation for a tradeable signal. This is isolated research in experiments_hermes/ — no effect on paper trading until validated.", "title": "Cross-Sector Relative Strength Rotation within 28-Symbol Universe", "workspace_path": "experiments_hermes/206-RESEARCH_sector_rotation_strength"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_sector_rotation_strength"], "needs": [{"need_type": "research", "priority": "high", "suggested_next_step": "Map 28 universe symbols to CoinGecko sectors, pull 7d returns from trading_context.json, compute sector cohort performance, check if sector-leading symbols outperform sector-lagging ones by >5% in forward window.", "title": "Cross-Sector Relative Strength Rotation", "why_needed": "Hermes symbol selection (identified as #1 PnL problem in Exp 173/177) lacks sector context. If sector-leading symbols systematically outperform sector-lagging ones, this is a simple additional signal layer that improves entry quality without new data sources."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Per-symbol funding rate snapshot with percentile flags (extreme_positive, moderate_positive, neutral, moderate_negative, extreme_negative) + summary of current market-level funding distribution", "hypothesis": "Extreme funding rates (>+0.05% or <-0.05% per 8h) are reliable contrarian signals. When funding is extreme positive and price weakens, it flags overleveraged longs about to liquidate. When funding is extreme negative, short-squeeze risk is elevated.", "risk": "Funding rates change every 8h on Gate.io; a single snapshot may not capture dynamic extremes. This is a discovery scan, not a full backtest. Kept isolated until real data confirms the signal exists at actionable magnitude in the current market.", "title": "Funding Rate Contrarian Fade as Non-GRU Signal Layer", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_contrarian_fade/ — one-shot ccxt query of funding rates for all 28 universe symbols from Gate.io. Compare to 0.03/0.05/0.10 bps thresholds. Log positive vs negative and extreme percentiles.", "title": "Funding Rate Contrarian Fade as Non-GRU Signal Layer", "why_needed": "GRU has been dead for 42+ days. Funding rates are the highest-ROI non-GRU signal class available: freely pollable via CCXT, documented contrarian edge at extremes, and integrate naturally into trading_context.json as a per-symbol funding_rate field with percentile flags."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot CCXT query output: per-symbol funding rate for all 28 universe symbols from Gate.io. Rates compared to 0.03/0.05/0.10 bps thresholds. Positive vs negative counts and extreme percentiles. Correlation with current price momentum and VWAP bands.", "hypothesis": "Extreme funding rates (>+0.05% or <-0.05% per 8h) are reliable contrarian signals independent of GRU. When funding is extreme positive and price weakens, it flags overleveraged longs about to liquidate. When funding is extreme negative, short-squeeze risk is elevated.", "risk": "Research-only. No trading decisions depend on this data scan. Funding rates are a single data point at a single timestamp — not a backtest. No deployment risk.", "title": "Funding Rate Contrarian Fade as Non-GRU Signal Layer", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ — one-shot ccxt query of funding rates for all 28 universe symbols from Gate.io. Compare to 0.03/0.05/0.10 bps thresholds. Log positive vs negative and extreme percentiles.", "title": "Funding Rate Contrarian Fade as Non-GRU Signal Layer", "why_needed": "GRU has been dead for 42+ days. Funding rates are the highest-ROI non-GRU signal class available: freely pollable via CCXT, documented contrarian edge at extremes, and integrate naturally into trading_context.json as a per-symbol funding_rate field with percentile flags."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot CCXT query of Gate.io perp funding rates for all 28 universe symbols. Log each symbol's current rate, compare to 0.03/0.05/0.10 bps thresholds, count positive vs negative, flag extreme percentiles. Compare to current price momentum to find divergences.", "hypothesis": "Extreme funding rates (>+0.05% or <-0.05% per 8h) are reliable contrarian signals independent of GRU. When funding is extreme positive and price weakens, it flags overleveraged longs about to liquidate. When funding is extreme negative, short-squeeze risk is elevated.", "risk": "Isolated research experiment — no live trading impact. Only observes current funding rates without executing any trades. Funding rate data is public and freely available via CCXT.", "title": "Funding Rate Diagnostics — One-Shot CCXT Poll of All 28 Universe Symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ — one-shot ccxt query of funding rates for all 28 universe symbols from Gate.io. Compare to 0.03/0.05/0.10 bps thresholds. Log positive vs negative and extreme percentiles.", "title": "Funding Rate Contrarian Fade as Non-GRU Signal Layer", "why_needed": "Funding rate flips are the highest-ROI non-GRU signal class available: freely pollable, documented contrarian edge at extremes, and integrate naturally into trading_context.json. GRU has been dead for 42+ days. This replaces one missing signal class with an orthogonal edge."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "JSON table: symbol → funding_rate, annualized, OI, percentile vs global distribution. Count of symbols above +0.05%, below -0.05%, neutral. Cross-reference with current Hermes positions.", "hypothesis": "Gate.io perpetual funding rates for Hermes' 28 symbols vary between positive and negative. Extreme rates (>+0.05% or <-0.05% per 8h) correlate with upcoming reversals. A one-shot poll will quantify how many symbols are at extremes.", "risk": "Gate.io may not support perps for all 28 symbols. Some altcoins may lack perpetual contracts. Need to handle missing data gracefully. Stays isolated in experiments_hermes/.", "title": "Funding Rate One-Shot Poll for All 28 Universe Symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}, {"action_type": "data_scan", "expected_output": "Sector map CSV, trailing 7d/30d sector returns, identification of currently-leading vs lagging sectors in the Hermes universe.", "hypothesis": "The 28 Hermes symbols cluster into 8-10 sectors. Sector-leading cohorts outperform sector-lagging ones by >5% over 7d windows. A sector_rank field in context would improve capital allocation decisions.", "risk": "Symbol-to-sector mapping is subjective. Some symbols (e.g., FET = AI? DeFAI? L1?) may have ambiguous sector classification. Stays in experiments_hermes/.", "title": "Sector Mapping of 28-Symbol Universe & 7d Relative Strength", "workspace_path": "experiments_hermes/206-RESEARCH_sector_rotation_mapping"}, {"action_type": "web_research", "expected_output": "Evidence digest with 3-5 quantifiable metrics (funding trajectory, depth balance, OI dominance, ETF flows) that can be added to Hermes' regime context as external cross-checks.", "hypothesis": "Coinbase's July 2026 Crypto Market Positioning report contains actionable evidence on BTC/ETH funding, order-book depth shifts, and altcoin OI dominance that can inform Hermes regime labels.", "risk": "Coinbase report may be behind login/paywall for full content. Treat as secondary evidence — primary data comes from CCXT polls. Stays in experiments_hermes/.", "title": "Coinbase July 2026 Positioning Deep-Read — Funding/Basis/Depth", "workspace_path": "experiments_hermes/207-RESEARCH_coinbase_positioning_july2026"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics", "experiments_hermes/206-RESEARCH_sector_rotation_mapping", "experiments_hermes/207-RESEARCH_coinbase_positioning_july2026"], "needs": [{"need_topic": "signal_backlog", "need_type": "tool", "priority": "high", "why_needed": "Funding rates are the highest-ROI non-GRU signal class — freely pollable, documented contrarian edge at extremes. This need has been open since Jul 13 (seen 30 iterations) with GRU dead 42+ days. Web evidence strongly supports immediate execution."}, {"need_topic": "signal_backlog", "need_type": "tool", "priority": "medium", "why_needed": "Liquidations are the most honest signal in crypto. A $100M cluster detection would give Hermes 2-4h post-event mean reversion edge. Currently zero liquidation awareness in the system."