WHALEOPSPAPERPERFJOURNALACCUMEXIT LABKNOWLEDGEROADMAPSPEC v5.5.0
HERMES ARCHITECTURE — Living System Map
SNAPSHOT 260706
Layers
10
infra → knowledge
Processes mapped
18
click any to drill in
Cron jobs
~113
+60 host crons
Est cost/day
$3
LLM + infra, data ≈ $0
Wallet DB
1,719
never deleted (P2)
Golden stack
+4.2%
net/trade · n=1,188
§ Reading this page 260811
This curated diagram below is hand-authored and drifts — it was last substantially revised 260706 and does not yet show the trading-decision layer as it exists today (admission_gate, the unified TDE decision boundary, decision_fabric health, the M20 human/script attention split, or the LLM wired into it 260811). For that layer specifically, see the dedicated DECISION ENGINE page — a fresh, fully-detailed drillable diagram of exactly that pipeline, including precisely how the new LLM fits in and what it does and does not affect. For a diagram that cannot go stale (every process this system actually runs, right now, auto-regenerated every 4h), jump to § Live System Map further down this page. For per-feed scanner/screener detail (thresholds, gates, expectancy — 50 feeds, generated fresh from live code), see SCANNER.
§ Master Architecture — the whole machine CLICK ANY NODE TO DRILL IN
🔍 detect · ✚ enrich · 🚪 gate · ⚡ boost · 🔁 learn — doctrine: it is not the token, it is the WALLET and the MOMENT.
flowchart TB
  subgraph L1["L1 · DATA INGESTION — free by design"]
    ING["GMGN · DexScreener · Dune · 30+ TG channels
ProTrending · FOMO Multi · dexLions/Cipher piggyback · chain data"] end subgraph L2["L2 · DETECTION 🔍"] ACC["accumulation engine"] FPS["fresh predecessor EP-7"] WMT["whalemover + sapphy"] DRM["dormancy scanner"] MOM["momentum: trending·sniper·launch·volume"] CAT["narrative catalyst detector"] end subgraph L3["L3 · ENRICHMENT ✚"] CARD["TOKEN CARD — wallet intel · deployer 🚪 · dormancy · volume · regime"] end subgraph L4["L4 · CONVERGENCE ⚡"] MS["master scanner 180 — 2+ sources"] APX["APEX siren (staged)"] end subgraph L5["L5 · DECISION + EXECUTION — full pipeline: see decision_engine.html"] CDE["decision engine — admission_gate → TDE → M20
COPY / FRONT-RUN / SKIP (+ LLM-assisted 260811)"] PPR["paper trader → real rails"] end subgraph L6["L6 · LEARNING 🔁"] KLOOP["outcome tracker → FDR evaluator → Karpathy review"] end subgraph L7["L7 · WALLET INTELLIGENCE"] WDB[("wallet_db 7,208 — the moat")] TDWA["TDWA + cluster discovery"] end subgraph L8["L8 · GOVERNANCE + BRAIN"] GOVN["queue 37479 🚪 · versioning · watchdogs · knowledge bridge"] end subgraph L9["L9 · META / SELF-IMPROVEMENT 🔁 — hover to reveal"] FLY["Wallet Flywheel"] EP1["GMGN Alpha Leaderboard
+ Edge Discovery"] SKSY["Claude Skill Synthesis"] INSC["Dashboard Insight Compounder"] end ING --> ACC & FPS & WMT & DRM & MOM & CAT ACC & FPS & WMT & DRM & MOM --> CARD --> MS --> CDE CAT -.->|narrative_flag join| ACC & WMT MS -.-> APX CDE --> PPR --> KLOOP KLOOP -->|weights| CDE & CARD & ACC TDWA -->|graduate 🚪| WDB --> CARD KLOOP --> GOVN TDWA -.->|reuses engine| FLY -.->|proposal| WDB EP1 -.->|proposal| WDB KLOOP -.->|same self-improve loop| SKSY APX -.->|weights + scorecard| INSC -.->|insight bus| KLOOP classDef meta fill:#1a1f26,stroke:#3a4048,color:#8891a0,stroke-dasharray:4 3 class FLY,EP1,SKSY,INSC meta click ACC call openProc("accum") click FPS call openProc("ep7") click WMT call openProc("whalemover") click DRM call openProc("dormancy") click MOM call openProc("momentum") click CAT call openProc("catalystdetector") click CARD call openProc("tokencard") click MS call openProc("scanner") click APX call openProc("apex") click CDE call openProc("copydecision") click PPR call openProc("paper") click KLOOP call openProc("karpathy") click WDB call openProc("walletdb") click TDWA call openProc("tdwa") click GOVN call openProc("governance") click ING call openProc("ingestion") click FLY call openProc("flywheel") click EP1 call openProc("gmgnleaderboard") click SKSY call openProc("skillsynth") click INSC call openProc("insightcompounder")
§ Process Index — filter by layer
§ Improvement Roadmap — ranked (impact ÷ effort) · from spec §17
  1. P1 outcome-truth repair✅ SHIPPED 260705 tracker v1.2.0: alpha view past SL, moonbag accounting, winner-follow 72h, class-aware wins. Next: evaluator consumes the new fields after ~2wk data accrual.
