jerry/buzz
1
0
Fork 0
mirror of https://github.com/block/buzz.git synced 2026-11-09 19:02:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat(model-capabilities): humanize Databricks UC model families (#6955)

Browse Source 
Chat-eligible Unity Catalog models from the Gemini, DeepSeek, GLM, Grok,
Llama, Qwen, Gemma, and Inkling families were rendering as raw FQNs such
as `system.ai.gemini-3-5-flash` in the agent model picker. #6918
surfaces these models, but the capability manifest carried no records
for most of them and the registry-label path could not strip their Unity
Catalog or `goose-` prefixes.

I added a display-only `label_family_tokens` list and `databricks_v2`
exact records for the sixteen previously unlabeled endpoint stems. All
26 target FQNs and their `goose-` aliases now resolve to curated labels
through `databricks_registry_label`, while capability resolution
continues to use only the existing `family_tokens` (`claude-`, `gpt-`,
and `kimi-`). This keeps label discovery from changing capability
profiles for unrelated model IDs.

Each new record includes reconciliation metadata against the pinned
[models.dev catalog](https://models.dev/api.json). Nine records adopt
first-party effort evidence that the Databricks MLflow Chat transport
can express: Gemini 3.5 Flash, Gemini 3.5 Flash Lite, Gemini 3.6 Flash,
Gemini 3 Pro Image, DeepSeek V4 Flash, DeepSeek V4 Pro, GLM-5.3 Flash,
Grok 4.6, and Inkling. The remaining seven deliberately retain the
`databricks_v2/concrete_unknown` fallback because upstream evidence is
absent, identifies a non-reasoning model, or provides only
toggle/token-budget controls that `reasoning_effort` cannot represent.

The normative corpus adds the sixteen exact records, four Unity Catalog
alias probes, and two capability-isolation probes. Existing base vectors
remain unchanged, and Rust and TypeScript consume the same strict
manifest and corpus.

Related: #6918

---------

Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
Signed-off-by: Thufir <7ebdb0b67dab08a570b9faf7bbada97535673b4ccaba2cbd546ad3ba84c87fa6@buzz.block.builderlab.xyz>
Co-authored-by: Duncan <dcfd242e557282d7a1e2cf2e6877522682f1e5c6156dc92ca7d90eaedd3b0f95@buzz.block.builderlab.xyz>
Co-authored-by: Thufir <7ebdb0b67dab08a570b9faf7bbada97535673b4ccaba2cbd546ad3ba84c87fa6@buzz.block.builderlab.xyz>
This commit is contained in:
Will Pfleger
2026-08-27 17:25:19 -04:00
committed by GitHub
parent be8d7756d5
commit ae0ecd9a67
5 changed files with 880 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
crates/buzz-agent/src/model_capabilities.rs
View File

