# Theological Credence Catalog: Documentation and Implementation Guide

**Documentation version:** 1.0  
**Catalog version:** 1.0  
**Prepared:** 2026-08-06  
**Status:** validated release documentation  

**Companion artifacts:**

- [Primary catalog](theological-credence-catalog.md)
- [Machine-ingestible JSONL](theological-credence-catalog.jsonl)
- [Compact TSV index](theological-credence-catalog-index.tsv)
- [Generated release statistics](theological-credence-catalog-stats.json)

## 1. Purpose

The Theological Credence Catalog is a structured inventory of Christian theological questions for which conservative, Scripture-affirming Christians may reach competing reasonable interpretations.

Its purpose is to replace unnecessarily Boolean identity statements with proposition-level judgments that can be expressed as:

- point credences, such as “I assign postmillennialism a 50% probability”;
- comparative odds, such as “I regard postmillennialism as twice as likely as amillennialism”;
- credence ranges, such as “My present confidence in believer-only baptism is between 60% and 75%”;
- conditional judgments, such as “If I became less confident in covenant continuity, my confidence in infant baptism would fall.”

The catalog is designed for use by:

- individuals examining their own theology;
- friends or churches trying to discuss disagreement without hiding conviction;
- teachers explaining how interpretive judgments connect;
- agents generating questionnaires or explanatory material;
- developers building a website, assessment, profile, or theological exploration tool.

The catalog is not a creed, a substitute for exegesis, a spiritual maturity score, or a claim that every listed alternative has equal merit.

## 2. Governing theological framework

The catalog operates within a conservative Christian framework with these assumptions:

1. Scripture is wholly true in the original writings.
2. Faithful copies and translations substantively preserve Scripture’s teaching.
3. Apparent uncertainty belongs to human interpretation, synthesis, textual judgment, or application—not to Scripture’s truthfulness.
4. The canonical text is the controlling evidence for doctrinal conclusions, while historical reception, confessions, philosophy, and experience may inform how evidence is weighed.
5. Historic creedal claims such as one God in three persons, Christ’s full deity and humanity, his bodily resurrection, and his future triumph form the catalog’s Christian boundary rather than a collection of equally weighted denominational alternatives.

A user may therefore say, “I have only 60% confidence in my interpretation of this passage,” without implying, “I think Scripture is only 60% reliable.”

## 3. Terminology

### 3.1 Credence

A credence is a person’s present degree of belief in a proposition. It is commonly represented from 0 to 1 or from 0% to 100%.

Credences are:

- personal but answerable to evidence;
- revisable;
- capable of being internally inconsistent;
- distinct from the objective truth of the proposition;
- distinct from the strength or sincerity of a person’s faith.

### 3.2 Odds

Odds compare two credences.

If postmillennialism is assigned twice the weight of amillennialism, then:

`P(postmillennialism) / P(amillennialism) = 2`

This ratio does not by itself establish either absolute probability. Probability assigned to premillennialism, uncertainty, or hybrid positions still matters.

### 3.3 Credence range

A credence range expresses uncertainty about the exact value of one’s own credence. For example:

`P(postmillennialism) = 45%–60%`

This is the closest implementation of the conversational phrase “confidence interval,” but it is not a frequentist statistical confidence interval produced by repeated sampling.

Recommended public language is:

- “credence”;
- “confidence level” in informal conversation;
- “credence range” when giving an interval;
- “odds” when comparing options.

### 3.4 Evidence cluster

An evidence cluster is a group of passages or observations that depend on substantially the same underlying judgment.

Parallel Gospel accounts should not automatically receive three independent evidential votes. A passage quoted in several records should not receive a fresh likelihood multiplier at every downstream node. Evidence clusters prevent this kind of accidental double-counting.

### 3.5 Interpretive fork

An interpretive fork is the specific question that causes the same text to support different conclusions. Examples include:

- whether prophetic imagery is primarily literal, typological, symbolic, or layered;
- whether a narrative establishes a permanent norm;
- whether a warning describes a regenerate person in danger or a false professor;
- whether covenant promises retain a distinct national referent;
- whether two biblical terms identify different realities or overlapping descriptions.

### 3.6 Hard constraint

A hard constraint identifies propositions that contradict one another as defined. Hard constraints should be rare.

Examples include:

- “Only professing believers are proper baptismal subjects” and “infants are proper baptismal subjects” as single answers to the same exclusive question;
- “Every regenerate person infallibly perseveres” and “some regenerate people finally fall away” as universal claims;
- “There is one public return with no distinct earlier phase” and “a distinct secret return removes the church before the final tribulation.”