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "One-shot diagnostic report in experiments_hermes/205-RESEARCH_funding_rate_diagnostics/: min/max/median/deciles across all 28 symbols, symbols in top/bottom decile flagged, comparison to 0.03/0.05/0.10 bps thresholds. Recommendations for field schema to add to trading_context.json.", "hypothesis": "Persistently positive funding rates on altcoins signal crowded longs and predict short-term reversals. Extreme positive funding (>0.05%/8h) is a contrarian short signal; extreme negative funding is a contrarian long squeeze signal. This asymmetry directly fits Hermes' bilateral strategy.", "risk": "One-shot snapshot only — needs sequential polling to assess persistence. Funding extremes alone without volume/price confirmation may produce false signals. Kept isolated as RESEARCH until validated against forward returns.", "title": "Funding Rate Extremes as Contrarian Signal for Altcoin Reversals", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ — one-shot ccxt query of funding rates for all 28 universe symbols from Gate.io. Compare to 0.03/0.05/0.10 bps thresholds. Log positive vs negative and extreme percentiles.", "title": "Funding Rate Diagnostics — execute exp 205", "why_needed": "Funding rate flips are the highest-ROI non-GRU signal class available: freely pollable, documented contrarian edge at extremes, and integrate naturally into trading_context.json. Need has been open since Jul 13 (seen 29 iterations). GRU dead 40+ days. This research tick queues the execution."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot CCXT scan of funding rates for all 28 Hermes universe symbols from Gate.io. Report: min/max/median/deciles, symbols in top/bottom 10% decile, symbols exceeding 0.03%/0.05%/0.10% per 8h thresholds, positive vs negative split, and whether current Hermes HOLD pattern coincides with neutral funding or if extreme positions exist that the system is blind to.", "hypothesis": "Persistently positive funding rates on altcoins signal crowded longs and predict short-term reversals. Extreme positive funding (>0.05%/8h) is a contrarian short signal; extreme negative funding (<-0.05%/8h) signals crowded shorts and potential long squeeze. This fits Hermes' bilateral strategy and can be integrated as a new signal family into trading_context.json.", "risk": "Funding rates are a snapshot, not time-series. A single snapshot cannot distinguish between persistent extreme and transient spike. This is a diagnostic/evidence-gathering step only — no trading decisions will be based on a snapshot alone. The data_scan is research-only and will not be deployed into live context without human review.", "title": "Funding Rate Extremes as Contrarian Signal for Altcoin Reversals", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "Funding rate flips are the highest-ROI non-GRU signal class available: freely pollable, documented contrarian edge at extremes, and integrate naturally into trading_context.json. Need has been open since Jul 13 (seen 29 iterations). GRU dead 40+ days."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot CCXT query of Gate.io funding rates for all 28 universe symbols. CSV/JSON with per-symbol: funding_rate_8h, funding_annualized_pct, percentile vs cross-section. Flags for extreme deciles (>0.10% positive, <-0.05% negative). Compare to 0.03%/0.05%/0.10% thresholds. Written to experiments_hermes/205-RESEARCH_funding_rate_diagnostics/out/", "hypothesis": "Persistently positive funding rates on altcoins signal crowded longs and predict short-term reversals. Extreme positive funding is a contrarian short signal; extreme negative funding is a contrarian long signal. Funding extremes can replace the dead GRU pipeline for bilateral direction bias.", "risk": "Isolated data scan — no trading, no pipeline changes. Funding rates are point-in-time; single snapshot may miss intraday extremes. But even one snapshot establishes baseline distribution for threshold calibration.", "title": "Funding Rate Extremes as Contrarian Signal for Altcoin Reversals", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "JSON report in experiments_hermes/205-RESEARCH_funding_rate_diagnostics/out/funding_rate_snapshot.json with per-symbol funding rate, annualized %, decile rank, and extreme flags", "hypothesis": "Persistently positive funding rates on altcoins signal crowded longs and predict short-term reversals. Extreme positive funding (>0.05%/8h) is a contrarian short signal. Extreme negative funding (<-0.05%/8h) is a contrarian long signal. Asymmetric: funding extremes predict short-side moves more reliably than long-side.", "risk": "Pure data scan — no live trading, no deployment. Risk: zero. Isolated in experiments_hermes/.", "title": "Funding Rate Diagnostics — fetch live funding rates for all 28 Hermes universe symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ — one-shot ccxt query of funding rates for all 28 universe symbols from Gate.io. Compare to 0.03/0.05/0.10 bps thresholds. Log positive vs negative and extreme percentiles.", "title": "Funding Rate Diagnostics — execute exp 205", "why_needed": "Funding rate flips are the highest-ROI non-GRU signal class available: freely pollable, documented contrarian edge at extremes, and integrate naturally into trading_context.json. Need has been open since Jul 13 (seen 28 iterations). GRU dead 40+ days."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot CSV/JSON with funding_rate, funding_annualized_pct per symbol; min/max/median/deciles cross-section; symbols in top/bottom 10% decile flagged. Compare to 0.03%/0.05%/0.10% per 8h thresholds. Determine whether current flat/HOLD portfolio coincides with neutral funding or if extreme positions exist that Hermes is blind to.", "hypothesis": "Funding rate extremes predict short-term reversals. Persistent positive funding signals crowded longs (short signal), persistent negative funding signals crowded shorts (long signal). Gate.io funding rates are freely pollable and zero-cost. This is the highest-ROI non-GRU signal class available during the 40+ day GRU outage.", "risk": "Isolated one-shot data pull — no live trading, no pipeline changes, no code deployment. Funding rates are point-in-time snapshots and may differ from realized per-bar averages.", "title": "Funding Rate Diagnostics — one-shot CCXT poll of all 28 Hermes universe symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Write and execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/run.py — one-shot CCXT query of funding rates for all 28 universe symbols from Gate.io. Report min/max/median/deciles. Flag extreme deciles. Compare to 0.03/0.05/0.10 bps thresholds.", "title": "Funding Rate Integration into Trading Context — execute exp 205", "why_needed": "Funding rate flips are the highest-ROI non-GRU signal class available. Freely pollable via CCXT for all 28 Hermes symbols. GRU has been dead 40+ days. This is the longest-standing open need (topic:signal_backlog, open since Jul 13, 28 iterations). Need is to execute the queued data scan."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "OBI snapshot CSV with bid_volume_10, ask_volume_10, OBI_ratio per symbol. Correlation scan vs 15m/30m/1h forward returns. Identification of extreme OBI threshold zones with >60% predictive accuracy.", "hypothesis": "Extreme OBI values (< 0.35 or > 0.65) at depth_10 predict short-term price direction within 15m-1h, with asymmetric strength for negative moves. This is a non-GRU microstructure signal that works independently of the dead GRU pipeline.", "risk": "Standalone OBI snapshot without order book streaming has limited predictive power — point-in-time OBI may not capture sustained pressure. However, a one-shot correlation scan establishes baseline EV before investing in streaming infrastructure.", "title": "Order Book Imbalance Snapshot for Top-10 Hermes Symbols", "workspace_path": "experiments_hermes/205-RESEARCH_order_book_imbalance"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_order_book_imbalance"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "OBI values for top-10 Hermes symbols (BTC, ETH, LTC, FIL, HBAR, FET, AAVE, TRX, XRP, SOL) from Gate.io order book at depth 10+. Correlation with 15m-1h forward price moves. Extreme value thresholds and hit rates. Sector-level OBI patterns.", "hypothesis": "Order book imbalance (OBI) at depth 10+ levels has statistically significant predictive power for short-term crypto returns, with asymmetric strength — stronger for predicting negative price changes than positive ones. Extreme OBI values (< 0.35 or > 0.65) predict short-term mean reversion, and the short-side asymmetry adds value to Hermes' bilateral strategy.", "risk": "One-shot snapshot may not capture temporal dynamics. OBI is a microstructure signal requiring sustained imbalance to be predictive. Single snapshot OBI may be noisy. This is a data scan, not a trading strategy — risk is limited to wasted compute time.", "title": "Order Book Imbalance (OBI) One-Shot Scan for Top-10 Hermes Symbols", "workspace_path": "experiments_hermes/206-RESEARCH_order_book_imbalance_scan"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_order_book_imbalance_scan"], "needs": [{"need_type": "research", "priority": "high", "suggested_next_step": "Create experiments_hermes/206-RESEARCH_order_book_imbalance_scan/ with IDEA.md and run.py. Run one-shot OBI scan for top-10 Hermes symbols from Gate.io. Compare OBI values to price direction over 15m-1h. Report extreme thresholds and hit rates.", "title": "Order Book Imbalance (OBI) One-Shot Scan — Queue exp 206", "why_needed": "Non-GRU signal class that is completely independent of the frozen GRU pipeline. Short-side asymmetry fits validated bilateral strategy. Leading indicator (moves before price). Cheap to implement via CCXT fetch_order_book()."