  2. E1 / EP-11 serial early-winner intersection — intersect earliest profitable buyers across independent 7d 10× tokens. Queue #53.
  3. A3 event bus✅ SHIPPED 260706 event_bus.emit() fire-time bus + apex_runner merge. Backtest: 69% of fires would fire earlier, median 160min saved (WATCH-LATE 94%).
  4. EP-11 serial early-winner intersection✅ LIVE 260706 daily proposal-only; #1 result self-validated (already in DB); 8 new/run → 7423.
  5. E2 sell-side intelligence✅ SHIPPED 260706 tracked-wallet sells captured; paper force-closes on SM_DISTRIB (≥2 sellers/6h). Live: caught 中公教育 +142% near peak. Next: net-position line on cards + live copy_decision SKIP.
  6. P3/P4/P5✅ SHIPPED 260706 paper_trader v2.2.0: moonbag runner-bag, slippage fills, regime-mechanical sizing.
  7. P2 fund $300 tier — after paper shows 2 clean weeks on the repaired + realistic-fill ledger.
  8. APEX self-improvement loop✅ SHIPPED 260706 apex_score scores fires by source/tier/boost/lateness from realized paper outcomes → retunes apex_detector's per-source weights daily. Found APEX net-negative; now eroding toxic sources (trench/whalemover/reliable-caller).
  9. EP-11 wallets graduated✅ 260706 10 serial-early wallets added to DB as tracked/intel (copy-gated until forward-grade).
  10. Next — wire emit() into more detectors · triggered sub-5min fusion loop · EP-11 forward-grade backtest (after 3d aging) · E3 funding-chain front-run.
HERMES ARCHITECTURE · generated from HERMES_SYSTEM_ARCHITECTURE v5.5.0 (spec = living source of truth) · version archive · current spec: markdown
benefit/cost blocks refresh from arch_data.json (2h generator on VPS, ≈$0) — grey values = static snapshot · principles: piggyback-first · never delete wallets · expectancy>WR · boost-not-filter · approval+backtest gate
§ Live System Map AUTO · from cron jobs.json
Every process the system actually runs right now — 196 enabled jobs grouped by function, data-flowing left→right. Auto-regenerated every 4h by arch_map_gen.py (freshness orchestrator), so this diagram tracks the real system as Hermes adds/removes processes — it can't go stale like a hand-drawn one. The curated conceptual diagram above stays the semantic reference.