### 3.7 Soft correlation

A soft correlation records a historical or hermeneutical tendency without asserting logical entailment.

Examples:

- premillennialism often correlates with futurist readings of Revelation;
- covenant continuity raises the probability of infant baptism;
- charismatic theology does not entail the prosperity gospel;
- postmillennialism does not entail theonomy.

## 4. Artifact inventory

The release contains four supporting files.

### 4.1 `theological-credence-catalog.md`

The primary human-readable catalog.

It contains:

- the framework and probability vocabulary;
- 229 issue records;
- 34 identity and denominational descriptor records;
- text clusters and interpretive forks;
- cross-record mirror guidance;
- 85 high-leverage dependency statements;
- normalization rules;
- denominational calibration sources;
- scope and safety cautions.

This is the best source for reading, theological review, retrieval-augmented generation, and human editorial work.

### 4.2 `theological-credence-catalog.jsonl`

The machine-ingestible projection.

Each line is one independent JSON object. The file contains 360 records:

- 1 metadata record;
- 229 issue records;
- 34 descriptor records;
- 11 mirror-group records;
- 85 dependency records.

JSONL is suitable for:

- streaming ingestion;
- embeddings and retrieval;
- indexing by record type or ID;
- agent context selection;
- loading into a document or graph database.

The issue options and theological detail remain inside `body_markdown`. The JSONL file supplies a reliable record wrapper but is not yet a fully normalized option-level database.

### 4.3 `theological-credence-catalog-index.tsv`

A compact four-column index with:

- `id`;
- `record_type`;
- `domain`;
- `title`.

The file has one header row and 359 data rows. The metadata record is intentionally omitted.

Use the TSV for:

- quick inspection;
- spreadsheet import;
- search menus;
- record selection;
- verifying coverage without loading full bodies.

### 4.4 `theological-credence-catalog-stats.json`

A small generated statistics file containing current counts and Markdown size. It is useful for release verification and automated tests.

## 5. Catalog coverage

The 229 issue records are distributed across seven domains:

- Authority, canon, and hermeneutics: 12
- Theology proper, Christology, and Mariology: 12
- Creation and human constitution: 9
- Moral theology and public theology: 18
- Salvation, grace, justification, and sanctification: 61
- Church, ministry, sacraments, Spirit, worship, and Sabbath: 78
- Eschatology and Israel/church questions: 39

The descriptor section contains 34 prototypes, including:

- broad boundary labels such as Nicene or evangelical;
- denominational families such as Baptist, Lutheran, Anglican, Roman Catholic, and Eastern Orthodox;
- soteriological labels such as Calvinist and Wesleyan;
- eschatological labels such as premillennial, amillennial, postmillennial, and dispensational;
- polity and ministry labels such as Presbyterian and complementarian;
- worship-style labels such as high-church and low-church;
- public-theology labels such as theonomic and two-kingdoms.

The catalog aims for broad recurring coverage rather than recording every disputed verse, historical theologian, local church custom, or modern ethical scenario.

## 6. Stable identifiers and namespaces

Every record has a stable identifier. Identifiers should be treated as permanent keys rather than regenerated from titles.

Major issue namespaces include:

- `AUTH`: authority, canon, confession, and interpretive authority;
- `HERM`: hermeneutics and canonical interpretation;
- `TEXT`: textual questions;
- `GOD`, `TRIN`, `CHRIST`, `MARY`: theology proper, Trinity, Christology, and Mariology;
- `CREA`, `ANTH`, `ANT`: creation and anthropology;
- `ETH`, `PUBLIC`, `MARR`: ethics, public theology, and marriage;
- `ELEC`, `CALL`, `ATN`, `CONV`, `JUS`, `PER`, `SAN`, `LAW`, `NCV`, `PROV`: salvation and covenant-related questions;
- `ECC`, `GOV`, `BAP`, `MEM`, `SAC`, `SUP`, `PNE`, `MIS`, `WOR`, `SAB`, `GEN`: church, government, baptism, membership, sacraments, the Lord’s Supper, Spirit, mission, worship, Sabbath, and sex-related ministry questions;
- `ESC-*`: eschatology, with topic-specific subnamespaces;
- `ID-*`: identity and denominational descriptors;
- `MIRROR-*`: duplicate-domain views of the same underlying question;
- `DEP-*`: cross-record dependency statements;
- `META-*`: catalog metadata.

New records should receive new IDs. Existing IDs should not be reassigned to a different question, even if a display title changes.