}, {"need_type": "research", "priority": "high", "suggested_next_step": "Create experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ with IDEA.md and run.py. One-shot ccxt query of funding rates for all 28 universe symbols from Gate.io. Compare to 0.03/0.05/0.10 bps thresholds. Log positive vs negative and extreme percentiles.", "title": "Funding Rate Diagnostics — Execute exp 205", "why_needed": "Highest-ROI non-GRU signal class. Freely pollable via CCXT for all 28 symbols. Topic:signal_backlog open since Jul 13 (28 seen_count)."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot data scan producing: (1) funding rate per symbol with threshold classification, (2) cross-sectional percentile distribution, (3) extreme decile flags, (4) integration feasibility assessment for trading_context.json", "hypothesis": "Extreme positive funding rates (>0.05%/8h) on altcoins signal crowded longs and predict short-term reversals; extreme negative funding rates signal crowded shorts and potential long squeezes. Funding rates provide an independent non-GRU signal class for bilateral positioning.", "risk": "Isolated data scan only — no trading decisions, no pipeline changes. Funding rates are point-in-time snapshots; multi-sample analysis needed before deployment.", "title": "Funding Rate Diagnostics — execute exp 205", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "JSON report of all 28 symbols' funding rates, annualized %s, decile ranks, and extreme flags. Covers the topic:signal_backlog need.", "hypothesis": "Funding rates across the Hermes universe show detectable extremes that could serve as contrarian signals. Top/bottom 10% decile symbols are actionable.", "risk": "Pure data scan — no trades, no state changes. Zero risk. Only risk is CCXT rate limiting, which is handled by retry logic.", "title": "Funding Rate Diagnostics for All 28 Hermes Symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}, {"action_type": "data_scan", "expected_output": "OBI ratio for each of top-10 symbols, depth profile, comparison to 0.5 neutral baseline. Flags symbols with extreme OBI (< 0.35 or > 0.65).", "hypothesis": "OBI at depth 10+ levels provides short-term directional signal, especially for short-side setups. Asymmetric predictive power (stronger for drops) fits Hermes bilateral strategy.", "risk": "Pure data scan — no trades, no state changes. Zero risk.", "title": "Order Book Imbalance Snapshot for Top-10 Hermes Symbols", "workspace_path": "experiments_hermes/206-RESEARCH_obi_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics", "experiments_hermes/206-RESEARCH_obi_diagnostics"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "Funding rate flips are the highest-ROI non-GRU signal class available: freely pollable, documented contrarian edge at extremes, and integrate naturally into trading_context.json. GRU dead 40+ days."}], "proposals": []}

{"actions": [{"action_type": "research_investigation", "expected_output": "One-shot data probe: Deribit DVOL current value, comparison to Hermes regime_summary, divergence log entry. If DVOL > 60 while regime_summary=neutral, produce vol red flag recommendation. Scrape VIX daily close from Yahoo Finance. Build experiments_hermes/209/RESULTS.md with probe results.", "hypothesis": "Deribit DVOL and CBOE VIX/VIX regime transitions provide an upstream signal for Hermes position sizing. When DVOL > 60 (shock pricing), entry notional should contract 30-50% and stop-loss width should expand. When DVOL < 40 (sleeping market), notional should expand 20-40% and stop-loss width contract. This is independent of GRU regime signals and complements existing SMA50/200 crossover logic.", "risk": "Isolated experiment — no live order risk. DVOL may be stale (daily index). VIX scrape may fail across weekends. The vol-regime sizing rule is a paper-only recommendation until validated against 30+ closed trades.", "title": "Cross-asset volatility regime overlay for Hermes position sizing", "workspace_path": "experiments_hermes/209-RESEARCH_vol_regime_hermes_overlay"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/209-RESEARCH_vol_regime_hermes_overlay"], "needs": [{"need_topic": "vol_regime_awareness", "need_type": "signal", "priority": "medium", "status": "open", "why_needed": "Hermes regime detection uses SMA50/200 crossover and market breadth only. No awareness of options-implied volatility (DVOL, VIX, MOVE). Adding a vol-regime overlay would directly improve entry sizing (smaller notional during high vol, larger during low vol) without requiring new data pipeline infrastructure."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "IDEA.md with current DVOL/VIX/VVIX/MOVE values, divergence analysis vs regime_summary, and a concrete validation plan for a lightweight periodic vol-check tool appended to the diagnostics pipeline.", "hypothesis": "Deribit DVOL, CBOE VIX/VVIX, and MOVE (treasury vol) provide a leading vol-regime signal that would flag when Hermes should contract vs expand entry sizing — independent of the dead SMA50/200 regime pipeline.", "risk": "Pure research tick — no paper trading impact. Data source reliability (free Deribit stats page may change format). Vol-regime overlay is advisory only, not an entry gate.", "title": "Cross-asset volatility regime leading indicator via VVIX/MOVE/DVOL", "workspace_path": "experiments_hermes/209-RESEARCH_vol_regime_hermes_overlay"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/209-RESEARCH_vol_regime_hermes_overlay"], "needs": [], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "IDEA.md, RESULTS.md with: (1) OBI values for BTC/USDT across 3+ REST pulls; (2) comparison of OBI sign vs concurrent 15m bar direction; (3) decision on whether to build a lightweight L2 polling loop appended to the trading context builder", "hypothesis": "Order book imbalance at the top 10 levels on Gate.io provides a non-zero directional signal for BTC/USDT on 15m bars. OBI > +0.15 (bid-heavy) correlates with upward mid-price movement within one bar; OBI < -0.15 (ask-heavy) correlates with downward movement. Even if the signal is weak (rho < 0.2 on 15m), the existence of any non-zero correlation in a dead-GRU period is valuable as an independent signal class.", "risk": "Experiment only — no live trading. Gate.io REST may rate-limit or return stale depth. OBI may be zero across all pulls (balanced book) proving it's a non-signal for now. Even negative evidence is useful: it would show OBI is not actionable without a websocket stream.", "title": "Order Book Imbalance (OBI) canary for BTC/USDT — proof-of-concept REST pull", "workspace_path": "experiments_hermes/205-RESEARCH_obi_canary"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_obi_canary"], "needs": [], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "experiments_hermes/206-RESEARCH_obi_canary/RESULTS.md with: (1) REST endpoint validation (OBI values, freshness, rate limits), (2) OBI vs 15m bar comparison for 3-5 ticks, (3) feasibility estimate for a lightweight L2 polling loop appended to the trading context builder, (4) proposed integration shape.", "hypothesis": "Order book imbalance (bid_volume - ask_volume) / (bid_volume + ask_volume) at top-10 levels on Gate.io predicts short-term price direction with >55% accuracy for BTC/USDT. Non-zero OBI values correlate with same-direction 15m bar movement.", "risk": "Experimental only. OBI on Gate.io may have lower signal quality than on Binance due to thinner books. The Kalena 58-62% band is for top-10 levels on liquid venues — BTC on Gate.io may or may not qualify. Does not execute trades or modify runtime.", "title": "OBI Canary — One-shot Gate.io order book imbalance scan for BTC/USDT", "workspace_path": "experiments_hermes/206-RESEARCH_obi_canary"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_obi_canary"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ — one-shot ccxt query of funding rates for all 28 universe symbols from Gate.io. Compare to 0.03/0.05/0.10 bps thresholds. Log positive vs negative and extreme percentiles.", "title": "Funding Rate Diagnostics — execute exp 205", "why_needed": "Funding rate flips are the highest-ROI non-GRU signal class available: freely pollable, documented contrarian edge at extremes, and integrate naturally into trading_context.json. Need has been open since Jul 13 (seen 27 iterations). GRU dead 40+ days."