> For the complete documentation index, see [llms.txt](https://matterhorn-doc.mometic.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://matterhorn-doc.mometic.com/understand-the-scores/evidence-quality.md).

# Evidence quality

Distinguish weak company results from missing Matterhorn evidence, stale facts, coverage gaps, conflicts, and model uncertainty.

An unanswered question and a failed test are different findings. Matterhorn makes that distinction visible, but you still need to read the state attached to a metric.

| State or label                  | Meaning for your research                                                         |
| ------------------------------- | --------------------------------------------------------------------------------- |
| Pass / met                      | The available measurement clears the applicable rule                              |
| Fail / not met                  | A measured result does not clear that rule                                        |
| Neutral / mixed                 | The result lies between stronger and weaker conditions or contains mixed evidence |
| Unavailable / not measurable    | Necessary evidence is absent or cannot support that calculation                   |
| Uncollected                     | The platform has not obtained the required record                                 |
| Stale                           | The observation is too old for its intended use                                   |
| Implausible / conflict / review | A validation concern needs resolution before relying on the value                 |
| Not applicable                  | The measure does not fit this business or context                                 |

The precise labels differ among the checklist, model, and CAN SLIM views. Read the accompanying rationale rather than mapping every blank to the same cause.

## Three different uses of “confidence”

**Standard evidence confidence** summarizes support for evaluated evidence. **ML evidence quality** describes support for the model's inputs. **Analyst confidence** is part of a qualitative commentary judgment. None is a calibrated probability of an investment succeeding.

**Completeness** describes how much required evidence is covered, with weighted categories in the standard system. The ML chart's **input coverage** is family-specific and can include observed absence/count fields. These percentages have different denominators.

## Missingness in standard and ML views

The standard score can redistribute an entirely unmeasured component's weight among measured components. The ML fixed component blend retains neutral midpoints for unavailable inputs, while learned features carry missingness information. A dash in the chart remains a dash; the midpoint used internally does not become an observed fact.

## A practical response to a gap

1. Decide whether the missing measure is central to this company's thesis.
2. Check **Evidence** and **Filings** for a definition, period, or source limitation.
3. Check [freshness](/keep-research-current/refresh-and-freshness.md) before treating absence as permanent.
4. Record the uncertainty or ask for the specific source to be researched.

Missing backlog at a pharmaceutical company is not automatically alarming. Missing cash-flow support in a cash-compounding thesis is more consequential. The research question determines the importance of the gap.

> **Pro Tip — Prefer a narrow, named uncertainty to a vague confidence score.** “The latest backlog comparison mixes current and total obligations” tells you exactly what to resolve.

Related: [evidence and citations](/investigate-the-business/evidence-and-citations.md), [FAQ](/reference/faq.md).