## 7. Issue-record model

An issue record represents a defined theological question, not a person or a denomination.

### 7.1 Human-readable fields

The Markdown records use these conceptual fields:

- `ID`: stable identifier;
- `Question`: the proposition-level question being estimated;
- `Type`: how option credences may be normalized;
- `Options`: serious interpretive alternatives;
- `Text clusters` or `Key texts`: primary biblical evidence;
- `Decision logic` or `Interpretive forks`: judgments that move evidence toward one option or another;
- `Dependencies`, `Upstream`, or `Downstream`: relations to other records;
- `Identity links`, `Relations`, or `Booleanized identity labels`: labels commonly associated with the options;
- `Guardrail`: a warning when a position approaches or crosses the catalog’s stated boundary.

Domain-specialist records sometimes use different headings while preserving the same function.

### 7.2 JSONL issue fields

An issue object has:

- `record_type`: always `issue`;
- `id`: stable issue ID;
- `title`: full display title;
- `short_label`: compact display label;
- `question`: explicit user-facing question;
- `domain`: top-level catalog domain;
- `body_markdown`: complete theological record body;
- `option_set_type`: normalization safety classification;
- `mirror_group_ids`: optional list of mirror-group IDs;
- `domain_view_of`: optional canonical issue ID when this is an alternate domain view.

### 7.3 Option-set types

The allowed `option_set_type` values are:

#### `exclusive`

Options are mutually exclusive and jointly exhaustive as presently defined. Point estimates should sum to 100%.

#### `exclusive_with_hybrid`

The principal options compete, but the record contains an explicit hybrid or underdetermined option. Point estimates should still sum to 100%.

#### `multi_select`

Several options may be true together. Each option receives an independent credence. Do not force the set to sum to 100%.

Atonement motifs are a typical example: penal substitution, Christus Victor, moral influence, recapitulation, and sacrificial representation may overlap.

#### `untyped_do_not_normalize`

The source record has not received a sufficiently explicit exclusivity classification.

A product may:

- collect independent credences;
- present the alternatives;
- ask the user whether they view them as exclusive;
- flag the record for later curation.

A product must not force these options to sum to 100%.

#### `typed_freeform_review_required`

The record declares a nonstandard structure such as a spectrum, mixture, layered horizon, or passage-by-passage weighting.

These records require a tailored interface or a later curation pass. They must not be treated as simple radio-button questions.

### 7.4 Current option-type distribution

The current release contains:

- 73 `exclusive_with_hybrid` records;
- 10 `multi_select` records;
- 139 `untyped_do_not_normalize` records;
- 7 `typed_freeform_review_required` records.

The large untyped group is deliberate. It is safer to postpone normalization than to encode false exclusivity.

## 8. Descriptor-record model

A descriptor record represents a label people commonly treat as Boolean, such as “Baptist,” “Calvinist,” “charismatic,” or “amillennial.”

Each descriptor separates:

- `Core`: features close to definitional for the label;
- `Common`: features frequently associated with it;
- `Variable`: questions on which people bearing the label may disagree;
- `False inference warning`: conclusions that must not be inferred merely from the label.

### 8.1 JSONL descriptor fields

A descriptor object has:

- `record_type`: always `descriptor`;
- `id`;
- `title`;
- `domain`;
- `body_markdown`.

### 8.2 Recommended use

Identity descriptors should be computed or discussed after issue-level credences are collected.

A descriptor is best treated as a prototype fit rather than a binary classification:

- core components may receive high weights;
- common components may receive moderate weights;
- variable components should normally receive little or no weight;
- contradictory core components may reduce fit;
- a low fit should not be presented as spiritual deficiency.

The user should be allowed to say:

- “I identify with this label despite an unusual component belief”;
- “My beliefs resemble this prototype, but I do not use the label”;
- “I am between these traditions”;
- “This label is organizational or cultural rather than doctrinal for me.”

## 9. Mirror groups

A mirror group identifies two records that ask substantially the same underlying question from different theological domains or at different levels of granularity.

The current release has 11 mirror groups.

A mirror-group object contains:

- `record_type`: `mirror_group`;
- `id`;
- `canonical_id`;
- `member_ids`;
- `note`.

Issue records participating in a mirror contain `mirror_group_ids`. A noncanonical member also contains `domain_view_of`.

### 9.1 Required ingestion behavior

For each mirror group:

1. Choose the canonical record for primary display or storage.
2. Preserve the alternate record as a domain-specific view.
3. Attach repeated passages to one shared evidence object.
4. Do not count agreement between the two records as independent confirmation.
5. If both records contain differently granular options, create an explicit mapping rather than assuming their option IDs correspond one-to-one.