}, {"need_type": "data_quality", "priority": "high", "suggested_next_step": "Check paper-trading/hermes-agent/state/ and paper-trading/data/ for historical decision JSON files from Jun 10-Jul 18. If files exist, write a backfill script. If not, accept structural limitation and build forward-compatible window.", "title": "Fix pre-Jul-22 legacy trade attribution gap (180 no_decision_found)", "why_needed": "~180 legacy trades from Jun 10-Jul 22 have no decision metadata. 10% coverage prevents any claim from generalizing to full population."}], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "experiments_hermes/208-RESEARCH_obi_canary/RESULTS.md with: (1) OBI values for 20+ consecutive ticks, (2) OBI vs concurrent bar direction correlation, (3) threshold crossing frequency at ±0.10/0.15/0.20, (4) feasibility estimate for lightweight L2 polling loop", "hypothesis": "Order book imbalance at top 10 levels on Gate.io predicts short-term (1-30min) mid-price direction at 58-62% accuracy for BTC/USDT. OBI divergence between near-price (5 levels) and mid-book (5-20 levels) is more predictive than raw ratio alone.", "risk": "REST depth is a point-in-time snapshot, not streaming. May miss microsecond-scale order book dynamics that actual tape-readers exploit. But for a 15m bar trader, even 1-min stale OBI snapshots can provide edge. GRU dead makes any non-correlated signal worth testing.", "title": "Order Book Imbalance (OBI) canary — Gate.io REST depth scan for BTC/USDT", "workspace_path": "experiments_hermes/208-RESEARCH_obi_canary"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/208-RESEARCH_obi_canary"], "needs": [{"need_type": "research", "priority": "high", "suggested_next_step": "Execute experiments_hermes/208-RESEARCH_obi_canary/ — ccxt Gate.io fetch_order_book for BTC/USDT, compute OBI(10), compare to 15m bar direction. 20+ consecutive ticks to estimate correlation. Report in RESULTS.md.", "title": "OBI canary experiment — gate.io order book scan for BTC/USDT", "why_needed": "Hermes has zero order book awareness despite published microstructure evidence that OBI predicts 58-62% short-term direction on liquid venues. With GRU dead 40+ days, a non-correlated signal class at within-bar horizon is the highest-ROI new signal to investigate."}], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "One-shot OBI(10) value for BTC/USDT from Gate.io REST. If non-zero, log value vs current 15m bar mid-price direction. Deliver IDEA.md and RESULTS.md with findings and feasibility assessment for a lightweight L2 polling loop appended to the trading context.", "hypothesis": "OBI(10) on Gate.io BTC/USDT top-10 levels produces non-zero values that correlate with 15m bar direction. A lightweight polling loop detecting OBI magnitude > 0.30 provides independent entry timing signal for Hermes paper trading decisions.", "risk": "OBI is a high-frequency microstructure signal applied to 15m bars — the time mismatch may dilute predictive power. Keep as isolated experiment, not pipeline integration, until at least 10 samples confirm correlation.", "title": "Order Book Imbalance (OBI) canary — single-shot Gate.io depth scan for BTC/USDT", "workspace_path": "experiments_hermes/208-RESEARCH_obi_canary"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/208-RESEARCH_obi_canary"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot OBI value for BTC/USDT from Gate.io REST. If OBI > 0.15 or < -0.15, log as threshold-crossing event. Compare to current 15m bar direction. Determine if REST endpoint works reliably.", "hypothesis": "OBI(10) on Gate.io BTC/USDT provides a non-correlated non-GRU short-horizon signal. If OBI != 0, it can be logged against current 15m bar direction to validate predictive power in a paper context.", "risk": "OBI is intra-tick (seconds-minutes) — 15m bars may be too coarse to capture edge. Published research targets 1-30min horizons. However, even weak correlation at 15m resolution would justify building a lightweight L2 polling loop.", "title": "Order Book Imbalance (OBI) Canary — Gate.io L2 Depth Scan", "workspace_path": "experiments_hermes/208-RESEARCH_obi_canary"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/208-RESEARCH_obi_canary"], "needs": [{"need_type": "research", "priority": "medium", "suggested_next_step": "Next execution tick: pull BTC/USDT order book from Gate.io via ccxt, compute OBI(10), log result, compare to current 15m bar direction. Write RESULTS.md in experiments_hermes/208-RESEARCH_obi_canary/", "title": "OBI Canary — order book imbalance feasibility scan", "why_needed": "With GRU dead 40+ days, Hermes has zero order book awareness and no short-horizon non-GRU entry-timing signal. OBI is a free, non-correlated class available from Gate.io REST — worth a one-shot validation before committing to a polling loop."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "experiments_hermes/209-RESEARCH_obi_canary/RESULTS.md with OBI(10) baseline, endpoint validation, and design spec for lightweight polling loop", "hypothesis": "Gate.io REST order book depth at top-10 levels produces a usable OBI value for BTC/USDT that can be compared against current 15m bar direction. If OBI != 0.00 and correlates with short-term price direction, a lightweight L2 polling loop can be designed as a non-GRU signal class.", "risk": "OBI at 15m snapshot cadence may be too slow for actionable signals (orderbook is transient). Gate.io REST depth may differ from websocket L2. Exp 042 (market depth idea, archived) flagged this concern. If one-shot scan shows OBI near zero for BTC/USDT, the idea may be low-value for 15m timeframe.", "title": "Order Book Imbalance (OBI) canary — Gate.io depth endpoint validation", "workspace_path": "experiments_hermes/209-RESEARCH_obi_canary"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/209-RESEARCH_obi_canary"], "needs": [], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "experiments_hermes/205/RESULTS.md with per-symbol funding rates, extreme thresholds flagged, and pipeline integration estimate", "hypothesis": "Current funding rates across the 28-symbol universe provide a non-correlated signal class for entry timing and squeeze detection. One-shot ccxt Gate.io scan will identify which symbols are in extreme funding zones (persistent positive = long crowding, persistent negative = short squeeze setup).", "risk": "Funding rate polling from cron may be blocked if ccxt requires non-standard network access. Fallback: use Gate.io public REST endpoint directly (api.gateio.ws) which is curl-accessible.", "title": "Execute exp 205 funding rate diagnostics scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/"}, {"action_type": "experiment", "expected_output": "experiments_hermes/208/IDEA.md + RESULTS.md showing OBI value, divergence from 15m bar, and feasibility assessment for periodic polling loop", "hypothesis": "OBI(10) for BTC/USDT on Gate.io can be computed from live REST depth snapshots. If OBI > +0.15 or < -0.15, OBI diverges from the current 15m bar enough to add signal resolution.", "risk": "Rate limits on Gate.io REST depth endpoint. Need at least 3 snapshots across a 30-min window to observe OBI evolution. Pure research — no deployment risk.", "title": "Order Book Imbalance canary probe", "workspace_path": "experiments_hermes/208-RESEARCH_obi_canary/"}, {"action_type": "experiment", "expected_output": "experiments_hermes/210/RESULTS.md with event age distribution, proposed λ values per event_type, and modified prompt template", "hypothesis": "Stale events (>6h old) dominate the event queue and dilute fresh signal. Applying exponential decay would shift decision weight to events < 2h old.", "risk": "Read-only analysis of existing event_log. No new data sources. No runtime changes.", "title": "News sentiment decay-weighting probe", "workspace_path": "experiments_hermes/210-RESEARCH_news_sentiment_decay/"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/", "experiments_hermes/208-RESEARCH_obi_canary/", "experiments_hermes/210-RESEARCH_news_sentiment_decay/"], "needs": [{"need_topic": "signal_backlog", "need_type": "research", "priority": "medium", "status": "open", "why_needed": "Hermes has zero order book awareness. OBI is validated in microstructure research (58-62% accuracy on 5-min horizons). With GRU dead for 40 days, any non-correlated short-term signal class is valuable."}, {"need_topic": "signal_backlog", "need_type": "research", "priority": "medium", "status": "open", "why_needed": "Hermes regime detection uses SMA and breadth only. Adding DVOL (Deribit's crypto VIX) would flag when IV is pricing shock (DVOL > 70) vs sleeping market (DVOL < 40). Cross-asset vol leads regime changes by days to weeks."}, {"need_topic": "strategy_and_signal_design", "need_type": "research", "priority": "medium", "status": "open", "why_needed": "32K events exist. Events hours/days old carry equal weight in decision prompt. News alpha has proven short half-life (minutes for high-impact, hours for low-impact). Decay-weighting would improve signal-to-noise ratio."