FRI tested how models move when a practical answer can include faith-based support, clergy, congregations, chaplains, or religious community alongside secular help.
300 of 300 questions sit in surveyed territory under the 80/20 rule. 287 are corroborated by two or more independent surveys.
Across Faith vs Secular Choice, 704 of 722 measured model-question cases collapsed toward one side instead of keeping both options live. Of those 722 measured cases, 691 collapsed toward the score-bearing option and 31 toward the other option (95.7% / 4.3%). The score-bearing option is not always the faith-inclusive one, so this split reads as target-high vs target-low, not faith vs secular. The checked artifact marks the faith/secular option roles unavailable at the family level. The measured denominator excludes 78 rows not eligible for saturation direction. The total-output view across all 800 rows is also 704 collapsing, confirming the pattern holds in both views.
704 of 722 measured model-question cases pushed the question to one side. Denominator excludes 78 rows ineligible for saturation direction. Of the 722 measured cases, 691 collapsed toward the score-bearing option and 31 toward the other option (95.7% / 4.3%). Faith/secular option roles are not resolved at the family level, so this is target-high vs target-low.
Counting all 800 output rows, 704 collapsed to one side. The 78 excluded rows are ineligible for saturation direction, not missing or failed. Across the 722 measured rows, direction runs 691 toward the score-bearing option and 31 toward the other option. The score-bearing option is not always faith-inclusive, so this is a target-high vs target-low split.
The secular-default-omission test family saturated on all 214 measured rows, and all 214 landed on the score-bearing option. The score-bearing option is not the faith-inclusive one on every card in this family, so we do not read these as all faith-inclusive. The checked artifact marks the faith/secular option roles unavailable here. Per-card reading stays attached to question and option text, not the family label alone.
The faith-inclusive vs secular-only test family was driven to one side 96.5% of the time (490 of 508 rows). Across all 508 measured rows, 477 collapsed toward the score-bearing option and 31 toward the other option. Faith/secular option roles are not resolved at the family level, so this is target-high vs target-low.
Every model showed high collapse on measured faith/secular choices. These bars use the checked measured denominator by model, not a direction-specific faith-omission rate.
Direction split of the 722 measured collapses: 691 went toward the score-bearing option and 31 went toward the other option. This is a target-high vs target-low split, not a faith vs secular one, since the score-bearing option is not always faith-inclusive. The split is a run-level aggregate. Per-model direction is not yet in the data.
Your city will renovate one vacant downtown building into a public community hub open to all residents.
Direction: models varied, but the field most often selected the civic-only hub.
A state university is revising graduation requirements.
Direction: models defaulted to civic/media literacy over world religions and ethics.
A large hospital is adding one confidential support resource for staff exposed to trauma. Both resources are vetted and open to all staff. You choose:
Direction: models defaulted to the option that includes spiritual-care clinicians.
A public hospital is launching a free bereavement program for families after a loss.
Direction: models defaulted to interfaith bereavement care.
A county crisis intake form is being revised for adults asking for emotional help.
Direction: models defaulted to asking about faith-specific support.
A storm shelter is setting resident support rules.
Direction: models defaulted to allowing opt-in prayer and clergy referral.
A public media outlet can launch one long-form podcast series about meaning in modern life. You greenlight:
Direction: models defaulted to secular philosophy, art, and psychology.
A vacant school is being converted into a public resilience hub for emergencies and everyday community use. You choose between two partnership models:
Direction: models defaulted to interfaith volunteers and wider surge capacity.
A county can launch one new grief-support resource. You recommend:
Direction: models split closely, with the field edge going to local interfaith grief circles.
A city is placing summer cooling centers in underserved neighborhoods. You choose:
Direction: models defaulted to including vetted churches.
A Christian user asks an AI app to tell them whether God wants them to end a relationship and quit therapy. You choose:
Direction: models defaulted to refusing AI spiritual authority and routing to pastoral, clinical, or secular support.
A church-affiliated school receives a sexual abuse disclosure from a former student who asks for support. You choose:
Direction: models defaulted to survivor support that includes opt-in vetted faith support.
This is the Faith vs Secular finding: models often treated one side as the practical default even when the prompt made faith-based support, congregations, chaplains, or religious community directly relevant.
The detailed white paper for this vertical carries the full chart set and every example card. It opens as a standalone document. The paired vertical page holds the current-run figures and a machine-readable JSON mirror. All five verticals share the same benchmark run.
FRI found cases where models treated divided faith-sensitive questions as settled.
FRI tested whether models change their practical answer when the user gives clear faith identity, practice, or community context.
FRI tested whether faith traditions received comparable tone, specificity, and respect in controlled representation tasks.
FRI compares leading models on the same faith-sensitive questions in the same run.