### 9.2 Related records are not automatically mirrors

Records may be strongly related without being the same question. For example:

- covenant continuity;
- infant baptism;
- Sabbath continuity;
- Israel/church relations;
- Mosaic-law continuity.

These should receive dependency edges or share an upstream latent judgment. They should not be collapsed into one record.

## 10. Dependency records

The 85 dependency records describe high-leverage relations between questions.

A dependency object contains:

- `record_type`: `dependency`;
- `id`;
- `statement_markdown`;
- `expression_format`;
- `machine_resolvable`;
- `option_alias_policy`;
- `node_ids`.

### 10.1 Critical parsing rule

Every current dependency is marked:

`machine_resolvable: false`

The `NODE=value` expressions are human-readable aliases. They are not guaranteed to match the exact option identifiers embedded in each issue body.

An implementation may safely use:

- the dependency ID;
- the prose statement;
- the resolved `node_ids`;
- the stated direction and qualitative strength for display or human review.

It must not execute option-level probability updates until a curated alias map has been created.

Recommended future mapping:

- `dependency_id`;
- `source_option_group_id`;
- `source_member_option_ids`;
- `target_option_group_id`;
- `target_member_option_ids`;
- `relation_type`;
- `strength`;
- `rationale`;
- `shared_evidence_cluster_ids`.

### 10.2 Relation meanings

#### `raises`

The source proposition makes the target more plausible, but does not prove it.

#### `lowers`

The source proposition makes the target less plausible, but does not normally reduce it to zero.

#### `constrains`

The source narrows the defensible target options or changes how they should be formulated.

#### `permits`

The source makes a target compatible without positively supporting it.

#### `hard conflict`

The propositions contradict one another under the definitions used by the catalog.

#### `no entailment`

The records are commonly conflated, but one must not be used as direct evidence for the other.

### 10.3 Cycles are expected

The dependency structure is not a one-way doctrinal tree.

Examples of legitimate feedback include:

- local exegesis shapes a doctrine;
- the developed doctrine changes the prior brought back to related passages;
- authority assumptions affect historical-reception weights;
- ecclesiology affects sacramental interpretation;
- sacramental interpretation affects ecclesiology.

A factor graph, iterative Bayesian model, or transparent rule-assisted conversation fits the catalog better than an acyclic questionnaire that silently treats every answer as independent.

## 11. Probability and normalization rules

### 11.1 Exclusive point estimates

For an exclusive node:

`sum(P(option_i)) = 100%`

If a user supplies only ratios and one fixed value, the remaining mass can be normalized.

Example:

- postmillennialism is twice as likely as amillennialism;
- premillennialism is fixed at 25%;
- no other options are active.

The remaining 75% is divided in a 2:1 ratio:

- postmillennialism: 50%;
- amillennialism: 25%;
- premillennialism: 25%.

### 11.2 Credence ranges

For each option:

`lower_i <= upper_i`

For an exclusive option set, the ranges must be jointly feasible:

`sum(lower_i) <= 100%`

`sum(upper_i) >= 100%`

The application should not require all lower bounds or all upper bounds to sum to exactly 100%. The intervals represent uncertainty about the eventual point assignment.

### 11.3 Multi-select nodes

For a multi-select node, each option receives its own credence.

The total may be below, equal to, or above 100%. No cross-option normalization should occur unless a smaller mutual-exclusion subgroup is explicitly declared.

### 11.4 Untyped and freeform nodes

For `untyped_do_not_normalize` and `typed_freeform_review_required`:

- preserve the user’s independent estimates;
- preserve their rationale;
- display a warning that the option geometry is not yet curated;
- do not silently rescale values;
- permit passage-level or dimension-level judgments when appropriate.

### 11.5 Comparative statements

Store comparative language as a ratio constraint rather than converting it prematurely.

Suggested representation:

```json
{
  "left_option_id": "postmillennial",
  "right_option_id": "amillennial",
  "odds_ratio": 2,
  "user_rationale": "My current reading of Revelation 20 and the kingdom texts"
}
```

### 11.6 Avoiding false precision

A website should make it easy to use:

- coarse verbal bands;
- five- or ten-point increments;
- ranges;
- “leaning” language;
- an underdetermined option;
- a statement of what evidence would change the estimate.

It should not imply that a person who chooses 63% has measured something more objectively than a person who chooses 60%.

