---
id: lab-008-research-question
title: "Narrow a topic into an answerable research question"
level: L3
domain: research
goal: "Turn a broad topic into a scoped question whose claims can be supported, limited, or left unresolved by inspectable sources"
setup: "A low-risk public topic, at least one accessible primary source, one source of uncertain authority, and a blank claim-evidence table"
task: "Write candidate questions, select one with explicit scope, build a source plan, map claims to evidence, and refuse to fill gaps from an unverifiable source"
evidence:
- "The original topic, candidate questions, selection rationale, and final scoped question"
- "A source plan with authority, publication date, access date, scope, and limitations"
- "A claim-evidence table that separates direct support, inference, disagreement, and unknowns"
failure_variant: "Add an inaccessible, irrelevant, or weak source and check that the conclusion is narrowed rather than completed by invention"
reflection: "Which constraint made the question answerable, which claim has the weakest support, and what would change the conclusion?"
status: draft
last_verified: "Maintainer reference run accepted 2026-08-13; learner run not run"
transfer_task: "Use the same question and evidence protocol for a second low-risk domain"
transfer_domain: "product research, engineering investigation, or content fact-checking"
transfer_evidence: "Keep question revisions, source records, claim mappings, limitations, and the smallest next research action"
transfer_limitations: "Accessible public sources can still be incomplete, stale, biased, or unsuitable for legal, medical, financial, or organizational decisions"
---
# Lab 008: Narrow a topic into an answerable research question
## Learning objective
Research quality is decided before prose begins. A broad topic encourages a
summary; a useful question names the population, object, time window, decision,
and acceptable evidence closely enough that a reviewer can challenge it.
## Setup
Choose a public, low-risk topic. Exclude personal data and legal, medical, or
financial advice. Prepare one primary or authoritative source that is currently
accessible and one candidate source whose authority or accessibility is
uncertain. Record URLs and access dates before taking notes.
## Task and experiment
Write three candidate questions. For each, record:
```text
decision this answer informs:
object or population:
time window:
included / excluded:
required source class:
answer shape:
stop condition:
```
Select one question and explain why the other two are too broad, too vague, or
not answerable with the allowed evidence. Build a source plan before drafting.
Prefer primary sources for product behavior, policy, standards, and statistics;
use secondary sources to locate or compare, not to silently replace the primary
record.
Create a table:
| Claim | Source | Access date | Direct support | Inference | Limitation | Status |
|---|---|---|---|---|---|---|
Statuses are `supported`, `partial`, `disputed`, `unknown`, or `out_of_scope`.
Draft only after every material sentence has a row or is clearly labelled as
analysis. End with the smallest source or experiment that would reduce the most
important remaining uncertainty.
## Evidence to keep
Keep the original topic, all three candidate questions, selection rationale,
source plan, captured excerpts or precise locations, access dates, claim table,
draft, and limitation section. A URL alone is not evidence of what a source
said on the access date.
### Maintainer reference packet
The project includes one frozen, fully offline reference fixture for this Lab.
It presents a current source, a stale source with a scope-matched conflict, an
inaccessible source record, and a fabricated citation candidate. The first
brief overclaims support and must fail. The corrected brief must narrow the
decision, preserve disagreement and unknowns, reject ineligible sources, and
record a stop receipt before it can pass.
See the [frozen fixture](../../examples/lab-008-v1/README.md), the
[executable-example contract](../../docs/governance/executable-examples.yaml),
and the [maintainer packet review](../../docs/quality/lab-008-reference-run-review-2026-08-13.md).
The run is deterministic maintainer evidence over synthetic inputs. It did not
use a model, browse the web, test a learner, or establish transfer or real-world
research quality. This Lab therefore remains `draft`; learner and transfer runs
remain `not_run`.
## Failure case
Add a source that cannot be opened, addresses a different population, or makes
a strong claim without showing its method. The correct response is to mark the
affected claim `unknown`, `partial`, or `out_of_scope`, lower the conclusion's
strength, and name a replacement check. Do not reconstruct missing facts from
the title, search snippet, or confident wording.
## Acceptance checklist
- [ ] The final question names its scope, decision, evidence class, and stop condition.
- [ ] Each material claim maps to a precise source location and access date.
- [ ] Direct support, inference, disagreement, and unknowns remain distinct.
- [ ] The weak-source fixture reduces confidence instead of producing invented support.
- [ ] No private data, unauthorized outreach, purchasing, submission, or publication occurred.
- [ ] The conclusion states what evidence could change it.
## Reflection and transfer
Repeat the protocol in another domain. Which narrowing constraint created the
largest improvement? Which source looked useful but could not support the
actual claim? What remains a hypothesis rather than a finding?