Skip to content

03 / Improve

Make the evidence easier to find without changing what it says.

Trace finds terminology gaps, ambiguous abstracts, missing machine-readable structure, and metadata issues. Every recommendation is checked against the underlying evidence before it is shown to you.

The constraint

A suggestion that strengthens a claim is not a suggestion.

Recommendations pass through a fidelity check before they reach you. If a clearer phrasing would also make the finding sound more certain, larger, or more general than the study supports, it is discarded rather than shown with a warning.

Strengthening language

Words like proven, confirms, or superior are rejected when the original text did not already carry them.

Removed qualifiers

Deleting a hedge changes a claim as surely as adding an assertion. Removed scope and uncertainty language is treated as a violation.

Causal overreach

Causal verbs are only permitted where the study design supports a causal reading.

What is reviewed

Four surfaces, checked against the extracted study.

Title clarity

  • Is the condition stated?
  • Is the intervention identifiable?
  • Is the study type clear where it helps?
  • Are acronyms expanded?

Abstract structure

  • Does it cover population, intervention, comparator, and outcome?
  • Are numeric results present?
  • Is the primary endpoint unambiguous?
  • Are absolute and relative effects both reported?
  • Is the conclusion proportional to the results?

Terminology

  • Are accepted subject headings used?
  • Are drug generic, brand, and class terms all present?
  • Are common clinical phrasings represented?

Machine readability

  • Is there a structured summary a system can parse?
  • Are dataset and code links present?
  • Is the trial registration linked?
  • Is there structured markup on a companion page?

How you review it

Accept, edit, or reject — with the evidence in view.

Each recommendation shows the current text, the suggestion, why it helps discoverability, and the fidelity check it passed. Nothing is applied automatically.

Trace does not promise a ranking improvement, and no recommendation carries a predicted uplift. Discoverability is measured before and after, and the change is reported as what it is: a measurement taken after an action, not proof the action caused it.