## 12. Evidence model

### 12.1 Separate source channels

Recommended evidence channels are:

- `biblical_text`;
- `canonical_pattern`;
- `historical_reception`;
- `confessional_authority`;
- `philosophical_coherence`;
- `lived_experience`.

These channels should remain visible. Historical consensus should not masquerade as an additional biblical passage, and an experience should not be counted as a lexical argument.

### 12.2 Suggested evidence object

A future normalized evidence object may contain:

```json
{
  "evidence_cluster_id": "EVID-EXAMPLE-001",
  "title": "Revelation 20 sequence or recapitulation",
  "source_channel": "biblical_text",
  "passages": ["Revelation 19:11–20:10"],
  "interpretive_judgment": "The visions are sequential",
  "user_confidence": 0.65,
  "supports": ["ESC-MIL-01.PREMIL"],
  "opposes": ["ESC-MIL-01.AMIL", "ESC-MIL-01.POSTMIL"]
}
```

The identifiers in this example are illustrative and are not guaranteed current catalog option IDs.

### 12.3 Double-counting controls

The application should:

1. assign one observation to one evidence cluster;
2. link that cluster to every affected node;
3. avoid multiplying its weight again at each downstream node;
4. merge parallel passages when they depend on the same judgment;
5. disclose when two apparent arguments share a hidden premise.

## 13. Suggested user experience

### 13.1 Stage 1: establish the frame

Ask the user whether they accept the catalog’s Scripture and creedal boundary. If not, the catalog may still be informative, but its priors and guardrails will not represent their framework.

Record separately:

- view of Scripture;
- canon;
- role of tradition;
- role of confessions;
- location of interpretive authority.

### 13.2 Stage 2: elicit upstream judgments

Prioritize high-leverage questions:

- covenant continuity;
- typology and fulfillment;
- apocalyptic symbolism;
- narrative normativity;
- Mosaic-law continuity;
- prophetic horizons;
- use of the Old Testament in the New.

### 13.3 Stage 3: move through theological domains

A recommended order is:

1. authority and hermeneutics;
2. theology proper and anthropology;
3. salvation;
4. church, ministry, and sacraments;
5. Spirit, worship, and Sabbath;
6. eschatology;
7. creation, ethics, and public theology.

The order is a usability recommendation, not a theological ranking.

### 13.4 Stage 4: show dependencies transparently

When an answer affects another record, say so.

Good:

“Your confidence in strong covenant continuity tends to raise the probability of infant baptism, but it does not determine it. Would you like to inspect the baptism texts independently?”

Bad:

“Because you chose covenant theology, your baptism answer has been set to paedobaptism.”

### 13.5 Stage 5: compute descriptor fit last

Only after enough component questions have been answered should the tool say:

- “Your current answers resemble the Reformed Baptist prototype”;
- “You have substantial overlap with both Anglican and confessional Reformed profiles”;
- “Your label fit is uncertain because several core questions are unanswered.”

The tool should distinguish:

- doctrinal resemblance;
- institutional membership;
- self-identification;
- local congregation;
- official denominational teaching.

### 13.6 Stage 6: invite counterfactual reflection

Useful prompts include:

- “Which passage is doing the most work in this conclusion?”
- “Which upstream assumption most influences this estimate?”
- “What evidence would move you by ten percentage points?”
- “If your confidence in this hermeneutical judgment fell, which downstream views would change?”
- “Are you assigning probability to the proposition or to the label?”

## 14. Conversational examples

### 14.1 Simple exclusive question

“At present I am approximately 50% postmillennial, 30% amillennial, and 20% premillennial.”

### 14.2 Odds with unresolved remainder

“I think postmillennialism is twice as likely as amillennialism, but I have not decided how much probability to reserve for premillennialism.”

### 14.3 Credence range

“I put believer-only baptism somewhere between 60% and 75%, because my estimate depends heavily on how I read covenant continuity.”

### 14.4 Multi-select doctrine

“I have high confidence that the atonement is both penal-substitutionary and Christus Victor. I do not treat those motifs as competitors for one hundred percentage points.”

### 14.5 Identity descriptor

“My beliefs currently fit much of the Reformed Baptist prototype, but I am less confident about cessationism and the Sabbath than that label might imply.”

### 14.6 Theological humility without indifference

“I assign one interpretation 90% and another 10%. I think the first is substantially better supported, while acknowledging that my interpretation remains corrigible.”

## 15. Agent behavior contract

An agent using the catalog should follow these rules.

