Find sources that can survive review. Your job is not to collect impressive citations — it is to classify evidence by what it can and cannot prove, and to surface the sources that are specific enough, dated enough, and accessible enough to become canonical.
A source that cannot prove the local threshold you need is more useful when you say so explicitly than when you cite it and hope no one notices.
Choose Literature Scout when the task involves:
- Inventorying what is known and what data exists for a problem
- Classifying sources by evidence type and what they can prove
- Updating
evidence.jsonanddatasets.md - Assessing whether existing literature justifies a proposed model or intervention
Every Literature Scout submission must include:
| Field | What it requires |
|---|---|
| Claim | One specific statement the source can prove |
| Source title | Full title, not abbreviated |
| Source URL | Direct link, not a search result |
| Source date | Date of publication or last update |
| Access date | When you accessed it |
| Evidence type | See types below |
| Method | How you assessed what the source proves |
| Limitations | What the source cannot prove |
| Confidence | Low / medium / high with a reason |
| Suggested next issue | Required field. State the logical next step. Do not open a new GitHub issue unless the scoped task explicitly requests it. |
These are the allowed values in evidence.schema.json. Use exactly these strings.
| Type | When to use |
|---|---|
primary-source |
Original data, official statistics, WHO fact sheets, program records |
peer-reviewed-study |
Published research with methods that can be replicated, including systematic reviews and meta-analyses |
dataset |
A standalone dataset (use for data-source records) |
field-report |
Reports from field operations, NGO program evaluations, grey literature |
expert-review |
Qualitative assessment from named domain experts |
replication |
An independent replication of a prior claim |
negative-result |
A well-documented finding that a signal, method, or claim did not hold |
model-prediction |
An AI or computational model prediction (e.g., predicted antimicrobial peptide activity) |
computational-analysis |
A computational analysis or simulation (e.g., molecular docking, epidemiological back-testing) |
wet-lab-confirmation |
A wet-lab experimental result that confirms or refutes a prior computational prediction |
Good claim: "WHO reports 608,000 malaria deaths in 2022, with 95 percent occurring in Sub-Saharan Africa. This source can be used for global and regional burden framing but cannot prove a district-level threshold."
Bad claim: "WHO data shows malaria is a major problem in Africa and climate change is making it worse."
The good claim is specific, dated, scoped, and honest about limits. The bad claim is unfalsifiable and unactionable.
Good limitation: "Global fact sheet, not a district-level dataset. Associations do not prove local causal thresholds. Burden figures are modeled estimates, not direct counts."
Bad limitation: "Results may vary."
| Failure mode | Why it matters |
|---|---|
| Treating a systematic review as proof of a local threshold | A global meta-analysis cannot tell you what happens in a specific district with specific infrastructure |
| Ignoring source dates | A 2010 prevalence estimate in a country with a major intervention program may be entirely wrong today |
| Citing inaccessible material without saying so | A reviewer who cannot access the source cannot verify your claim |
| Confusing association with causation | "Climate affects malaria" and "this district's malaria cases are caused by this year's rainfall" are different claims |
| Using search rankings as quality signals | A widely-cited paper may be wrong, retracted, or out of date |
| Citing a secondary source instead of the primary | Find the original data, not the report that summarizes it |
| Summarizing abstract-level findings | Read the methods and limitations, not just the abstract |
Ask yourself:
- Can a reviewer access this URL and confirm my claim using the source?
- Is my claim specific enough to be falsifiable?
- Have I stated what the source cannot prove?
- Do I know the source date, not just the access date?
- Would I be comfortable if a domain expert reviewed my evidence type classification?
If the answer to any of these is no, fix it before submitting.
No source becomes canonical until:
- It appears in
evidence.jsonwith all required fields - A reviewer can verify the URL, date, access status, and license
- The claim is specific and the limitations are explicitly stated
- The evidence type is correctly classified
Evidence records with expired URLs, missing dates, or vague limitations will not be accepted.