Institutional capstone handbook

Project and boundaries

Manage an initial $1,000,000 virtual multi-strategy portfolio with at least five independent Strategy IDs covering Trend, Carry, Relative Value, Flow, and Event. Two Trend parameter variants are not two categories. Volatility/Arbitrage are optional additions, not substitutes.

Complete everything offline or in simulated ledgers, without accounts, keys, real money, or trading authorization. Live Small, Production, and Scaled names merely rehearse approvals. Evidence, risk, and discipline determine scores; profit is not required.

Finish A–F projects first and preserve the initial protocol against later overwrites. Deliver a full simulated quarter or explicitly labeled “teaching historical replay.” Compressed drills are not real-time operation. Inadequate data/fill/valuation models require more evidence, not invented profits.

Step 1: freeze the charter and roles

Assign research, trading, independent risk, and portfolio approval. Individuals may ask peers/reviewers to act independently; without review, mark “Awaiting review,” not graduated. Record start date, quarter, currency, clocks, and strategy versions.

The following teaching budgets may be revised once before starting with reasons. Later changes require approval, effective times, and backtest impacts, without rewriting earlier violations.

These thresholds standardize teaching scores, not real-fund advice. Gaps may cross them; notional compliance does not ensure margin sufficiency. Reconcile liquid cash/obligations per venue. Two strategies cannot reuse the same reserve.

Step 2: create five complete Strategy Objects

Copy the 23-field template five times. Complete Strategy ID, Name, Market, Instrument, Horizon, Category, Hypothesis, Mechanism, Data, Features, Signal, Entry, Exit, Sizing, Risk, Execution, Cost, Backtest, OOS, Regime, Failure, Capacity, and Status. Include inputs/code/reports or explicit failures. Empty “to finish later” fields cannot enter a portfolio.

This complete structural example lacks real supporting inputs, so Status remains Idea. It cannot request capital. Replace every evidence reference and validate sequentially for actual submission.

Strategy ID: EDU-TREND-001
Name: Daily momentum research v1
Market: Virtual dollar spot market, UTC daily boundary
Instrument: One specified teaching asset, spot, no borrowing
Horizon: Daily decisions; at most one day, reassess next day
Category: Trend / Momentum
Hypothesis: One-day net returns after positive past-20-day returns exceed equal-risk baseline
Mechanism: Staged capital needs may persist; reversals invalidate
Data: Daily bars, actual availability, next executable quotes, fee versions required
Features: Past 20 complete-day returns and volatility; insufficient history produces no signal
Signal: Request long if past return exceeds 0 and inputs are fresh; otherwise flat
Entry: Next executable quote after visibility; timeout cancels and reconciles state
Exit: Next decision, risk trigger, or deadline; stops do not guarantee fill prices
Sizing: Scale within capital/risk budgets, gross-notional and capacity limits
Risk: Project strategy stress-loss cap 20,000; portfolio limits take priority
Execution: Teaching aggressive orders; unfilled remainders are never invented fills
Cost: Actual per-leg fees and arrival deviations, without double-counting impact
Backtest: No executable inputs yet; unverified, requires complete report before promotion
OOS: Freeze train, validation, final-test intervals before research; not yet run
Regime: Past-visible volatility/liquidity only; invalid inputs pause
Failure: No incremental net edge after delay/costs or over-budget stress losses reject
Capacity: Unverified; recompute costs, participation, and stress exits across sizes
Status: Idea; capital not activated, awaiting independent validation and approval

Complete the other four individually. One “fee constant” cannot replace different legs' financing, Funding, spreads, and execution. See Advanced B for eight methods.

Step 3: research-to-lifecycle decisions

Each candidate submits a Memo, attempts, Backtest Report, and final OOS. Promotion requires economic logic, statistics, OOS, robustness, cost, risk, execution, and Regime together. Insufficient evidence retains state.

Mainline: Idea → Research → Testing → Validated → Paper → Shadow → Live Small → Production → Scaled. Last three are simulated approval only. Degraded, Paused, and Retired exits are allowed. Keep retired records; renamed IDs cannot hide failures.