### 15.1 Retrieval

1. Retrieve by issue ID, title, domain, referenced passages, and linked node IDs.
2. Retrieve mirror-group metadata with any mirrored issue.
3. Retrieve relevant dependency statements when explaining downstream effects.
4. Retrieve descriptor records only when the user asks about labels or after component beliefs are known.

### 15.2 Reasoning

The agent should:

- distinguish Scripture from interpretations of Scripture;
- identify the specific interpretive fork;
- present serious alternatives in their strongest recognizable form;
- distinguish entailment from historical correlation;
- treat dependencies as defeasible unless marked hard;
- avoid prooftext counting;
- avoid converting human-readable dependency aliases into executable option updates;
- avoid forcing untyped or freeform nodes into exclusive normalization;
- state when historical, linguistic, philosophical, or confessional evidence is doing work.

### 15.3 Communication

The agent should:

- display the proposition beside every percentage;
- explain odds in plain language;
- allow “underdetermined,” “hybrid,” and “I do not know” responses;
- describe changes as changes in the user’s interpretive confidence;
- avoid implying that low confidence is weak faith;
- avoid presenting descriptor fit as a verdict about orthodoxy or character.

### 15.4 Prohibited shortcuts

The agent must not infer:

- Calvinist means Baptist or paedobaptist;
- premillennial means pretribulational;
- postmillennial means theonomic;
- complementarian means cessationist;
- egalitarian means affirming same-sex ethics;
- charismatic means prosperity-oriented;
- Catholic means one specific election theory;
- belief in a future conversion of Israel determines modern-state policy.

## 16. Suggested normalized application model

The current artifacts are intentionally text-first. A production application may normalize them into the following entities.

### 16.1 `issue`

Suggested fields:

- `issue_id`;
- `title`;
- `question`;
- `domain`;
- `option_set_type`;
- `body_markdown`;
- `status`;
- `version`.

### 16.2 `option`

Suggested fields:

- `option_id`;
- `issue_id`;
- `display_name`;
- `plain_definition`;
- `mutual_exclusion_group`;
- `hybrid_flag`;
- `boundary_note`.

### 16.3 `text_cluster`

Suggested fields:

- `text_cluster_id`;
- `issue_id`;
- `passage_references`;
- `summary`;
- `shared_cluster_id`.

### 16.4 `interpretive_fork`

Suggested fields:

- `fork_id`;
- `issue_id`;
- `question`;
- `affected_option_ids`;
- `notes`.

### 16.5 `dependency`

Suggested fields:

- `dependency_id`;
- `source_issue_ids`;
- `source_option_ids`;
- `target_issue_ids`;
- `target_option_ids`;
- `relation_type`;
- `qualitative_strength`;
- `rationale`;
- `machine_resolvable`;
- `shared_evidence_cluster_ids`.

### 16.6 `descriptor`

Suggested fields:

- `descriptor_id`;
- `title`;
- `kind`;
- `core_components`;
- `common_components`;
- `variable_components`;
- `false_inference_warnings`.

### 16.7 `user_credence`

Suggested fields:

- `user_id`;
- `option_id`;
- `point_credence`;
- `credence_lower`;
- `credence_upper`;
- `rationale`;
- `counterevidence_that_would_move_me`;
- `last_revised`;
- `visibility`.

### 16.8 `user_odds_constraint`

Suggested fields:

- `user_id`;
- `left_option_id`;
- `right_option_id`;
- `odds_ratio`;
- `rationale`;
- `last_revised`.

### 16.9 `user_evidence_judgment`

Suggested fields:

- `user_id`;
- `evidence_cluster_id`;
- `interpretive_judgment`;
- `confidence`;
- `notes`;
- `last_revised`.

### 16.10 `descriptor_fit`

Suggested fields:

- `user_id`;
- `descriptor_id`;
- `computed_fit`;
- `coverage`;
- `contradictory_core_count`;
- `self_identifies`;
- `explanation`;
- `last_computed`.

## 17. Descriptor-fit guidance

A descriptor-fit score should not be a naïve count of matching positions.

A safer method is:

1. distinguish core, common, and variable components;
2. assign the highest weights to definitional core components;
3. assign lower weights to common correlations;
4. exclude variable components from the denominator;
5. report how much of the prototype was actually measured;
6. show the strongest matches and mismatches;
7. permit self-identification to differ from computed resemblance.

Example language:

“Your answered positions have 78% weighted resemblance to this prototype, with 65% of its core dimensions measured. This is a descriptive comparison, not a claim about membership.”

The numeric value in that sentence is illustrative. The catalog does not prescribe one universal scoring formula.