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Funding rate snapshot for all 28 symbols, threshold zone classification, historical flip correlation coefficients, pipeline integration feasibility estimate", "hypothesis": "When funding flips from positive to negative, overleveraged longs have been flushed. Waiting 6-12h after the flip, if funding stays negative and price stabilizes, creates a long entry zone with squeeze potential. At 0.05%+ extreme funding, the crowded side is vulnerable to a violent squeeze within 24h. This is a non-correlated signal class Hermes has never tested.", "risk": "Gate.io funding rate data quality unknown; historical flip correlation sample may be small if data fetch is rate-limited; remains isolated in experiments_hermes/ until VALIDATED", "title": "Funding Rate Flip as Contrarian Entry Signal", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/IDEA.md, RESULTS.md with current funding snapshot by symbol, threshold zone counts, and pipeline integration feasibility notes", "hypothesis": "When funding flips from positive to negative (or passes 0.05%+ extreme zones), the crowded side is vulnerable to a squeeze within 24h. A one-shot scan of current Gate.io funding rates across all 28 Hermes symbols will validate data availability, compute $0.03/0.05/0.10 threshold zones, and assess integration feasibility.", "risk": "Gate.io may have incomplete funding rate data or non-standard update intervals for some symbols; research-only — no trading decisions depend on this scan", "title": "Funding Rate Flip as Contrarian Entry Signal — Gate.io Funding Rate Diagnostics Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ — one-shot Gate.io funding rate scan across all 28 symbols, compare current rates to 0.03/0.05/0.10 zones, compute historical flip correlations if possible, produce pipeline integration feasibility estimate", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan (queued to exp 205)", "why_needed": "Hermes has zero funding rate awareness. GRU has been dead for 24+ days. Funding rate flips provide a non-correlated, non-GRU signal class for entry timing and squeeze risk detection. Queue assigned to experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ for next execution tick."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot Gate.io funding rate scan across all 28 symbols; current rates compared to 0.03/0.05/0.10 threshold zones; historical flip correlations with 15m forward returns from Hermes own data; pipeline integration feasibility estimate.", "hypothesis": "Funding rate flips from positive to negative (or vice versa) produce mean-reversion/squeeze opportunities within 6-12h. Extreme funding (>0.05%) makes the crowded side vulnerable to violent squeeze within 24h. This is a non-correlated signal class that can inform Hermes entry timing independent of GRU and price action.", "risk": "Funding rate data quality unknown — Gate.io may have lower liquidity/sampling frequency than Binance for some symbols. Historical flip data may be sparse. This is a research scan only, not a deployment.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Funding rate diagnostics scan now queued to experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ for next research tick execution.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan (now queued)", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries. GRU has been dead for 24+ days, making alternative signal sources critical."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "IDEA.md and RESULTS.md: per-symbol funding rate snapshot, threshold zone comparison, historical flip correlation with forward returns, pipeline integration feasibility estimate", "hypothesis": "Funding rate flips from positive to negative identify long-flush exhaustion zones where entering LONG 6-12h after stabilization has positive EV. At 0.05%+ extreme funding, the crowded side is vulnerable to violent squeeze within 24h.", "risk": "Gate.io funding rate data may be delayed or only available for perpetuals subset. Some symbols may not have funding rate endpoint. Exploratory only — no strategy deployment without validation.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan Across 28 Symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ — one-shot Gate.io funding rate scan across all 28 symbols, compare current rates to 0.03/0.05/0.10 zones, compute historical flip correlations with forward returns, pipeline integration feasibility estimate.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries. GRU has been dead for 24+ days, making alternative signal sources critical."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "One-shot Gate.io funding rate scan across all 28 symbols, comparison of current rates to 0.03/0.05/0.10 threshold zones, correlation of historical funding flips with 15m forward returns from Hermes own data, pipeline integration feasibility estimate.", "hypothesis": "When funding flips from positive to negative, overleveraged longs have been flushed. Waiting 6-12h after the flip, if funding stays negative and price stabilizes, creates a long entry zone with squeeze potential. At 0.05%+ extreme funding, the crowded side is vulnerable to a violent squeeze within 24h. This is a non-correlated signal class Hermes has never tested.", "risk": "Isolated to experiments_hermes/. Requires CCXT Gate.io API access (already available). No live trading impact. If funding rate data is unreliable or too sparse, the experiment will say so and the idea can be abandoned.", "title": "Funding Rate Flip as Contrarian Entry Signal — Gate.io Endpoint Scan & Pipeline Integration Feasibility", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "research", "priority": "high", "suggested_next_step": "Queue experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ run. Scan Gate.io endpoint for all 28 symbols, compare to 0.03/0.05/0.10 threshold zones, produce pipeline integration feasibility.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries. GRU has been dead for 24+ days, making alternative signal sources critical."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot Gate.io funding rate scan across all 28 Hermes symbols. Compare current rates to 0.03/0.05/0.10 threshold zones. Compute historical flip correlations with 15m forward returns from Hermes own data. Pipeline integration feasibility estimate.", "hypothesis": "When funding flips from positive to negative, overleveraged longs have been flushed. Waiting 6-12h after the flip, if funding stays negative and price stabilizes, creates a long entry zone with squeeze potential. At 0.05%+ extreme funding, the crowded side is vulnerable to a violent squeeze within 24h. This is a non-correlated signal class Hermes has never tested.", "risk": "Gate.io CCXT funding rate endpoint may not expose historical data or may rate-limit. Stays isolated in experiments_hermes/ until validated against real data.", "title": "Funding Rate Flip as Contrarian Entry Signal — Gate.io Diagnostics Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Gate.io funding rate snapshot for all 28 symbols, current rate vs 0.03/0.05/0.10 threshold zones, historical flip correlation with 15m forward returns from Hermes own data cache, pipeline integration feasibility estimate", "hypothesis": "Funding rate flips (positive→negative or extreme >0.05%) are predictive of short-term 15m-1h forward returns for Hermes' 28-symbol universe. When funding flips from positive to negative and price stabilizes over 6-12h, a long entry zone with squeeze potential exists. At extreme funding (>0.05%), the crowded side is vulnerable to violent squeeze within 24h.", "risk": "Funding rate data is perp-specific — spot-only symbols in Hermes universe will return null. Gate.io CCXT endpoint reliability for historical funding data is unproven. This is a pure data-scan experiment with no trading pipeline changes, isolated in experiments_hermes/.", "title": "Funding Rate Flip as Contrarian Entry Signal — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Queue experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ run. Scan Gate.io endpoint for all 28 symbols, compare to 0.03/0.05/0.10 threshold zones, produce pipeline integration feasibility.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries. GRU has been dead for 24+ days, making alternative signal sources critical."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "JSON report with current funding rates for all 28 symbols, threshold zones (0.03/0.05/0.10), and pipeline integration feasibility assessment.", "hypothesis": "Gate.io funding rate data is available and actionable. Hermes should acquire a single funding rate snapshot to determine data quality, accessibility, and format.", "risk": "Stays isolated — no PM2, no trading config changes, no code deployment. Pure data acquisition and analysis.