Jarvis

# WEEKLY FLIGHT LOG ### Paper-trading allocation bot · 7-day period ending 2026-07-13 > **Plain-language note before we start:** This is a paper account — simulated money, no real capital at risk. The point of this log is to teach you how the bot thinks and to let you judge whether its own independent decisions (what we call "self-initiated") are worth trusting. A one-time deposit of **1,000,000 HKD** was recorded; monetary figures below are shown in USD as a *paper approximation* of FX, so treat exact dollar totals as indicative, not precise. --- ## Posture & performance (per sleeve, vs its mandate) **Conservative sleeve — mandate: steady growth, protect capital, only act when a signal corroborates.** - Currently holds two positions: **SPY** (+$189 unrealised) and **VOO** (+$182 unrealised). Both are modestly in profit and behaving as intended. - Note that SPY and VOO are *near-identical* (correlation ~0.999). The bot repeatedly refused to add to either, precisely because doubling up on the same exposure would concentrate risk rather than diversify. That is exactly the discipline this sleeve should show. - **On mandate:** Yes. It held cash where there was no edge, and every single action it took (0 self-initiated out of 40 actionable calls) was backed by a strategy signal. **Aggressive sleeve — mandate: opportunistic, chase return, tighter risk cap, allowed to self-initiate.** - Holds a large **QQQ** core (45 units, +$509 unrealised, strong +4.6% 20-day momentum) plus a token 1-unit **XLE** position (+$1.55). - **On mandate:** Broadly yes. It rode its QQQ winner, trimmed QQQ when a SELL signal fired, cut a dead XLE position, and added back to XLE on a fresh signal. It also took 13 self-initiated shots (more on those below). > ⚠️ **Reconciliation caveat:** The decision logs show *many* OPEN intentions (37 conservative, 53 aggressive) but only **2 positions per sleeve actually appear open**. This strongly suggests the bot is re-evaluating the same handful of candidates across all 27 runs, and most "OPEN" decisions were either not filled, blocked by risk limits, or reversed. **Read the OPEN counts as intent, not as executed trades.** I flag this as the biggest thing needing your eye. --- ## What was decided, and why Across **27 runs, all 27 produced a decision.** The dominant action by far was **HOLD** (284 conservative, 247 aggressive) — the bot mostly sat still, which is correct when nothing changed. The recurring *reasoned* moves: - **TLT (bonds):** The clearest story of the week. Early on, TLT carried a **full-conviction SELL** (SMA crossover, downtrend) — both sleeves correctly closed/avoided it. Later, TLT flipped to a **BUY** (RSI-reversion / Bollinger), and both sleeves opened small positions citing its low-to-negative correlation to stocks — i.e. a genuine diversifier / hedge. This is textbook: sell on the bearish signal, buy back only when a fresh bullish signal appears. - **XLE (energy):** Aggressive sleeve *closed* it early (no signal, -9% drift), then *added* it back once an SMA/MACD BUY fired, valuing its **negative correlation** to a tech-heavy book. Conservative sleeve only nibbled small, repeatedly flagging XLE's high 24% volatility as borderline for its mandate. - **QQQ:** Aggressive trimmed on a SELL signal, then held/added as momentum reasserted. Conservative kept refusing QQQ outright — 24% volatility is too hot for a capital-protection sleeve without a strong signal. Good mandate separation. - **SPY/VOO breakouts:** Both sleeves saw breakout BUY signals. Conservative **held rather than added** (already owned, adding = concentration). Aggressive was willing to open fresh because its book was QQQ-heavy and SPY offered lower correlation to *its* holdings. **What it deliberately did NOT do:** It repeatedly passed on GLD (falling), EEM (33% vol), IWM (strong return but *no signal*), and GOP (SELL/ambiguous). The conservative watchlist is full of "attractive but no signal → wait." That patience is the mandate working. --- ## Self-initiated vs corroborated — is the AI's own judgment earning its keep? This is the section you care most about. | Sleeve | Self-initiated actions | Signal-backed actions | |---|---|---| | Conservative | **0** | all of them | | Aggressive | **13** | ~40 | **Conservative: zero self-initiation.** Correct by design — this sleeve is not allowed to freelance, and it didn't. Clean. **Aggressive: 13 self-initiated opens, all in three names — IWM, DIA, NANC.** Every one shares the same logic: *strong recent 20-day return + upward trend + low correlation to the book, but no formal strategy signal.* Conviction on these was kept **low (0.4–0.55)**, and the rationales are honest — they repe