## 18. Data-ingestion procedure

### 18.1 JSONL ingestion

1. Read one line at a time.
2. Parse each line as an independent JSON object.
3. Index by `id`.
4. Partition or filter by `record_type`.
5. Preserve unknown future fields.
6. Reject duplicate IDs.
7. Validate every mirror member and dependency `node_id` against existing issue IDs.
8. Never execute a dependency whose `machine_resolvable` value is false.

### 18.2 Markdown ingestion

Use the Markdown catalog when:

- full theological context matters;
- passages and interpretive forks must be visible;
- an agent needs retrieval context;
- a human is editing or reviewing a record.

Record headings begin with a stable ID followed by a title. Do not infer IDs from titles.

### 18.3 TSV ingestion

Use the TSV as a projection only. It does not contain:

- record bodies;
- option-set classifications;
- mirror memberships;
- dependency node arrays;
- theological rationale.

## 19. Validation invariants

The release was checked against these invariants:

- all JSONL lines parse;
- all record IDs are unique;
- record counts agree across generated artifacts;
- every issue has an explicit question;
- every issue has a valid `option_set_type`;
- every mirror member exists;
- reverse mirror metadata is consistent;
- no issue belongs to multiple current mirror groups;
- every dependency `node_id` resolves to an issue;
- every full node ID named by a dependency appears in its `node_ids`;
- dependency aliases remain marked non-machine-resolvable;
- Markdown headings and JSONL issue/descriptor records agree;
- TSV rows have exactly four columns;
- no record body ends with a stray section separator;
- the primary catalog remains a flat text artifact without embedded diagrams.

Current validated counts are:

- 229 issue records;
- 34 descriptor records;
- 11 mirror groups;
- 85 dependency records;
- 360 total JSONL records including metadata.

## 20. Versioning and extension policy

### 20.1 Adding an issue

When adding a new issue:

1. confirm that it is not already represented;
2. decide whether it is a new question, a finer-grained domain view, or only an option within an existing question;
3. assign a new stable ID;
4. write an explicit question;
5. define the option-set type conservatively;
6. identify text clusters;
7. state the interpretive forks;
8. add upstream and downstream relations;
9. distinguish hard conflicts from soft correlations;
10. add identity links and guardrails where relevant.

### 20.2 Revising an issue

Preserve the existing ID when:

- clarifying prose;
- correcting a passage reference;
- improving option definitions without changing the underlying question;
- adding dependencies or identity cautions.

Create a new ID or versioned migration when the underlying question changes materially.

### 20.3 Deprecation

Do not silently delete a published ID. Mark it deprecated and supply:

- replacement ID;
- migration note;
- reason;
- version in which it changed.

### 20.4 Curating option aliases

Before making dependencies executable:

1. extract every declared option into a stable option table;
2. map dependency display aliases to one or more exact option IDs;
3. verify direction and strength;
4. record whether the link shares evidence with another link;
5. review the mapping theologically;
6. set `machine_resolvable: true` only after validation.

## 21. Limitations

The catalog has important limits:

- It is broad but not exhaustive of every theological dispute.
- It does not assign default probabilities.
- It does not claim that every listed option is equally plausible.
- Many records still require option-level normalization.
- Dependency strengths are qualitative and have not been empirically calibrated.
- Denominational descriptors describe prototypes, not every congregation or individual.
- Catholic and Orthodox positions depend on authority premises that differ from Protestant premises.
- Historical and linguistic questions sometimes require evidence beyond the biblical text itself.
- Ethical application often requires prudential facts not supplied by a doctrinal proposition.
- A probability interface can create false precision if it omits reasons, ranges, and revision history.

## 22. Pastoral, social, and privacy safeguards

A responsible implementation should:

- make profiles private by default;
- let users hide sensitive issue-level answers;
- avoid public leaderboards;
- avoid ranking users by orthodoxy, certainty, or denominational fit;
- distinguish exploratory answers from formal confession;
- allow revision without embarrassment;
- provide a way to explain unusual combinations;
- never treat uncertainty as moral failure;
- never use a naked percentage without displaying the proposition;
- avoid recommending church discipline, membership decisions, or ministerial fitness solely from automated scores.

The catalog is best used to improve truth-seeking, mutual understanding, and the quality of disagreement.