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/"}, {"action_type": "data_scan", "expected_output": "OBI z-score for BTC + top 5 Hermes symbols (SOL, XRP, ADA, TRX, FIL based on current position + volume), compared with 15m price action from current context.", "hypothesis": "Current OBI from Gate.io L2 order book is measurable and correlates with 15m forward returns in Hermes' timeframe.", "risk": "Stays isolated — pure data analysis using CCXT and existing Hermes data. No trading decisions.", "title": "Order Book Imbalance Snapshot for Top 5 Hermes Symbols", "workspace_path": "experiments_hermes/206-RESEARCH_order_book_imbalance_scan/"}, {"action_type": "data_scan", "expected_output": "24h liquidation volume for BTC and SOL, liquidation velocity trend, and risk assessment for the current SOL/USDT long position.", "hypothesis": "Current SOL liquidation data from Gate.io shows whether recent long liquidations create squeeze risk for the open SOL position.", "risk": "Stays isolated — pure data analysis. The output may inform the trading decision tick but does not modify any code.", "title": "Liquidation Signal Scan for SOL/BTC", "workspace_path": "experiments_hermes/207-RESEARCH_liquidation_signal_scan/"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/", "experiments_hermes/206-RESEARCH_order_book_imbalance_scan/", "experiments_hermes/207-RESEARCH_liquidation_signal_scan/"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries."}], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "Gate.io funding rate CSV for 5 symbols, 7d rolling mean vs current rate for each, threshold zone proposal (0.03/0.05/0.10), pipeline integration feasibility report, and recommendation for inclusion in trading_context.json as a new signal family.", "hypothesis": "Gate.io perpetual funding rates for SOL, BTC, ETH, FET, TRX show extreme deviations (>0.05% from 7d rolling mean) that serve as mean-reversion entry/exit signals. Negative extremes → LONG setup. Positive extremes → SHORT setup. Orthogonal to GRU proba and composite-score logic.", "risk": "Funding rate data may be sparse (1h snapshots) or require additional perp-specific endpoint calls. If Gate.io API does not expose historical rate series, the diagnostic will be limited to current-snapshot-only. Isolated experiment; no pipeline changes until validated.", "title": "Funding Rate Diagnostic — Gate.io endpoint scan, threshold zones, pipeline integration feasibility", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "Gate.io funding rate snapshot across SOL, BTC, ETH, FET, TRX; 7d rolling mean vs current deviation; 0.03/0.05/0.10 threshold zone analysis; pipeline integration feasibility for trading_context.json; IDEA.md and RESULTS.md artifacts", "hypothesis": "Gate.io funding rate deviations beyond 2-sigma from 7d rolling mean provide a non-correlated mean-reversion entry signal that improves Hermes paper trading decisions, especially for short-side entries where Hermes has the weakest empirical track record.", "risk": "Funding rate is a perpetual-futures metric; if the paper account uses spot-only positions, the signal is advisory rather than directly executable. However, Hermes' sandbox supports futures shorts (isolated 1x), and funding rate awareness would improve short-side entry timing. Data fetching is read-only and isolated within experiments_hermes/.", "title": "Gate.io funding rate signal for Hermes top-10 symbols — pipeline integration feasibility", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ run.py to scan Gate.io funding endpoints across SOL, BTC, ETH, FET, TRX; compute 7d rolling mean vs current rate; produce threshold zones and pipeline integration feasibility.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "why_needed": "Hermes has zero funding-rate awareness. This is the first non-correlated signal class Hermes would acquire. The GRU pipeline has been dead for 24+ days, and funding rate is orthogonal to both GRU proba and composite-score entry logic. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Gate.io funding rate time series for 5 symbols, 7d rolling mean, current-rate-vs-mean deviation. If 2+ symbols show extreme deviation (>0.05% from mean), produce integration feasibility for trading_context.json diagnostic pipeline.", "hypothesis": "Funding rate extreme deviations (beyond 2-sigma from 7d rolling mean) on Gate.io for SOL, BTC, ETH could serve as mean-reversion entry signal: extreme negative → long setup, extreme positive → short setup. This is orthogonal to both GRU proba and composite-score entry logic.", "risk": "Gate.io funding rate endpoint may require separate API call or subscription tier beyond standard market data. Even without pipeline integration, the diagnostic establishes the first non-correlated signal baseline in the Hermes experiment index.", "title": "Gate.io funding rate endpoint scan for SOL, BTC, ETH, FET, TRX — establish threshold zones and pipeline integration feasibility", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire — orthogonal to GRU proba, composite-score logic, and market breadth. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries."}, {"need_topic": "data_quality_backlog", "need_type": "data_quality", "priority": "high", "status": "open", "why_needed": "179/200 trades (89.5%) are unattributed because 'no decision found' — they closed without Hermes writing pending_trading_decision.json. This has been the single largest bottleneck across 20+ consecutive reflection windows with zero structural progress. Coverage at 10.5% prevents any claim from generalizing to full Hermes performance."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "One-shot Gate.io funding rate history scan across 5+ symbols. Compare 7d rolling mean vs current rate. If 2+ symbols show extreme deviation (>0.05% from mean), produce integration feasibility for the diagnostic pipeline. IDEA.md + RESULTS.md with threshold zones (0.03/0.05/0.10 basis points) and pipeline integration analysis.", "hypothesis": "Gate.io perpetual swap funding rate extreme deviations (beyond 2-sigma from 7d rolling mean) for SOL, BTC, ETH, FET, TRX serve as mean-reversion entry signals: extreme negative → long setup, extreme positive → short setup. This signal family is orthogonal to both GRU proba and composite-score entry logic.", "risk": "Funding rate data is specific to each exchange — Gate.io rates may not reflect broader market. CCXT rate limits and data availability for historical funding need verification. This stays isolated because deployment requires pipeline changes, not just policy adjustment.", "title": "Gate.io funding rate diagnostics — endpoint scan and threshold zone analysis", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Funding rate table per symbol: current rate, 7d mean, deviation sigma, threshold zone classification. Feasibility assessment for pipeline integration into trading_context.json", "hypothesis": "Funding rate extreme deviations (beyond 2-sigma from 7d rolling mean) on Gate.io for SOL, BTC, ETH, FET, TRX serve as mean-reversion entry signal: extreme negative → long setup, extreme positive → short setup. Phemex analysis confirms sustained negative funding rates since early 2026 — the longest streak since Nov 2022.", "risk": "Pure data scan — no trade execution, no pipeline changes. Gate.io may not expose historical funding rate data via public endpoints; may require CCXT or cached snapshot. If endpoint unavailable, fall back to report documenting the limitation.", "title": "Gate.io funding rate diagnostics — endpoint scan across Hermes top symbols", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ with one-shot Gate.io endpoint scan across SOL, BTC, ETH, FET, TRX. Establish 0.03/0.05/0.10 threshold zones. Produce pipeline integration feasibility.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries. The open SOL position needs an empirical exit trigger."