Quantitative development: applied workshop

Programming and data structures

Use programming and data structures to make decisions inspectable.

Implement a trade-summary function with boundary checks.

Teaching example: Volume, price and currency units need explicit definitions.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Market data and time

Use market data and time to make decisions inspectable.

Distinguish event time, receive time and time zones.

Teaching example: The latest received record may not have the latest event time.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Numerics and tolerance

Use numerics and tolerance to make decisions inspectable.

Compare floating-point results using a defined tolerance.

Teaching example: The floating-point result of 0.1+0.2 may not equal 0.3 exactly.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Tests for strategy implementation

Use tests for strategy implementation to make decisions inspectable.

Test order states, empty inputs and repeated events.

Teaching example: A duplicate fill event should not double-count the fill.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Performance measurement

Use performance measurement to make decisions inspectable.

Record baseline, input size and environment before comparing optimisation.

Teaching example: One faster run may reflect caching or noise.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Failures, limits and recovery

Use failures, limits and recovery to make decisions inspectable.

Define explicit handling for invalid data and connection failures.

Teaching example: Retries require state and idempotency considerations.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Implementation notes and review

Use implementation notes and review to make decisions inspectable.

Explain model assumptions, interfaces, tests and limitations.

Teaching example: Model validity and code correctness need separate checks.. Predict the result, independently check the key step, and compare these assumptions with the actual role. Do not use demonstration results as real-world advice.

Worked example · fictional

From a problem to inspectable work

Synthetic fills: A is 100 shares at 10, B is 200 at 11, and A is received again; count IDs once, excluding fees.

1. 100×10+200×11=3200; VWAP is 3200/300, not the simple mean of 10 and 11.
2. Further judgment: Can lower latency justify omitting duplicate and sequence checks?
3. Retain correctness invariants and measure latency distributions and recovery

Deliver quant data tool and test notes: the calculation reference is 3200; the recommendation must respect Retain correctness invariants and measure latency distributions and recovery. Separate numbers, inference and unknowns.

Failure case and repair

In Tests for strategy implementation, copy tool output without definitions, then retain the conclusion when asked: Can lower latency justify omitting duplicate and sequence checks?

Even a result of 3200 does not explain when Retain correctness invariants and measure latency distributions and recovery is supported.

100×10+200×11=3200; VWAP is 3200/300, not the simple mean of 10 and 11. Then test assumptions in Failures, limits and recovery.