Stage 0 - framing

Pick a question framework before searching, not after.

deep-research routes every question to one of six frameworks, each with its own required fields. Restating a vague question in framework form catches missing scope before PubMed search strings are built.

Six Frameworks

Question type	Framework	Required fields
Intervention effectiveness/efficacy	PICO	Population, Intervention, Comparator, Outcome
Etiology, risk factors, prognosis	PECO	Population, Exposure, Comparator, Outcome
Prevalence, incidence, association with no comparator	PEO	Population, Exposure, Outcome
Scoping review of an evidence base	PCC	Population, Concept, Context
Qualitative or mixed-methods experience/process	SPIDER	Sample, Phenomenon of Interest, Design, Evaluation, Research type
Service delivery, implementation, evaluation	SPICE	Setting, Perspective, Intervention, Comparison, Evaluation

Default to PICO for a comparator-bearing intervention question and PECO for an etiological question with no intervention. Reach for PEO/PCC/SPIDER/SPICE only when the question genuinely does not fit PICO/PECO's shape - do not force a scoping or qualitative aim into intervention columns just because PICO is the familiar path.

Worked Examples

Vague to PICO

"Does exercise help with depression?"

Population:   Adults with a clinical diagnosis of major depressive disorder
Intervention: Structured aerobic exercise program
Comparator:   Treatment as usual / waitlist control
Outcome:      Depressive symptom severity (standardized scale)

Vague to PECO

"Is screen time bad for kids' sleep?"

Population:  Children and adolescents aged 6-17
Exposure:    Recreational screen time in the two hours before bed
Comparator:  Low or no pre-bed screen time
Outcome:     Sleep onset latency and total sleep duration

Vague to PEO

"How common is burnout among ICU nurses?"

Population: Registered nurses working in intensive care units
Exposure:   ICU nursing role (no comparator group sought)
Outcome:    Burnout prevalence (validated inventory, e.g. Maslach Burnout Inventory)

Vague to PCC (scoping)

"What has been studied about telehealth in rural mental health care?"

Population: Rural and remote populations
Concept:    Telehealth-delivered mental health services
Context:    Primary and community mental health care settings

Vague to SPIDER (qualitative)

"What do patients think about being diagnosed with a chronic illness?"

Sample:                  Adults recently diagnosed with a chronic illness
Phenomenon of Interest:  Lived experience of receiving a chronic-illness diagnosis
Design:                  Interviews, focus groups
Evaluation:              Patient-reported meaning, coping, and identity change
Research type:           Qualitative

Vague to SPICE (service evaluation)

"Does adding a pharmacist to the discharge team reduce readmissions?"

Setting:      Acute hospital discharge process
Perspective:  Hospital system / patient
Intervention: Pharmacist-led discharge medication review
Comparison:   Standard discharge process without pharmacist review
Evaluation:   30-day all-cause readmission rate

Ambiguous fit: an intervention question with no real comparator can be written as PICO with Comparator: "none (single-arm / descriptive)" rather than switched to PEO - keep the comparator column unless the review is explicitly scoping or qualitative.

Recording The Framework

The agent records the chosen framework and fields in config.json at Stage 0, alongside the existing filters:

{
  "question_framework": "PICO",
  "framework_fields": {
    "population": "Adolescents 12-17y with moderate MDD, outpatient",
    "intervention": "Manualized group CBT",
    "comparator": "Waitlist control",
    "outcome": "Depressive symptom severity"
  }
}

Every required field for the chosen framework needs a concrete answer or an explicit "none"/"not applicable" with a one-line reason. If the question does not obviously map to one framework, the agent restates it back using the framework it picked and asks for confirmation before proceeding - never guesses a value the user did not give.

Search Axes Follow The Framework

Framework	Query axes
PICO	population, intervention, comparator (if not "none"), outcome
PECO	population, exposure, comparator (if not "none"), outcome
PEO	population, exposure, outcome
PCC	population/context, concept - broader and less outcome-constrained, no outcome axis
SPIDER	sample, phenomenon of interest, design (qualitative method terms), evaluation
SPICE	setting, perspective, intervention, comparison (if not "none"), evaluation

For PICO, PECO, PEO, and SPICE, at least one query should target null/negative and non-significant results - PCC maps evidence rather than testing an effect, so this rule does not apply there.

Source manual: references/question-frameworks.md. See also appraisal frameworks, which route off the study design produced by whichever question framework you pick.