## 23. Denominational calibration

Descriptor prototypes were oriented using official confessional or denominational sources, including:

- [Westminster Confession with Scripture proofs](https://opc.org/WCF-WIP.html), Orthodox Presbyterian Church;
- [Canons of Dort](https://www.crcna.org/welcome/beliefs/confessions/canons-dort), Christian Reformed Church;
- [Baptist Faith and Message 2000](https://bfm.sbc.net/bfm2000/), Southern Baptist Convention;
- [Brief Statement of Doctrinal Position](https://www.lcms.org/about/beliefs/doctrine/brief-statement-of-lcms-doctrinal-position), Lutheran Church—Missouri Synod;
- [*To Be a Christian: An Anglican Catechism*](https://anglicanchurch.net/catechism/), Anglican Church in North America;
- [Articles of Religion](https://www.wesleyan.org/about/articles-of-religion), The Wesleyan Church;
- [Statement of Fundamental Truths](https://ag.org/en/Beliefs/Statement-of-Fundamental-Truths), Assemblies of God USA;
- [Confession of Faith in a Mennonite Perspective](https://www.mennoniteusa.org/who-are-mennonites/what-we-believe/confession-of-faith/), Mennonite Church USA;
- [Catechism of the Catholic Church](https://www.vatican.va/content/catechism/en.html);
- [*The Orthodox Faith*](https://www.oca.org/orthodoxy/the-orthodox-faith), Orthodox Church in America;
- [28 Fundamental Beliefs](https://adventist.org/en/beliefs), Seventh-day Adventist Church.

These sources calibrate label definitions. They are not treated as an additional layer of inspired evidence.

## 24. Recommended minimum viable product

A sound first release can remain simple.

It should:

1. load the JSONL and Markdown artifacts;
2. let users browse by domain;
3. show one explicit question at a time;
4. display options, passages, and interpretive forks;
5. collect a point credence or range;
6. respect the option-set type;
7. store a rationale and “what would change my mind” note;
8. show related issues without automatically changing answers;
9. calculate descriptor resemblance only after component answers exist;
10. export a human-readable profile with propositions beside percentages.

It should postpone:

- automatic dependency propagation;
- executable Bayesian updates;
- authoritative denominational classification;
- public comparison or ranking;
- option-level automation based on the current human-readable aliases.

## 25. Final implementation checklist

Before launch, verify:

- [ ] Scripture’s truth and human interpretive uncertainty are clearly distinguished.
- [ ] The interface uses “credence” or “credence range” accurately.
- [ ] Exclusive nodes normalize to 100%.
- [ ] Multi-select nodes do not normalize across options.
- [ ] Untyped and freeform nodes are not forced into radio-button geometry.
- [ ] Every percentage appears beside a proposition.
- [ ] Passage clusters are not counted repeatedly.
- [ ] Mirror records share evidence.
- [ ] Dependency aliases remain advisory until curated.
- [ ] Hard constraints are limited to genuine contradictions.
- [ ] Descriptor fit is presented as resemblance, not identity or worth.
- [ ] Users can record reasons and counterevidence.
- [ ] Users can revise their estimates.
- [ ] Sensitive answers are private by default.
- [ ] Exported profiles distinguish belief, membership, and self-identification.
- [ ] The product explains its limitations.

## 26. Short glossary

- **Boolean label:** an identity term treated as simply true or false of a person.
- **Canonical pattern:** a theological pattern inferred across multiple biblical books.
- **Credence:** present degree of belief in a proposition.
- **Credence range:** range expressing uncertainty about one’s exact credence.
- **Dependency:** a relationship in which one judgment changes the plausibility or interpretation of another.
- **Descriptor:** a denominational, theological, ecclesial, or stylistic prototype.
- **Domain view:** a second treatment of the same underlying question for another theological domain.
- **Evidence cluster:** passages or observations sharing one underlying evidential judgment.
- **Hard constraint:** direct contradiction under the stated definitions.
- **Interpretive fork:** the reading decision that moves a passage toward different conclusions.
- **Mirror group:** records that represent the same underlying question and must not be double-counted.
- **Multi-select node:** question whose options can be jointly true.
- **Odds:** ratio comparing two credences.
- **Option-set type:** rule governing whether and how option credences should normalize.
- **Prior:** initial plausibility before considering the present evidence item.
- **Prototype fit:** degree to which component beliefs resemble a descriptor.
- **Soft correlation:** tendency for positions to travel together without logical entailment.

## 27. Summary

The catalog’s central principle is simple:

> Hold Scripture as wholly true, define the proposition carefully, identify the interpretive fork, state one’s present credence honestly, and remain willing to revise the human judgment.

The data model exists to make that practice possible without reducing theology to slogans or turning humility into indifference.