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "IDEA.md for experiments_hermes/206-RESEARCH_ob_imbalance_btc_filter with: (1) Gate.io BTC/USDT L2 order book fetch method via CCXT; (2) imbalance formula and threshold zones; (3) cross-correlation with BTC 1h return over 3 snapshots; (4) integration feasibility for trading_context.json", "hypothesis": "BTC/USDT L2 order book imbalance (bid_vol / (bid_vol + ask_vol)) from Gate.io CCXT snapshots produces a microstructure signal orthogonal to OHLCV-based macro tools. AlgoTick 90-day backtest shows Sharpe 2.78, 59% WR, 2.58 profit factor at avg 7h hold. At extreme imbalance (<0.35 or >0.65), BTC 1h forward return should confirm micro-correlation, making imbalance usable as a regime override for alt entries and as an exit trigger for the open SOL position.", "risk": "This is entirely exploratory. Order book data may be noisy on Gate.io's L2 snapshot depth. No trading decisions depend on this research. Deferred to Codex/human review before any pipeline integration.", "title": "Order book imbalance as ensemble-compatible signal family for BTC regime filter", "workspace_path": "experiments_hermes/206-RESEARCH_ob_imbalance_btc_filter"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/206-RESEARCH_ob_imbalance_btc_filter"], "needs": [], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "Threshold zones per symbol, symbol-level funding rate profiles, comparison with 7d rolling mean, integration feasibility assessment for trading_context.json diagnostic pipeline", "hypothesis": "Funding rate extreme deviations (beyond 2-sigma from 7d rolling mean) on Gate.io for SOL, BTC, ETH, FET, TRX can serve as mean-reversion entry/exit signal orthogonal to Hermes' current OHLCV-only composite-score logic — extreme negative → long setup, extreme positive → short setup", "risk": "Funding rates are perp-specific, not spot — may not correlate with entry/exit timing in Hermes' spot paper account. Gate.io may not expose historical funding rate data through CCXT. Old exp 041 was never executed; this validates whether the signal class is worth the integration cost.", "title": "Gate.io funding rate diagnostics — one-shot endpoint scan and pipeline feasibility", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "IDEA.md + RESULTS.md showing 7d funding rate rolling mean vs current by symbol, extreme deviations flagged, integration feasibility for trading_context.json", "hypothesis": "Gate.io perpetual funding rate extreme deviations have signal value for Hermes entry timing", "risk": "Stays isolated — no deployment proposal without human review. No PM2 or registry changes.", "title": "Funding rate Gate.io endpoint scan for SOL/BTC/ETH", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/"}, {"action_type": "data_scan", "expected_output": "RESULTS.md with N SOL trades, holding times, PnL at h=2/4/8/12. Recommendation for current SOL position exit plan.", "hypothesis": "Hermes SOL trades have an empirically optimal holding time that differs from the generic 240-min review window", "risk": "Small sample problem — if N<5 SOL trades with linked decisions, result may be inconclusive. Stays isolated.", "title": "SOL historical trade exit analysis from decision logs", "workspace_path": "experiments_hermes/208-sol-exit-timing-empirical/"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/", "experiments_hermes/208-sol-exit-timing-empirical/"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "Hermes has zero funding rate awareness. Funding rates on Gate.io for SOL, BTC, ETH are orthogonal to both GRU proba and composite-score entry logic. This is a genuinely new signal family."}, {"need_topic": "data_quality_backlog", "need_type": "data_quality", "priority": "high", "status": "open", "why_needed": "Attribution coverage at 9% means 91% of all trades are invisible to reflection. The legacy 'no_decision_found' gap (178 trades, Jun 10-Jul 18) can only be fixed by reconstructing decision metadata from historical logs."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "Per-symbol funding rate table, 30d z-scores, threshold zones (0.03/0.05/0.10), pipeline integration feasibility assessment, and decision to add funding_rate_advisory field to symbol snippets.", "hypothesis": "Gate.io perp funding rate z-scores and negative-to-positive flips provide a non-correlated entry timing signal for Hermes' 28-symbol universe. Persistently negative funding (>1 week at -4% annualized) with rising OI signals latent short-squeeze fuel; a flip to positive signals exhaustion.", "risk": "Gate.io may not have perp contracts for all 28 symbols. Funding rate history may be limited. Risk is data availability, not strategy safety.", "title": "Funding Rate Flip as Non-Correlated Entry Timing Edge", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "One-shot Gate.io endpoint scan across all 28 Hermes symbols: per-symbol funding rate, 30d z-score, establishment of 0.03/0.05/0.10 threshold zones, and pipeline integration feasibility assessment.", "hypothesis": "Funding rate flips (negative→positive after persistent short positioning) precede significant price reversals. Adding a funding-rate advisory column to Hermes symbol snippets would give a second opinion on overbought/oversold crowding, completely non-correlated to existing RSI+VWAP+momentum+meanrev composite scoring.", "risk": "Funding rate data quality and availability on Gate.io may differ from Kraken/Hyperliquid. Cross-exchange spread requires additional endpoint access. Isolated to experiments_hermes/ — no runtime impact until validated.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open"}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ with IDEA.md, RESULTS.md, a one-shot run.py that fetches funding rates and produces a symbol-vs-rate matrix with threshold zones marked, and pipeline integration feasibility notes", "hypothesis": "Funding rate flips (negative→positive after persistent short positioning) precede significant price reversals. Gate.io perp funding rates across Hermes' 28-symbol universe can be fetched and threshold-zoned (0.03/0.05/0.10) for integration into symbol snippets as a non-correlated advisory column.", "risk": "Gate.io may not offer funding rate on all 28 symbols (some may be spot-only). Funding rate history may be unavailable on first fetch (cold start). Pure paper research — no deployment risk.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries. All 5 GRU slots dead since Jul 3 (23+ days)."}], "proposals": []}

{"actions": [{"action_type": "code", "expected_output": "Per-symbol funding rate table (28 Hermes symbols), 30d z-scores, threshold zones (0.03/0.05/0.10), comparison vs existing composite score entries, and pipeline integration feasibility assessment (as IDEA.md + RESULTS.md + out/funding_rate_data.json).", "hypothesis": "Funding rate flips (negative→positive after persistent short positioning) precede significant price reversals. Gate.io funding rate data can be integrated into Hermes as a second opinion on overbought/oversold crowding.", "risk": "Funding rates can stay extreme in strong trends — contrarian signals without trend context are dangerous. This is isolated in experiments_hermes/ and will not affect live trading until validated.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_type": "signal", "priority": "high", "suggested_next_step": "Execute experiments_hermes/205-RESEARCH_funding_rate_diagnostics/ with one-shot Gate.io endpoint scan per the explore research plan.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "why_needed": "Hermes has zero funding rate awareness (confirmed by topic:signal_backlog need). Adding a funding-rate advisory column to symbol snippets would give Hermes a second opinion on overbought/oversold crowding — the first non-correlated signal class."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Status report: (a) whether sector fields exist in live context, (b) 28-symbol sector mapping, (c) per-symbol sector divergence scores, (d) recommendations for live integration or diagnostic addition.", "hypothesis": "Exp 207 (sector rotation) was marked implemented but the sector mapping + 7d sector return computation may not be producing live sector_relative_strength fields in trading_context.json. The topic:signal_fusion need was resolved for FET interpretation advisory, not sector rotation — leaving the rotation gap unaddressed.", "risk": "Uses existing OHLCV data only — zero upstream risk. If sector mapping reveals no significant divergences, the audit still produces a documented baseline for the universe.", "title": "Sector Rotation Audit — Verify Exp 207 Sector Mapping in Live trading_context.json", "workspace_path": "experiments_hermes/205-RESEARCH_sector_rotation_live_audit"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_sector_rotation_live_audit"], "needs": [{"need_type": "research", "priority": "high", "status": "open", "why_needed": "A need closed for the wrong reason leaves a genuine pipeline gap unaddressed. The sector rotation audit will confirm whether live data exists or needs to be built."}], "proposals": []}

{"actions": [{"action_type": "data_scan", "expected_output": "Per-symbol funding rate, 30d z-score, 0.03/0.05/0.10 threshold zone mapping, cross-exchange spread feasibility note, and pipeline integration report marking whether funding-rate advisory column can be added to trading_context.json symbol snippets.", "hypothesis": "Persistent negative funding in rising market confirms bull entries; persistent positive funding validates SHORT candidates. Funding rate flip from negative to positive signals short-squeeze exhaustion. This is a completely non-correlated signal to Hermes' existing composite scoring.", "risk": "Isolated to data scan — no trading logic changes. Gate.io endpoint rate limits may paginate slowly. No Binance/Hyperliquid cross-exchange spread if those endpoints are not accessible.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "Hermes has zero funding rate awareness. This would be the first non-correlated signal class. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries."}, {"need_topic": "signal_fusion", "priority": "high", "status": "open", "why_needed": "Explore research flagged that Exp 207 (sector rotation) was marked implemented but live sector data may not be present in trading_context.json. Sector divergence would be a second orthogonal signal class."}], "proposals": []}