Rehearse one complete promotion request, one rejection, and one paused-to-Testing recovery. Record states, reasons, budgets, approvers, time, and rollback. Five categories need reviewable studies, not forced funded failures. Missing qualified research requires more research; capital may remain cash.

Step 4: capital, capacity, and risk reports

Approve capital first, then calculate net/gross exposure, contributions, factors, and venues. Allocated $700,000 need not be fully traded; $300,000 reserves are not instant inter-venue transfers.

Use consistent windows/costs for correlation and contributions with estimation errors. Without covariance evidence, constrain conservatively through scenarios/exposure rather than inventing precise parity weights. Run three or more sizes per strategy and report normal/stress-exit capacity.

Submit the Risk Report. Reviewers halve any strategy limit; recalculate capital, notional, contributions, and stress PnL, not pie-chart labels alone.

Step 5: at least four stress scenarios

The opening protocol fixes shock size, sequence, and duration. Recalculate from inputs/models. Submit actions as well as losses; “stop promptly” cannot replace executable conditions. Without nonlinear liquidation, option, or freeze models, list unquantified risks and reduce/stop affected limits.

At least one deliberately injected drill breaches limits to test shutdown; it is not permission for actual violations. Track positions and unconfirmed orders afterwards. Independent risk approves restart after reconciliation. Repair/replay failed drills and retain earlier failures.

Step 6: daily ledgers and nine-component attribution

For every trade or fixed-position interval, retain decision, arrival, fill, valuation, quantity, benchmark, costs, flows, and independent net PnL. Split size changes into intervals. Unify currencies and retain conversion sources.

Use Advanced F definitions: Directional, Signal, Carry, Funding, Basis, Execution, Fee, Slippage, Residual. Sum equals independent net PnL. Carry excludes separate Funding/Basis. Residual never automatically becomes Alpha. Investigate errors; specify floating-point tolerance before acceptable ranges, without broad tolerances hiding missing accounts.

Submit daily/strategy totals, cash/valuation reconciliation, and one investigated nonzero Residual. Retain old versions and explanations after corrections rather than overwriting history to appear flawless.

Step 7: quarterly review

Daily checks cover PnL, risk, positions, execution, errors; weekly checks expectancy, win rate, payoffs, regimes, drift; monthly checks Alpha/Beta, drawdowns, correlation, risk-adjusted results; quarterly decisions concern portfolio, capital, research, and structure. Different cadences require different decisions, not copied summaries.

Choose retain, downgrade, pause, or retire per strategy with evidence; submit next-quarter capital and reasons. Score decisions separately from results. Profitable advance-limit violations still fail. Thorough failed research preventing losses can earn research/discipline points.

Use Strategy Review. An independent reviewer inspects one decision without future outcomes, then compares actual outcomes and records differences.

Submission structure and acceptance actions

institutional-project/
  charter.md                Opening tasks, capital, limits, roles
  strategies/               Five-plus full objects and lifecycles
  research/                 Memos, attempts, code, input versions, OOS
  execution/                Parents/children, remainders, TCA, exceptions
  ledger/                   Cash, positions, valuations, fees, nine-component attribution
  risk/                     Allocation, capacity, correlation, stress, shutdown
  reviews/                  Daily/weekly/monthly/quarterly, minutes, approvals
  reproduce.md              Environment, commands, expected outputs, limits

Reviewers trace one random signal per category to availability; rerun a test/cost stress; recompute capital and one day's attribution; replay shutdown/restart; remove a venue and inspect residual risk; and reconcile quarterly decisions with evidence. Missing inputs/code or irreproducible charts cannot replace these actions.

Scores and hard failures

Teaching pass requires at least 80 total and at least 60% of every dimension. Hard requirements: initial $1,000,000, five named categories, complete objects, capital/risk reports, four stresses, nine-component attribution, full quarterly review, reproducible materials, and independent review.

Fabricated inputs/fills, unrepaired future leakage, irreconcilable ledgers, duplicated returns, hidden failures, or unauthorized hard-limit breaches cannot be offset by high returns. Injected stress failures may be discovered and handled correctly. If stops fail, repair/replay before assessment. “Project complete, strategy not promoted” is valid; record the two conclusions separately.