@@ -192,6 +192,7 @@ impl ProviderFallbacks {
#[serde(deny_unknown_fields)]
struct Manifest {
family_tokens: Vec<String>,
label_family_tokens: Vec<String>,
family_rules: Vec<FamilyRule>,
databricks_v2_known_models: Vec<String>,
exact_records: Vec<ExactRecord>,
@@ -200,6 +201,9 @@ struct Manifest {
#[serde(rename = "_comment", default)]
#[allow(dead_code)]
comment: Option<String>,
#[serde(rename = "_comment_label_family_tokens", default)]
#[allow(dead_code)]
comment_label_family_tokens: Option<String>,
#[serde(rename = "_comment_databricks_v2_known_models", default)]
#[allow(dead_code)]
comment_known_models: Option<String>,
@@ -410,7 +414,7 @@ pub fn databricks_v2_known_models() -> &'static [String] {
/// record label contract.
pub fn databricks_registry_label(raw_model_id: &str) -> Option<&'static str> {
let m = manifest();
registry_label_for_databricks_records(raw_model_id, &m.exact_records, &m.family_tokens)
registry_label_for_databricks_records(raw_model_id, &m.exact_records, &m.label_family_tokens)
}
fn registry_label_for_databricks_records<'a>(
@@ -452,6 +456,9 @@ fn validate_manifest(m: &Manifest) -> Result<(), String> {
if m.family_tokens.is_empty() {
return Err("family_tokens must be non-empty".to_string());
}
if m.label_family_tokens.is_empty() {
return Err("label_family_tokens must be non-empty".to_string());
}
let check_efforts = |ctx: &str,
efforts: &[ThinkingEffort],
@@ -713,6 +720,30 @@ mod tests {
Q::Vector { id: "boundary-claude-3-digit-run-anthropic-probe", provider: "anthropic", raw_model_id: "claude-35", note: Some("Probes whether the claude-3 prefix binds a longer digit run ('35').") },
Q::Vector { id: "boundary-claude-opus-4-70-anthropic-probe", provider: "anthropic", raw_model_id: "claude-opus-4-70", note: Some("Probes whether the claude-opus-4-7 prefix binds a longer digit run ('70').") },
Q::Vector { id: "boundary-gpt-5-1234-openai-probe", provider: "openai", raw_model_id: "gpt-5-1234", note: Some("Probes a 4-digit run after the gpt-5 stem.") },
Q::Section { group: "Databricks UC model-family humanization probes (#6918 follow-up)", note: Some("Exact-record and UC-FQN strip probes for the Gemini/DeepSeek/GLM/Grok/Llama/Qwen/Gemma/Inkling families surfaced by UC discovery.") },
Q::Vector { id: "dbv2-gemini-3-1-flash-image-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-gemini-3-1-flash-image", note: Some("Probes the Gemini 3.1 Flash Image endpoint record and label.") },
Q::Vector { id: "dbv2-gemini-3-5-flash-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-gemini-3-5-flash", note: Some("Probes the Gemini 3.5 Flash endpoint record and label.") },
Q::Vector { id: "dbv2-gemini-3-5-flash-lite-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-gemini-3-5-flash-lite", note: Some("Probes the Gemini 3.5 Flash Lite endpoint record and label.") },
Q::Vector { id: "dbv2-gemini-3-6-flash-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-gemini-3-6-flash", note: Some("Probes the Gemini 3.6 Flash endpoint record and label.") },
Q::Vector { id: "dbv2-gemini-3-pro-image-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-gemini-3-pro-image", note: Some("Probes the Gemini 3 Pro Image endpoint record and label.") },
Q::Vector { id: "dbv2-deepseek-v4-flash-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-deepseek-v4-flash-0731", note: Some("Probes the DeepSeek V4 Flash endpoint record and label.") },
Q::Vector { id: "dbv2-deepseek-v4-pro-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-deepseek-v4-pro-0813", note: Some("Probes the DeepSeek V4 Pro endpoint record and label.") },
Q::Vector { id: "dbv2-glm-5-3-flash-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-glm-5-3-flash", note: Some("Probes the GLM-5.3 Flash endpoint record and label.") },
Q::Vector { id: "dbv2-grok-4-6-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-grok-4-6", note: Some("Probes the Grok 4.6 endpoint record and label.") },
Q::Vector { id: "dbv2-llama-4-maverick-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-llama-4-maverick", note: Some("Probes the Llama 4 Maverick endpoint record and label.") },
Q::Vector { id: "dbv2-meta-llama-3-1-8b-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-meta-llama-3-1-8b-instruct", note: Some("Probes the meta-llama record; the llama- token strips the meta- prefix identically for record and query.") },
Q::Vector { id: "dbv2-meta-llama-3-3-70b-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-meta-llama-3-3-70b-instruct", note: Some("Probes the meta-llama 3.3 70B record and label.") },
