Engineering foundations research

The machinery underneath the research: execution realism, borrow and financing replay, corporate-action lifecycle, and the quality contracts that gate what gets published.

Most of what decides whether a research result survives contact with a market is not research. It is whether the borrow was available, whether the corporate action was applied, whether the financing was charged, whether the venue was open, and whether the fill price was one anybody could have got. These documents are about that layer, and it is where more published edges die than in the signal itself.

The failures at this layer share a characteristic that makes them dangerous: they are silent and they flatter. An unapplied split turns a price series into a profit. A fill modelled at the mid earns a spread nobody paid. A backtest that ignores borrow shorts names that could not be shorted. None of these produce an error message; they produce a better number, which is why they survive review by people who are checking the strategy rather than the plumbing.

So each foundation is a contract with a test behind it rather than a description. Execution realism, borrow and financing replay, corporate-action lifecycle and market-status handling are specified. Mutation tests break the guards deliberately to confirm that each check can fail.

The same discipline applies to what these documents admit they cannot do. The cost model's one fully checkable component, crypto commission, matched exactly at 5.00 basis points against a modelled 5.0. The equity components could not be checked because recorded fills carry a padded limit price rather than a decision price. A fill that appears to beat it is beating the padding. That is published as a schema defect with a one-field fix, not smoothed over as a cost estimate.

The engineering foundations documents