Result Interpretation

Result interpretation compares what happened with what was predicted before the run. A result is not just a number; it is evidence for or against a hypothesis under a stated protocol.

When a result is used publicly, summarize the supported statement with a Claim evidence record.

The basic comparison is:

$$\text{observed }\Delta M\text{vs.}\text{predicted }\Delta M_{\mathrm{expected}}$$

Outcomes

• Confirmed: the result meets the pre-registered success threshold.
• Not confirmed: the result does not meet the threshold.
• Surprising: the direction, size, or failure mode differs from prediction.
• Inconclusive: the run failed, variance is too large, or the protocol is flawed.

Checks

• Was the metric computed on the intended split?
• Did any data, code, or environment drift occur?
• Is the effect larger than run-to-run variance?
• Does the result beat the baseline that matters?
• Is the conclusion limited to the tested setting?
• What should be updated: hypothesis, data, model, metric, or workflow?

Related

• Claim extraction
• Claim evidence record
• Negative result
• Experiment ledger
• Research log
• Error analysis
• Confidence interval
• Statistical significance
• Robustness
• Reproducibility
• Metric