EmpeiriaCoverage
What this dataset covers, what backs it, and where it is thin. 44 markers with 9 reference ranges, 32 conditions and 33 symptoms, in dataset 2.1.0.
Checked by a program, not yet by a person. Every cited range has its
quote matched, word for word, against a saved copy of its source. No person has yet
reviewed any citation, and the citations were transcribed by AI.
How done is it
Each subgoal's definition of done, from the roadmap, tested live.
Overall 88%. A criterion is measured from the
dataset, a gate that passes or fails now, enforced by a test that runs
on every change, or needs a person to judge it. Also as JSON at
/api/v1/coverage.
Look up any condition or symptom
in progress-
About 100 conditions
measured
32 of about 100
-
Every condition cited
measured
32 of 32
-
Every condition explained in plain and clinical language
measured
32 of 32
Find normal levels for any marker
done-
LOINC codes verified
measured
44 of 44
-
At least 80% of reference ranges cited
measured
9 of 9
-
No quality figure has fallen
gate
the ratchet holds
Understand what a substance does and how it is measured
done-
Every marker has its chemistry
measured
44 of 44
-
Every marker has a primary assay that explains why labs differ
measured
44 of 44
Navigate by body
done-
Every marker reports on a part of the body
measured
44 of 44
-
Every organ belongs to a body system
measured
24 of 24
-
No internal link is dead
enforced by a test
every page reachable from the home page is crawled on each test run ·
tests/test_link_crawl.py::test_no_internal_link_is_dead
Assess your own results
waiting on a person-
Lab report import works
enforced by a test
paste a report on My results; uncertain rows are flagged, never guessed ·
tests/test_report_import.py::test_confident_rows_are_read_right -
A result's fasting state is read
enforced by a test
a fasting glucose is judged against the fasting range ·
tests/test_p1_bugs.py::test_fasting_observation_uses_the_fasting_interval -
Every marker-to-organ link cited
measured
90 of 90; each link's tier (primary or contributing) is what its source states, and sets its weight
-
Link tiers checked by a person
needs a person
0 of 90 links signed off
Be trustworthy: every claim traceable
waiting on a person-
The three data gates are required in CI
gate
CI runs `mra audit --strict`, `mra data-quality --check`, `mra verify-changes`
-
No value has moved without its evidence
gate
change control holds
-
Coverage only rises
gate
the ratchet holds
-
Citations checked by a person
needs a person
0 of 457 cited rows signed off; the rest were transcribed or researched by AI and checked only by program (`mra review <key> --by <you>`)
Graphics and interface
| Measure | Done | Note | |
|---|---|---|---|
| Molecules and atoms drawn in 2D | 37/37 | 100% | |
| Molecules and atoms with a 3D model | 35/37 | 95% | no checkable 3D source: cobamamide, methylcobalamin |
| Protein markers with a verified 3D structure | 11/11 | 100% | each chain aligned to its UniProt sequence |
| Our organs and tissues drawn on the body figure | 24/30 | 80% | 46 shapes linked through UBERON |
Interface goals, as the roadmap records them:
- awaiting review Range bar
- done My results
- done Coverage page
- awaiting review "In practice"
- done Body map
- done Line graph
- done 2D molecular structure
- done 2D structures from verified coordinates
- done Baseline: every molecule and atom in 2D and 3D
- done Link every biomarker mention
- done 3D molecular model
- done Header on every page, API docs included
- done Logo pipeline
- awaiting review Caduceus side columns
Over time
Each dataset release's figures, then today's. Releases before 2.0.0 were reconstructed from the project history.
| Measure | 0.1.0 | 1.0.0 | 2.0.0 | 2.1.0 | now | |
|---|---|---|---|---|---|---|
| Reference intervals cited | 0% | 84% | 95% | 97% | 100% | |
| LOINC codes reconciled | 0% | 0% | 0% | 0% | 100% | |
| Markers with plain English | 39% | 39% | 39% | 39% | 39% | |
| Markers with chemistry | 52% | 52% | 100% | 100% | 100% | |
| Markers with an assay method | 52% | 52% | 52% | 100% | 100% | |
| Markers explaining lab disagreement | 39% | 39% | 39% | 100% | 100% | |
| Structures ingested from ChEBI | 100% | 100% | 100% | 100% | 100% | |
| Conditions with a MONDO id | 95% | 95% | 95% | 95% | 84% | |
| Symptoms with an HPO id | 91% | 91% | 91% | 91% | 91% |
How complete the data is
| Measure | Done | What a gap costs | |
|---|---|---|---|
| Reference intervals cited | 9/9 | 100% | The app tells every reader it shows what backs each claim. |
| LOINC codes reconciled | 44/44 | 100% | Lab codes are what an integrator would rely on, and ours are unchecked. |
| Markers with plain English | 17/44 | 39% | Plain language is the differentiator; without it a page is a worse version of what exists elsewhere. |
| Markers with chemistry | 44/44 | 100% | No answer to what the substance is. |
| Markers with an assay method | 44/44 | 100% | No answer to how it is measured. |
| Markers explaining measurement in practice | 44/44 | 100% | No answer to what actually happens: the sample, the preparation, the setting. |
| Markers explaining lab disagreement | 44/44 | 100% | The assay-specific caveat stays a disclaimer instead of a reason. |
| Structures ingested from ChEBI | 17/17 | 100% | Curated ChEBI ids with no structure pulled through. |
| Conditions with a MONDO id | 27/32 | 84% | No external handle to verify the condition against. |
| Symptoms with an HPO id | 30/33 | 91% | Cannot diff our symptom set against the phenotype ontology, so gaps stay invisible. |
Marker by marker
✓ present · – missing. "Ranges cited" counts reference ranges with a source quote; "lab differences" means the page explains why labs print different ranges for this marker.
Not here by design
- Diagnosis. Conditions describe typical patterns; nothing takes your results and names a disease.
- Ranking conditions by likelihood. A sorted list of causes is a diagnosis by another name.
- Your data on our server. Your results stay in your browser and in files you download ().