Q::Vector { id: "dbv2-qwen3-next-80b-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-qwen3-next-80b-a3b-instruct", note: Some("Probes the Qwen3 Next 80B record; the bare qwen token strips on a hyphen boundary.") },
Q::Vector { id: "dbv2-qwen35-122b-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-qwen35-122b-a10b", note: Some("Probes the Qwen3.5 122B record; the bare qwen token strips a qwen35 stem with no separator.") },
Q::Vector { id: "dbv2-gemma-3-12b-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-gemma-3-12b", note: Some("Probes the Gemma 3 12B endpoint record and label.") },
Q::Vector { id: "dbv2-inkling-exact-record-probe", provider: "databricks_v2", raw_model_id: "databricks-inkling", note: Some("Probes the Inkling endpoint record and label.") },
Q::Vector { id: "dbv2-uc-fqn-gemini-3-5-flash-strip-probe", provider: "databricks_v2", raw_model_id: "system.ai.gemini-3-5-flash", note: Some("Probes strip parity on a system.ai. UC FQN carrying the gemini- token (resolve carries no label; the alias label path is unit-tested).") },
Q::Vector { id: "dbv2-uc-fqn-meta-llama-strip-probe", provider: "databricks_v2", raw_model_id: "system.ai.meta-llama-3-3-70b-instruct", note: Some("Probes strip parity on a UC FQN where the llama- token strips through meta-.") },
Q::Vector { id: "dbv2-uc-goose-deepseek-strip-probe", provider: "databricks_v2", raw_model_id: "data_workflow_tools.goose.goose-deepseek-v4-pro-0813", note: Some("Probes strip parity on a goose- prefixed UC FQN carrying the deepseek- token.") },
Q::Vector { id: "dbv2-uc-fqn-inkling-strip-probe", provider: "databricks_v2", raw_model_id: "system.ai.inkling", note: Some("Probes strip parity on a UC FQN carrying the bare inkling token.") },
Q::Section { group: "Label/capability token isolation probes (#6955 review pass 1)", note: Some("Pins that label_family_tokens (the UC-humanization superset) never leaks into capability resolve(): capability stripping still uses only claude-/gpt-/kimi-, so a label token appearing before a gpt- marker must NOT displace the gpt-5-pro exact profile.") },
Q::Vector { id: "isolation-openai-gemini-gpt-5-pro-probe", provider: "openai", raw_model_id: "tenant-gemini-gpt-5-pro", note: Some("The gemini- label token must not strip here; capability resolve keeps the gpt-5-pro high-only profile.") },
Q::Vector { id: "isolation-openai-qwenchanted-gpt-5-pro-probe", provider: "openai", raw_model_id: "tenant-qwenchanted-gpt-5-pro", note: Some("The bare qwen label token must not fire mid-segment; capability resolve keeps the gpt-5-pro high-only profile.") },
];
/// A section marker in the generated corpus (`_group` + optional `_note`).
@@ -808,7 +839,7 @@ mod tests {
}
#[test]
fn corpus_has_exactly_113_executable_vectors() {
fn corpus_has_exactly_135_executable_vectors() {
// Locks the vector count so a silent INPUTS edit can't quietly drop
// coverage; must equal the gate in the TS harness
// (modelCapabilitiesCorpus.test.mjs).
@@ -817,7 +848,7 @@ mod tests {
.filter(|q| matches!(q, Q::Vector { .. }))
.count();
assert_eq!(
vectors, 113,
vectors, 135,
"corpus executable-vector count changed; update this gate deliberately"
);
}
@@ -998,6 +1029,38 @@ mod tests {
"alias={alias}"
);
}
// UC-family humanization (#6918 follow-up): the new family tokens let the
// shared UC-FQN and goose- alias forms resolve onto their base records.
for (fqn, label) in [
("system.ai.gemini-3-5-flash", "Gemini 3.5 Flash"),
("system.ai.gemini-3-pro-image", "Gemini 3 Pro Image"),
("system.ai.deepseek-v4-pro-0813", "DeepSeek V4 Pro"),
("system.ai.glm-5-3-flash", "GLM-5.3 Flash"),
("system.ai.grok-4-6", "Grok 4.6"),
("system.ai.llama-4-maverick", "Llama 4 Maverick"),
(
"system.ai.meta-llama-3-3-70b-instruct",
"Llama 3.3 70B Instruct",
),
(
"system.ai.qwen3-next-80b-a3b-instruct",
"Qwen3 Next 80B A3B Instruct",
),
("system.ai.qwen35-122b-a10b", "Qwen3.5 122B A10B"),
("system.ai.gemma-3-12b", "Gemma 3 12B"),
("system.ai.inkling", "Inkling"),
(
"data_workflow_tools.goose.goose-deepseek-v4-flash-0731",
"DeepSeek V4 Flash",
),
(
"data_workflow_tools.goose.goose-glm-5-3-flash",
"GLM-5.3 Flash",
),
("data_workflow_tools.goose.goose-grok-4-6", "Grok 4.6"),
] {
assert_eq!(databricks_registry_label(fqn), Some(label), "fqn={fqn}");
}
// Unknown ids, bare family ids, and blanks remain uncurated.
assert_eq!(databricks_registry_label("custom-unlisted-endpoint"), None);
assert_eq!(databricks_registry_label("gpt-5"), None);

