Strategies & Risk

The strategies/ folder is where the desk's discipline is written down. `CLAUDE.md` requires that every proposed trade map to a written rule here, and the Risk Manager reads this folder before clearing any trade. It is not decoration: the caps in these files are the numbers the Risk Manager checks a candidate against, and the entry/exit rules are what the Portfolio Manager must cite when it proposes an order.

The design intent is that you own these limits and the agent never rewrites them. If a cap below is removed, blanked, or left as a TODO, the Risk Manager does not guess a safe value, it VETOes. Guardrails are structural, so the safe failure mode is *no trade*, not an improvised one.

Risk caps (active)

These seven caps live in strategies/README.md and apply to every strategy. The Risk Manager evaluates a candidate order against all of them; a breach of any one is grounds to reject the trade before it ever reaches a preview.

When the Risk Manager must VETO

A VETO is stronger than a flag: it stops the run, and the trade does not proceed to a preview. strategies/README.md lists the conditions under which the Risk Manager must VETO rather than merely raise a concern.

• Any cap above would be breached by the proposed order.
• The entry has no defined stop-loss.
• A cap value is missing, blank, or `TODO`, an unset limit is treated as no permission, not as unlimited.
• The trade would touch any account other than the Agentic account.
• Account data looks inconsistent, or a tool errored unexpectedly, stop and report rather than retry blindly.

These mirror the hard guardrails in `CLAUDE.md`. The Risk Manager is an independent sub-agent with veto power and no order tools, see The Desk Team.

No averaging into losers

Adding to a position that is underwater is forbidden, unless a specific strategy file explicitly permits it *and* defines the add limits. This mirrors `CLAUDE.md`. The only strategy that currently carries that exception is left-side-accumulation.md, and only under its pre-planned ladder, fixed total-risk budget, and whole-position kill-stop. Any add outside those written limits is still forbidden.

Mean Reversion

Thesis: in a stock that is in a confirmed uptrend, short-term oversold dips tend to revert toward the trend. Buy the dip, sell the snap-back, but trade *with* the higher-timeframe trend only, never catch a falling knife in a downtrend. strategies/README.md calls this the right-side-lite setup because it still requires the trend intact plus stabilization before any entry.

Universe: liquid US equities only, average daily volume >=1M shares and price >=$10 (avoid illiquid/penny names). Confirm tradability with get_equity_tradability. Candidates come from a watchlist or the Technical Analyst's scan.

Entry signals (ALL must hold)

1. Uptrend intact: price above its rising ~50-day moving average (higher-timeframe trend is up).
2. Short-term oversold: a clear pullback, RSI(14) <= 30, *or* price tagging the lower end of its recent range / a defined support level.
3. Stabilization: the most recent bar shows the dip slowing, not a vertical breakdown on expanding volume.
4. Backdrop not risk-off: the Macro/News brief is not risk-off, there is no earnings within the next ~10 trading days, and no material adverse headline (no unflagged INJECTION ATTEMPTS).
5. Fundamentals not broken: Fundamental score >= 0, buy dips in healthy names, not falling fundamentals.

Exit signals (any one triggers an exit proposal)

• Target: price reverts to the ~20-day moving average or prior resistance, or RSI(14) >= 55, take profit.
• Time stop: no reversion after 10 trading days, exit; the capital is better used elsewhere.
• Trend break: a daily close back below the ~50-day MA, thesis invalidated, exit.
• Hard stop: the sizing stop, defined below.

Position sizing

Sizing is risk-based and then capped. Here "equity" means NAV (get_portfolio.total_value), per the README, not equity_value. Risk 1% of NAV between entry and the hard stop, convert to shares, then clip the notional to the per-trade and concentration caps.

risk budget  = 1% of NAV, measured between entry and the hard stop
shares       = floor( (0.01 × NAV) / (entry − stop_price) )

then cap the notional at:
  - 15%  per-trade cap
  - 25%  single-symbol concentration cap
and take the SMALLER share count.

hard stop = tighter of { strategy invalidation level , -8% from average entry }
            (never looser than the -8% per-position stop in README.md)

Averaging into a loser is not permitted in this strategy. One entry per setup; if it hits the stop, the trade is over, do not add to recover. On any signal conflict (technical says buy but fundamentals score < 0, or macro is risk-off), stop and surface it rather than overriding a rule.

Left-Side Accumulation

Thesis: for a high-quality name caught in a deep, fear-/macro-driven drawdown, build a position in pre-planned tranches as price falls into a defined value zone, before a confirmed reversal. You accept you won't catch the exact bottom; the goal is a good average cost basis across the zone, sized so the whole position is survivable if you're early. Style: contrarian swing/position, long only, equities only.

Universe (stricter than mean-reversion)

• Quality only: Fundamental score >= +1 (a higher bar than mean-reversion's >= 0). Profitable (positive TTM earnings) or a fortress balance sheet; ADV >= 2M shares; price >= $15. Large/established names.
• The drop must be FEAR, not a broken thesis. Only selloffs driven by macro / sector rotation / sentiment / overdone reaction qualify.
• Do NOT left-side a thesis-breaker: a guidance cut that changes the story, a dividend cut, accounting/fraud/going-concern risk, a structural patent cliff, or a dilution spiral. No falling-knife juniors, no meme/penny names.

The ladder (define ALL of it before buying tranche 1)

Hard limits (the "with limits" the rule requires)

• Total position (sum of all tranches) <= 25% of NAV, the concentration cap. Each individual tranche order <= 15% of NAV, the per-trade cap.
• Total risk on the FULL position <= 2% of NAV. Risk = (planned avg cost basis − kill-stop) × total shares. With a wide kill-stop this 2% cap usually binds before the 25% concentration cap, use the smaller size.
• Counts as one position toward the max-6 open-positions cap; each tranche counts toward the 4-orders/day cap, so spread tranches across levels/days.
• No size beyond the plan, ever. The pre-defined total is the maximum; you cannot "top up" later.

The kill-stop and fundamental kill (whole position)

• Price kill-stop: a hard stop below the deepest tranche (e.g. -8% under T3, or a weekly close below the multi-year support that would mean the bottom-call is simply wrong). If hit -> exit the entire position; no further adds. This is what bounds the strategy.
• Fundamental kill: if the cause of the drop turns into a thesis-breaker while you're building, stop adding and exit, even above the price stop.

Low-touch operating notes

• Fundamental kill comes first. The primary exit is *thesis broken*, not a price tick; the price stop is the backstop.
• Stops trigger on a weekly close, not intraday wicks, ignore noise between reviews.
• Partial fills are the expected case, not a failure. Most setups will only ever fill T1; never "complete" a ladder just to be fully sized.
• Reviews are scheduled, not constant, the desk surfaces a name only when it genuinely qualifies, and you act on the rare approval.
• Two tranches max, fewer, wider, pre-planned adds mean fewer moments you must be present.

Adding new strategies

Momentum, event-driven, and other approaches can be added as new files in strategies/. Keep each one self-contained and testable, a full entry/exit spec plus sizing that references the shared caps in strategies/README.md. The Portfolio Manager must always cite the specific strategy a trade comes from, so a rule that isn't written down cannot be traded.