Confidence is a review signal, not an automatic verdict
Every response starts with full confidence. Dayalogs subtracts points only when enabled signals suggest the response may deserve review.
A 100 score means no enabled quality check reduced confidence. It is not proof that the respondent is human or that the response is authentic.
What the confidence layer does
The confidence layer runs quietly while people answer a survey. It combines several non-blocking quality checks around response behavior, answer consistency and repeated collection patterns.
The respondent does not see a CAPTCHA or an extra task. Researchers see the result later in the admin Results and Confidence tabs.
Important: Dayalogs stores the signals even if you later disable one of the challenges for scoring. That makes the confidence score recomputable instead of destructive.
Privacy note: confidence checks can include behavioral and device-derived quality signals, including hashed technical identifiers. Treat them as survey quality signals and describe them in the survey privacy notice when appropriate.
Response signals and cross-response checks
Dayalogs exposes configurable quality challenges. Some are evaluated from the submitted response itself. Others compare responses across the same survey, so they only become meaningful once there is enough fieldwork volume.
Completion time
Looks for response timing that is unusually far from the expected survey experience.
Open-answer behavior
Looks for open-text behavior that may deserve review in context.
Session behavior
Uses lightweight browser-session signals to flag responses that may not reflect normal attention.
Integrity checks
Uses passive checks that help distinguish ordinary survey use from unusual submission behavior.
Client integrity
Looks for technical request patterns that are unusual for ordinary browser traffic.
Repeated answer patterns
Looks for mechanical answer patterns in batteries while avoiding penalties for non-substantive answers such as don't know.
Cross-response checks look for repeated technical patterns across the survey. They are useful for spotting collection setups that a single-response check cannot see, but they are treated as review signals rather than proof of fraud.
Internal record: Dayalogs stores the underlying quality payload for auditability. It is not another public score; it is the retained record that makes confidence recomputable when settings change.
Configure what counts
Open the survey Confidence tab to enable or disable the available checks. They are enabled by default, but a survey can turn off a check when it is not appropriate for that fieldwork setup.
For example, a kiosk or expo tablet may legitimately produce unusual timing or focus behavior. In that case you can disable the check that creates false positives and recalculate the survey confidence without losing the original signal history.
- Disabled challenges disappear from the Confidence breakdown.
- Disabled challenges stop affecting response confidence in Results.
- The original signals remain stored and can be counted again later.
Read the Results and Confidence tabs
The Results tab shows one compact bar per response. Green means high confidence, orange means review, and red means risky. If a response is below 100, hover the bar to see the active reasons.
The Confidence tab shows the survey-level picture: global score, responses sampled, challenge counters and the current challenge configuration.
Review confidence
From the Confidence tab, or from the equivalent API and MCP tools, you can review confidence for the collected responses. The view uses the survey's active challenge configuration, so saving different challenge settings updates what counts.
API and MCP
External tools can read the same confidence summary without scraping the admin interface.
GET /api/v1/surveys/{slug}/qualityreturns the confidence summary and challenge settings.- Survey stats also include confidence under
stats.quality.
MCP tools use the same structured output so an AI assistant can explain the confidence status with the proper permissions.
Current limits
The current layer is intentionally simple and explainable. It does not try to prove identity, and it does not replace fieldwork judgement or respondent-source controls.
It is probabilistic and continuous, not a gate. Sophisticated or distributed low-volume fraud can avoid some checks, and cross-response checks need enough fieldwork volume before they become useful.
Those can be added later using the same pattern: store the raw signal, persist the normalized penalty, expose the challenge in configuration, and make all survey-level views respect that configuration.