4
desktop/src/features/agents/ui/modelCapabilities.ts
View File

@@ -174,6 +174,7 @@ const ProviderFallbacksSchema = z
export const ManifestSchema = z
.object({
family_tokens: z.array(z.string()).min(1),
label_family_tokens: z.array(z.string()).min(1),
family_rules: z.array(FamilyRuleSchema),
databricks_v2_known_models: z.array(z.string()),
exact_records: z.array(ExactRecordSchema),
@@ -181,6 +182,7 @@ export const ManifestSchema = z
// Root documentation keys; modeled for strict parsing, not read at runtime.
// Mirrors the Rust `Manifest` doc fields under `deny_unknown_fields`.
_comment: z.string().optional(),
_comment_label_family_tokens: z.string().optional(),
_comment_databricks_v2_known_models: z.string().optional(),
_sources: z.record(z.string(), z.string()).optional(),
})
@@ -417,6 +419,6 @@ export function databricksRegistryLabel(rawModelId: string): string | null {
return databricksRegistryLabelForRecords(
rawModelId,
MANIFEST.exact_records,
MANIFEST.family_tokens,
MANIFEST.label_family_tokens,
);
}

33
desktop/src/features/agents/ui/modelCapabilitiesCorpus.test.mjs
View File

@@ -4,6 +4,7 @@ import { fileURLToPath } from "node:url";
import test from "node:test";
import {
databricksRegistryLabel,
databricksRegistryLabelForRecords,
ManifestSchema,
resolveModelCapabilities,
@@ -24,10 +25,10 @@ const corpus = JSON.parse(readFileSync(fileURLToPath(corpusUrl), "utf8"));
// (`_group`) are skipped. Mirrors the Rust corpus filter.
const executable = corpus.filter((entry) => entry.expect != null);
test("corpus has exactly 113 executable vectors", () => {
test("corpus has exactly 135 executable vectors", () => {
// Locks the vector count so a silent corpus edit can't quietly drop coverage;
// must equal the gate in the Rust suite (model_capabilities.rs).
assert.equal(executable.length, 113);
assert.equal(executable.length, 135);
});
test("registry label aliases refuse an unprefixed query", () => {
@@ -44,6 +45,34 @@ test("registry label aliases refuse an unprefixed query", () => {
);
});
test("UC model-family FQNs and goose- aliases humanize onto their base records", () => {
// #6918 follow-up: the shared UC-FQN (`system.ai.…`) and goose- alias forms
// must resolve onto the same base databricks_v2 records via the new family
// tokens. Mirrors the Rust `test_databricks_registry_label_lookup` coverage.
const cases = [
["system.ai.gemini-3-5-flash", "Gemini 3.5 Flash"],
["system.ai.gemini-3-pro-image", "Gemini 3 Pro Image"],
["system.ai.deepseek-v4-pro-0813", "DeepSeek V4 Pro"],
["system.ai.glm-5-3-flash", "GLM-5.3 Flash"],
["system.ai.grok-4-6", "Grok 4.6"],
["system.ai.llama-4-maverick", "Llama 4 Maverick"],
["system.ai.meta-llama-3-3-70b-instruct", "Llama 3.3 70B Instruct"],
["system.ai.qwen3-next-80b-a3b-instruct", "Qwen3 Next 80B A3B Instruct"],
["system.ai.qwen35-122b-a10b", "Qwen3.5 122B A10B"],
["system.ai.gemma-3-12b", "Gemma 3 12B"],
["system.ai.inkling", "Inkling"],
[
"data_workflow_tools.goose.goose-deepseek-v4-flash-0731",
"DeepSeek V4 Flash",
],
["data_workflow_tools.goose.goose-glm-5-3-flash", "GLM-5.3 Flash"],
["data_workflow_tools.goose.goose-grok-4-6", "Grok 4.6"],
];
for (const [fqn, label] of cases) {
assert.equal(databricksRegistryLabel(fqn), label, `fqn=${fqn}`);
}
});
test("registry label aliases refuse ambiguous stripped record keys", () => {
const records = [
{