{"actions": [{"action_type": "code", "expected_output": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/out/funding_rates.csv with per-symbol funding rate, 30d z-score, flip detection, and validation against recent ENTER/EXIT decisions", "hypothesis": "Persistently negative funding rates (>1 week) followed by a flip to positive precede short-term price reversals. Gate.io funding rate data across the 28-symbol Hermes universe can provide a non-correlated entry/exit advisory independent of dead GRU probes.", "risk": "Isolated to experiments_hermes/; no pipeline integration. Gate.io may return stale funding rates during low-liquidity hours. If funding rates are uniformly near-zero for all Hermes symbols, the signal class is not actionable.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/RESULTS.md with per-symbol funding rate snapshot, 30d z-score, threshold zones (neutral 0.0-0.03, elevated 0.03-0.10, extreme >0.10 / negative <-0.10), and pipeline integration feasibility assessment", "hypothesis": "Persistent negative funding (>1 week) with rising OI predicts short squeezes; persistent positive funding predicts exhaustion. Funding rate data from Gate.io can produce a non-correlated advisory layer independent of dead GRU pipeline.", "risk": "Gate.io funding rate endpoint may not support all 28 Hermes symbols (only perpetual futures contracts). Cross-exchange spread vs Binance/Hyperliquid may require multiple API keys. This is isolated research in experiments_hermes/ with no runtime impact.", "title": "Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "status": "open", "why_needed": "Queued as the selected research action from explore consumption gate. Hermes has zero funding rate awareness. This is the first non-correlated signal class Hermes would acquire. Without it, the agent cannot distinguish genuine reversal from spoofed support for perp entries."}], "proposals": []}

{"actions": [{"action_type": "experiment", "expected_output": "Markdown report file: FUNDING_SCAN_RESULTS.md with per-symbol funding rate, 30d z-score (if historical data available), and pipeline integration feasibility assessment.", "hypothesis": "Gate.io public REST API returns per-symbol funding rates for all 28 Hermes symbols. A one-shot scan establishes baselines, identifies symbols with persistently extreme funding (|rate| > 0.05% 8h), and determines pipeline integration feasibility.", "risk": "Isolated experiment, no deployment. Gate.io API rate-limited but within free tier. Output may show mostly zero/small funding for altcoins — the signal lives on BTC/ETH majors.", "title": "205-RESEARCH Funding Rate Diagnostics — Gate.io Endpoint Scan", "workspace_path": "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/"}, {"action_type": "experiment", "expected_output": "PROXY_IMBALANCE_RESULTS.md with per-decision alignment classification and WR split (aligned vs divergent). If divergent < 40% WR, recommends adding VWAP-momentum alignment to entry screen.", "hypothesis": "Hermes ENTER decisions where VWAP delta and momentum velocity were aligned (VVAP discount + rising mom = bullish; VWAP premium + falling mom = bearish) have higher WR than divergent entries (VWAP discount + falling mom or VWAP premium + rising mom).", "risk": "Historical decision JSONs may not have both VWAP delta and momentum at entry bar. Sample may be small if only agent_trading_decisions.jsonl is used.", "title": "209-RESEARCH Proxy Imbalance Divergence — Entry Quality Scan", "workspace_path": "experiments_hermes/209-RESEARCH_proxy_imbalance_divergence/"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_funding_rate_diagnostics/", "experiments_hermes/209-RESEARCH_proxy_imbalance_divergence/"], "needs": [{"need_topic": "signal_backlog", "need_type": "signal", "priority": "high", "why_needed": "Funding rate is the first non-correlated signal class Hermes would acquire. The persistent open need has produced zero concrete action. The experiment path (205-RESEARCH) is now defined with a run.py and expected output. This need requires the experiment to be executed."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "IDEA.md with hypothesis and method; one-shot Gate.io REST scan results showing imbalance ratio distribution across symbols; 15m forward return correlation by imbalance threshold zone; pipeline integration feasibility assessment", "hypothesis": "Order book imbalance ratio = (bid_vol_5levels - ask_vol_5levels) / (bid_vol_5levels + ask_vol_5levels) from Gate.io spot orderbook correlates with subsequent 15m return direction for liquid pairs. An imbalance_ratio > 0.3 or < -0.3 could serve as a binary entry-timing veto overlay.", "risk": "Order book data is ephemeral — a single REST snapshot may not capture the imbalance dynamics. Gate.io orderbook depth and precision for lower-cap symbols may be insufficient. This is a purely diagnostic research task with no trading execution; no capital at risk.", "title": "Order Book Imbalance as Short-Term Entry Timing Overlay for Liquid Pairs", "workspace_path": "experiments_hermes/208-RESEARCH_order_book_imbalance"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/208-RESEARCH_order_book_imbalance"], "needs": [], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "Imbalance ratio distribution per symbol; threshold zones; correlation snapshot against 15m forward return; feasibility assessment for pipeline integration", "hypothesis": "Gate.io order book imbalance ratio (bid_vol_5levels - ask_vol_5levels) / (bid_vol_5levels + ask_vol_5levels) > 0.3 correlates with positive 15m returns > 50% of the time for liquid pairs (BTC, ETH, SOL, XRP, LTC). An imbalance-ratio binary veto could improve entry timing on composite-score candidates.", "risk": "One-shot snapshot may not generalize across market regimes. Gate.io order book depth may be thin for mid-cap symbols (ALGO, RUNE, ARB). Single-venue bias (Gate.io may not reflect broader market).", "title": "Order Book Imbalance — Gate.io One-Shot Scan for Top 10 Liquid Hermes Symbols", "workspace_path": "experiments_hermes/205-RESEARCH_order_book_imbalance_gateio_scan"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_order_book_imbalance_gateio_scan"], "needs": [{"need_type": "research", "priority": "high", "suggested_next_step": "Implement experiments_hermes/205-RESEARCH_order_book_imbalance_gateio_scan: one-shot Gate.io orderbook scan for top 10 liquid Hermes symbols. Compute imbalance ratio. Compare against subsequent 15m return. If imbalance >0.3 correlates with positive returns >50%, write pipeline integration feasibility assessment.", "title": "Gate.io Order Book Imbalance Scan — Short-Term Entry Timing Overlay", "why_needed": "Hermes has zero order-book awareness. Adding a simple imbalance-ratio binary veto (from single Gate.io REST call) would provide a new signal family completely uncorrelated with OHLCV, GRU, or regime signals. This is the first non-OHLCV signal dimension."}], "proposals": []}

{"actions": [{"action_type": "web_research", "expected_output": "experiments_hermes/205-RESEARCH_order_book_imbalance_overlay/RESULTS.md with imbalance ratios, 15m return correlations, threshold recommendations, and pipeline integration feasibility assessment.", "hypothesis": "Gate.io order book imbalance ratio (bid_vol_top5 - ask_vol_top5) / (bid_vol_top5 + ask_vol_top5) provides a leading short-term timing signal for liquid Hermes pairs. Imbalance > 0.3 should correlate with positive 15m returns > 50% of the time, offering a binary entry-timing veto (reject LONG if imbalance negative, reject SHORT if imbalance positive).", "risk": "Gate.io spot order book depth may be insufficient for less liquid pairs. 15m forward return window is short — noise dominates. Single time snapshot may not generalize across market regimes. Stays isolated in experiments_hermes/ until validated.", "title": "Order Book Imbalance as Short-Term Entry Timing Overlay for Liquid Pairs", "workspace_path": "experiments_hermes/205-RESEARCH_order_book_imbalance_overlay"}], "artifacts": ["paper-trading/hermes-agent/state/research_latest.json", "paper-trading/hermes-agent/logs/session_events.jsonl", "experiments_hermes/205-RESEARCH_order_book_imbalance_overlay"], "needs": [], "proposals": []}