# Speed–accuracy worked examples: methods and claim sheet

Published with the Altamens evidence review **“Faster Is Not Always Better: Reading Speed and Accuracy Together”** on 7 October 2026.

## Provenance

- `worked-examples.csv` contains five deliberately invented session summaries.
- The rows are not people, accounts, trials, game runs or estimates of any population.
- No Altamens customer data, analytics or game telemetry were accessed.
- `calculate.py` uses the Python standard library to recalculate every derived value.

## Definitions

Let `RT` be mean correct-response time in milliseconds and `PC` be proportion correct.

- Accuracy percentage: `100 × PC`.
- Inverse efficiency score (IES): `RT / PC`; lower is conventionally better.
- Rate-correct score (RCS): `PC / (RT / 1000)`; higher is conventionally better and the unit here is correct responses per second.
- Balanced integration score (BIS): `z(PC) − z(RT)`; higher is conventionally better. This sheet uses population standard deviations (`ddof = 0`) across these five example rows only.

IES and RCS are reciprocal transformations of the same two session-level inputs apart from units. BIS depends on the chosen reference rows: adding or removing a row changes the means, standard deviations and therefore all BIS values.

## Reproduction

Run from this directory:

```bash
python3 calculate.py
```

The printed values should match the stored CSV to the displayed precision.

## Appropriate claims

- A session can rank differently when judged by speed alone, accuracy alone or a declared integrated score.
- In these examples, the 60%-accurate fast-guessing-risk row has a slightly better IES and RCS than the 95%-accurate baseline.
- A session that improves both speed and accuracy is less dependent on a compensation rule than a session that trades one for the other.

## Claims these files do not support

- How common any pattern is among players.
- A training, learning, transfer, health, diagnostic or everyday-life benefit.
- A universal choice of metric, cutoff, reliable-change threshold or clinically meaningful difference.
- Comparisons between people, devices, games or task versions.

Research, drafting, calculation checks and publication preparation used an OpenAI automated assistant. The assets received automated consistency checks but no independent human methods review or journal peer review.