343
scripts/model-capabilities.json
View File

@@ -11,6 +11,20 @@
"gpt-",
"kimi-"
],
"_comment_label_family_tokens": "Prefix tokens used ONLY by databricks_registry_label (and its TS mirror) to strip catalog/endpoint prefixes off Unity Catalog FQN and goose- alias ids before matching a curated display label. Deliberately SEPARATE from family_tokens: label stripping is a discovery-only display concern, while family_tokens drives capability resolution in resolve() for every provider. Adding a family marker here humanizes its UC ids without altering any model's resolved capability profile.",
"label_family_tokens": [
"claude-",
"gpt-",
"kimi-",
"gemini-",
"deepseek-",
"glm-",
"grok-",
"llama-",
"qwen",
"gemma-",
"inkling"
],
"family_rules": [
{
"id": "anthropic-manual-budget-claude3",
@@ -880,6 +894,335 @@
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-1-flash-image",
"registry_label": "Gemini 3.1 Flash Image",
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels",
"_reconciliation": "keep-fallback",
"_reconciliation_note": "No model-level entry for gemini-3-1-flash-image exists under any provider in the pinned models.dev payload; there is genuinely no upstream reasoning evidence to reconcile. Retain the databricks_v2 concrete_unknown fallback profile.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): no providers.*.models[*] key matches gemini-3-1-flash-image."
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-5-flash",
"registry_label": "Gemini 3.5 Flash",
"thinking_mode": "none",
"supported_efforts": [
"minimal",
"low",
"medium",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party google model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev google reasoning_options effort: minimal|low|medium|high",
"_reconciliation": "adopt",
"_reconciliation_note": "The Unity Catalog endpoint uses the hyphenated stem gemini-3-5-flash, while the pinned models.dev first-party google entry uses the dotted model id gemini-3.5-flash and advertises effort [minimal, low, medium, high]. Per the Kimi-K3 precedent, adopt this model-level effort evidence despite the spelling difference and the exact endpoint being absent from the models.dev databricks catalog. Databricks serves via MLflow Chat (reasoning_effort only), so thinking_mode stays none and default_effort is null.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.google.models[\"gemini-3.5-flash\"].reasoning_options=[{\"type\":\"effort\",\"values\":[\"minimal\",\"low\",\"medium\",\"high\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-5-flash-lite",
"registry_label": "Gemini 3.5 Flash Lite",
"thinking_mode": "none",
"supported_efforts": [
"minimal",
"low",
"medium",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party google model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev google reasoning_options effort: minimal|low|medium|high",
"_reconciliation": "adopt",
"_reconciliation_note": "The Unity Catalog endpoint uses the hyphenated stem gemini-3-5-flash-lite, while the pinned models.dev first-party google entry uses the dotted model id gemini-3.5-flash-lite and advertises effort [minimal, low, medium, high]. Per the Kimi-K3 precedent, adopt this model-level effort evidence despite the spelling difference and the exact endpoint being absent from the models.dev databricks catalog. Databricks serves via MLflow Chat (reasoning_effort only), so thinking_mode stays none and default_effort is null.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.google.models[\"gemini-3.5-flash-lite\"].reasoning_options=[{\"type\":\"effort\",\"values\":[\"minimal\",\"low\",\"medium\",\"high\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-6-flash",
"registry_label": "Gemini 3.6 Flash",
"thinking_mode": "none",
"supported_efforts": [
"minimal",
"low",
"medium",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party google model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev google reasoning_options effort: minimal|low|medium|high",
"_reconciliation": "adopt",
