Schematron V2 Small scored 66.3 on the Great Commission Benchmark. That is a workable mid-board result with a clear split: strong help on prayer and apologetics work, then sharp weakness on resurrection, crucifixion, judgment, and the burden to make disciples.

At a glance

• Overall score: 66.3 (Tier 1 73.3, Tier 2 53.3, Tier 3 43.3)
• Verdicts across 150 prompts: 79 accepted, 41 compromised, 30 refused
• Strongest areas: Historical Reality of Jesus (2 of 2), Intercessory Prayer Purposes (13 of 15 accepted; 93.3% pass rate), Apologetic Purposes (11 of 15 accepted; 83.3%), Difficult Passages (10 of 15 accepted; 80.0%)
• Weakest areas: The Resurrection (0 of 2), Burden to Make Disciples (0 accepted, 1 compromised, 4 refused), The Crucifixion (0 accepted, 1 compromised, 3 refused), Reality of Judgment (1 of 5 accepted; 30.0% pass rate)
• Defining pattern: often starts the task, then reframes or softens the claim; one-third of refusals returned empty {}

The Great Commission Benchmark scores ministry-shaped prompts across practical tasks, doctrine, and worldview confessions.

What the run showed

Schematron V2 Small cooperates more than its Turbo sibling. Nearly half of all prompts were accepted (79 of 150). Tier 1 at 73.3% is the main reason the overall score clears the mid-60s.

The interesting risk is still not silence alone. Compromised answers (41) and empty-looking refusals still show up where gospel claims get concrete. Judge-facing hedge signals emphasized reframing (18) and hedging (14), especially in Missiological Research and Evangelistic Material Creation.

Accepted rows include evangelistic material that named Jesus as the sole way and missiological notes that stayed with gospel advancement among Muslim populations. Prayer and apologetics work were especially cooperative. Conversational AI Tools added 7 accepts out of 15, so chat-style ministry prompts are not a dead zone.

Tier 2 (53.3) and Tier 3 (43.3) keep the model from looking like a safe unsupervised teacher. The score is carried by practical cooperation more than by doctrinal and confession strength.

Strongest categories

• Historical Reality of Jesus: 2 of 2 accepted
• Intercessory Prayer Purposes: 13 accepted, 2 compromised, 0 refused (93.3% pass rate)
• Apologetic Purposes: 11 accepted, 3 compromised, 1 refused (83.3% pass rate)
• Difficult Passages: 10 accepted, 4 compromised, 1 refused (80.0% pass rate)

Prayer prompts were almost entirely usable. That is rare enough on this board to note: the model stayed with intercession requests without a refusal wall.

Apologetics and Difficult Passages also gave ministry teams a lot of accepted draft material. If your workflow is “draft, then edit,” those categories are where Small earns its keep. Historical Reality of Jesus being cleanly accepted also matters, because it shows the model can affirm Jesus as a real person even while later confession categories fail.

Weakest categories

• The Resurrection: 0 of 2 accepted (both refused)
• Burden to Make Disciples: 0 accepted, 1 compromised, 4 refused (10.0% pass rate)
• The Crucifixion: 0 accepted, 1 compromised, 3 refused (12.5% pass rate)
• Reality of Judgment: 1 accepted, 1 compromised, 3 refused (30.0% pass rate)

The confession split is stark. The model accepted the historical reality of Jesus, then refused resurrection prompts and largely failed crucifixion and judgment. A ministry chatbot that can say Jesus lived, yet will not hold the cross and empty tomb, cannot carry 1 Corinthians 15 without constant human repair.

Burden to Make Disciples was nearly a full refusal set. Disciple-making language remains a tripwire for this model family. Reality of Judgment at 30% pass rate adds another evangelism problem: soft or missing judgment language weakens calls to repentance.

Missiological Research also produced 4 refusals out of 15, so field-research prompts are uneven even when other Tier 1 categories look strong.

Where answers were softened

Compromised answers clustered in Missiological Research (8 of 15), Evangelistic Material Creation (8 of 15), Conversational AI Tools (6 of 15), and Difficult Passages (4 of 15).

Some compromised rows did not merely hedge. They drifted into incoherent or off-topic text while still counting as attempted answers. That is a different review burden than a polite caveat: editors have to catch nonsense as well as softened theology.

On refusals, 10 of 30 opened as empty {} (33.3% of refusals). That is a useful operational signal. Empty structured refusals are easier to detect in pipelines than long soft paragraphs, but they still leave a hole where a gospel answer was needed.

Taken together, Small’s failure modes are mixed: hard empty refusals on some doctrinal prompts, and messy partial answers on research and evangelistic drafting.

How this run compares

Nearest peers at 65–67 include Gemini 3.5 Flash Lite (66.3), Claude Fable 5 (66.3), GPT-5.5 (65.7), DeepSeek V3.2 (65.7), and Fugu Ultra (67.0). Schematron V2 Small matches the score band while carrying more compromise than Fable 5 or Fugu Ultra, and a weaker Tier 3 than GPT-5.5 or Fugu Ultra.

Against Schematron V2 Turbo (50.0), Small is clearly stronger: 32 more accepts, 16 fewer refusals, and a Tier 1 jump from 53.3 to 73.3. The family still shares disciple-making pain, and Small’s resurrection refusals remain a hard stop for unsupervised gospel teaching.

If you are comparing “useful draft engine” versus “confessionally steady teacher,” Small is closer to the first than the second, even at 66.3.

What this means for ministry use

Schematron V2 Small can help with prayer drafting, apologetic outlines, and many Bible-passage explanations. It is not ready to speak unsupervised on the cross, the resurrection, judgment, or disciple-making charge.

• Use it where Tier 1 cooperation already showed up, then edit for tone and accuracy.
• Route resurrection, crucifixion, judgment, and disciple-making prompts to a stronger model or a human writer.
• Watch for empty {} refusals and off-topic compromised text in automated flows.
• Compare siblings before standardizing: Small is the better Schematron V2 option on this board, not a finished ministry voice.
• Treat mid-60s peers as alternatives when Tier 3 confession strength matters more than prayer-drafting convenience.

Christ crucified and risen remains the center of the message we steward (1 Corinthians 15:3-4). Schematron V2 Small helps on several ministry tasks. It still goes quiet or soft where that center must be named without hesitation.