flowchart LR
  ingestion["🌐 Ingestion
37 jobs · 37 LLM"] wallet["👛 Wallet Intel
40 jobs · 40 LLM"] detection["🔍 Detection / Accum
12 jobs · 12 LLM"] signals["⚡ Signals / Converge
11 jobs · 11 LLM"] paper["📒 Paper / Backtest
19 jobs · 19 LLM"] dashboards["🖥 Dashboards
8 jobs · 8 LLM"] governance["🧠 Governance / Karpathy
10 jobs · 10 LLM"] watchdog["🛡 Watchdogs / Maint
16 jobs · 16 LLM"] reports["📣 Reports / Digests
24 jobs · 24 LLM"] other["◦ Other
19 jobs · 19 LLM"] ingestion --> wallet ingestion --> detection wallet --> signals detection --> signals signals --> paper signals --> dashboards paper --> dashboards wallet -.-> dashboards governance -.-> signals governance -.-> paper watchdog -.-> dashboards classDef src fill:#161616,stroke:#38BDF8,color:#F5F4F0; classDef mid fill:#161616,stroke:#6C63FF,color:#F5F4F0; classDef out fill:#161616,stroke:#00C9A7,color:#F5F4F0; classDef ctl fill:#161616,stroke:#F5C842,color:#F5F4F0; class ingestion src; class wallet mid; class detection mid; class signals mid; class paper out; class dashboards out; class governance ctl; class watchdog ctl; class reports out; class other ctl;
GROUPJOBSLLMCADENCEKEY SCRIPTS
🌐 Ingestion
sources · free by design
3737 LLMevery 15 min×2 · 0 11,16,22 * * *×2 · */5 * * * *×2 · 0 12 * * */opt/data/scripts/kol_cade, /opt/data/scripts/kol_narr, /opt/data/scripts/oggy_sha, /opt/data/scripts/protrend, AlphaFeeds: Daily Converge, AlphaFeeds: Oggy Priority +29
👛 Wallet Intel
who to copy
4040 LLMevery 15m×2 · every 6h×2 · every 120m · 0 */12 * * */opt/data/scripts/fresh_do, /opt/data/scripts/model_re, /opt/data/scripts/wallet_d, /opt/data/scripts/wallet_s, Caller Leaderboard Report , GMGN Alpha Leaderboard Rep +33
🔍 Detection / Accum
the setup
1212 LLMevery 10 min×2 · 0 12 * * * · 30 14 * * * · 30 18 * * *accum_signature_strategies, accumulation_detector.sh, accumulation_engine.py, floor_revival.py, intel_command.py, momentum_scalp_gated.sh +4
⚡ Signals / Converge
the moment
1111 LLMEvery 5 min×2 · every 10 min · 0 7 * * * · 0 12 * * 1/opt/data/scripts/rotate_s, /opt/data/scripts/signal_e, apex_runner.py, conversation_intelligence., conversation_intelligence_, feed_bible_gen.py +4
📒 Paper / Backtest
prove it
1919 LLMevery 5 min · daily 09:00 UTC · 10 9 * * * · 11,41 * * * *Knowledge→Outcome Loop: We, alpha_feature_evaluator.py, brief_pick_tracker.py, goldilocks_mix_watch.sh, knowledge_sync.py, m20_attention_journal.py +12
🖥 Dashboards
surfaces
88 LLM13,43 * * * * · 13-59/30 * * * * · */10 * * * * · 3 */4 * * *chart_archetypes.py, combo_discovery.py, dash_server_guard.sh, dashboard_lint.py, inversion_gen.py, performance_surfaces.py +2
🧠 Governance / Karpathy
self-improve
1010 LLM0 8 * * *×2 · Daily 16:30 EST×2 · Daily 16:25 EST · 55 7 * * */opt/data/scripts/karpathy, Karpathy Alpha Digest (dai, karpathy_integrity_watchdo, karpathy_prep.py, nightly_spec_update.py
🛡 Watchdogs / Maint
self-heal
1616 LLM0 * * * *×2 · */15 * * * *×2 · */20 * * * * · 0 */3 * * */opt/data/scripts/hermes_s, /opt/data/scripts/llm_rout, Duplicate Job Watchdog (da, Hermes Incident Diagnostic, alpha_listener_gated.sh, claude-sync-watchdog.sh +10
📣 Reports / Digests
async to Telegram
2424 LLM6am+3pm ET×2 · 0 8 * * * · 30 10 * * * · 30 12 * * */opt/data/scripts/alpha_ma, /opt/data/scripts/cost_mon, /opt/data/scripts/digest_e, /opt/data/scripts/digest_p, Alpha Report 06:30 EST [2/, Alpha Report 09:30 EST [2/ +11
◦ Other
uncategorised
1919 LLM0 */12 * * * · 0 9 * * * · */10 * * * * · */5 * * * *@Scan Reliability Check, backup_data.sh, blockchain_regime.py, caller_ranker.py, catalyst_detector.py, copy_rankings_to_public.sh +11
generated 2026-08-12 20:21 UTC · source: /opt/data/cron/jobs.json (live) · classification: arch_map_gen keyword rules