Methodology

Allocate is built for questions where evidence is uneven. This method keeps research readable, sourced, and explicit about confidence.

Why this method exists

Some questions have strong academic coverage. Other questions draw on fragments such as city plans, public records, local reporting, images, testimonies, or direct observation. The method covers both conditions.

The goal is practical research that is:

• readable by non-specialists
• grounded in verifiable sources
• explicit about uncertainty
• useful in places where institutions have not done the work

Allocate treats research as a staged evidence process. A model output is one input to that process.

Evidence stack

Allocate uses a layered evidence stack. Higher-trust sources should inform the baseline. Local questions still require local evidence.

1. Literature: peer-reviewed research, open-access papers, and formal institutional reports.
2. Administrative data: budgets, permits, plans, procurement records, agency dashboards, meeting agendas, collision data, and service maps.
3. Operational evidence: schedules, inspections, maintenance records, implementation status, public notices, and local program materials.
4. Direct observation: field audits, repeated counts, imagery review, route checks, and timestamped measurements.
5. Local testimony: resident, worker, organizer, and community accounts with clear scope and attribution rules.
6. Inference: conclusions drawn by combining multiple cited facts.

Allocate keeps these evidence types distinct. Each claim should preserve its evidence type.

Research flow

Each report should move through the same sequence:

1. Frame the question: define topic, place, time horizon, and decision context.
2. Review what already exists: complete a literature review and open-source baseline.
3. Map the gaps: identify what is missing, stale, disputed, or too coarse.
4. Collect additional evidence: use public data, documents, observation, and local verification.
5. Synthesize claims: state what is known, likely, contested, and unknown.
6. Attach confidence: make the support for each claim visible.
7. Translate into action: show what can be done now and what still needs verification.

This structure matters most where evidence is scarce.

Claim states

Allocate separates the status of a claim from the confidence of a claim.

Claim states:

• observed: directly seen or measured
• reported: stated by a source or witness
• documented: established by a primary artifact or record
• inferred: reasoned from multiple cited facts

Every major claim in a report should indicate its state. Readers can then see when a conclusion depends on synthesis. The report should state when direct proof is unavailable.

Confidence and uncertainty

Confidence measures evidentiary support. It does not measure rhetorical tone.

• Low: sparse support, weak metadata, single-source dependency, or unresolved contradiction.
• Medium: good metadata and partial corroboration with meaningful uncertainty remaining.
• High: multiple independent sources, clear method, low ambiguity, and no major conflict in the record.

Reports should also surface uncertainty directly:

• what is missing
• what may be outdated
• what depends on inference
• what would most improve the conclusion if verified next

Ground-truth loop

Allocate is designed for a feedback loop between desk research and field evidence.

The loop is:

1. Desk research establishes the baseline
2. The system identifies the highest-value unknowns
3. Local contributors gather structured evidence
4. Editors or reviewers validate the submission
5. Findings are updated and confidence changes if warranted

This loop adds local evidence to the institutional record. Source review remains necessary for every claim.

Actionability

Every report should end with a decision-oriented section alongside its conclusion.

The standard output is:

1. What we know
2. What we do not know
3. What is most likely true
4. What is actionable now
5. What evidence would improve the decision most

Recommended actions should be grouped by owner and difficulty:

• immediate operational actions
• policy or governance actions
• capital-intensive actions
• research follow-ups
• local verification tasks

Allocate's output should be usable research: clear about what is known, what remains uncertain, and what someone can do next.