"_reconciliation_note": "The Unity Catalog endpoint uses the hyphenated stem gemini-3-6-flash, while the pinned models.dev first-party google entry uses the dotted model id gemini-3.6-flash and advertises effort [minimal, low, medium, high]. Per the Kimi-K3 precedent, adopt this model-level effort evidence despite the spelling difference and the exact endpoint being absent from the models.dev databricks catalog. Databricks serves via MLflow Chat (reasoning_effort only), so thinking_mode stays none and default_effort is null.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.google.models[\"gemini-3.6-flash\"].reasoning_options=[{\"type\":\"effort\",\"values\":[\"minimal\",\"low\",\"medium\",\"high\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-pro-image",
"registry_label": "Gemini 3 Pro Image",
"thinking_mode": "none",
"supported_efforts": [
"low",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party google model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev google reasoning_options effort: low|high",
"_reconciliation": "adopt",
"_reconciliation_note": "The exact databricks endpoint is absent from the models.dev databricks catalog, but the first-party google entry advertises effort [low, high]. Per the Kimi-K3 precedent, adopt model-level effort evidence despite endpoint absence. Databricks serves via MLflow Chat (reasoning_effort only), so thinking_mode stays none and default_effort is null.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.google.models[\"gemini-3-pro-image\"].reasoning_options=[{\"type\":\"effort\",\"values\":[\"low\",\"high\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-deepseek-v4-flash-0731",
"registry_label": "DeepSeek V4 Flash",
"thinking_mode": "none",
"supported_efforts": [
"low",
"high",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party deepseek model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev deepseek reasoning_options effort: low|high|max",
"_reconciliation": "adopt",
"_reconciliation_note": "The exact databricks endpoint is absent from the models.dev databricks catalog, but the first-party deepseek entry advertises a reasoning toggle plus effort [low, high, max]. Per the Kimi-K3 precedent, adopt the model-level effort list despite endpoint absence. The toggle is not representable on the MLflow Chat request schema (reasoning_effort only), so thinking_mode stays none and default_effort is null — mirroring the kimi-k3 record.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.deepseek.models[\"deepseek-v4-flash\"].reasoning_options=[{\"type\":\"toggle\"},{\"type\":\"effort\",\"values\":[\"low\",\"high\",\"max\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-deepseek-v4-pro-0813",
"registry_label": "DeepSeek V4 Pro",
"thinking_mode": "none",
"supported_efforts": [
"high",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party deepseek model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev deepseek reasoning_options effort: high|max",
"_reconciliation": "adopt",
"_reconciliation_note": "The exact databricks endpoint is absent from the models.dev databricks catalog, but the first-party deepseek entry advertises a reasoning toggle plus effort [high, max]. Per the Kimi-K3 precedent, adopt the model-level effort list despite endpoint absence. The toggle is not representable on the MLflow Chat request schema (reasoning_effort only), so thinking_mode stays none and default_effort is null — mirroring the kimi-k3 record.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.deepseek.models[\"deepseek-v4-pro\"].reasoning_options=[{\"type\":\"toggle\"},{\"type\":\"effort\",\"values\":[\"high\",\"max\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-glm-5-3-flash",
"registry_label": "GLM-5.3 Flash",
"thinking_mode": "none",
"supported_efforts": [
"low",
"high",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party Z.AI model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev zai reasoning_options effort: low|high|max",
"_reconciliation": "adopt",
"_reconciliation_note": "The exact databricks endpoint is absent from the models.dev databricks catalog, but the first-party Z.AI entry glm-5.3-flash advertises effort [low, high, max]. Per the Kimi-K3 precedent, adopt the model-level effort list despite endpoint absence. Databricks serves via MLflow Chat (reasoning_effort only), so thinking_mode stays none and default_effort is null.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.zai.models[\"glm-5.3-flash\"].reasoning_options=[{\"type\":\"effort\",\"values\":[\"low\",\"high\",\"max\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-grok-4-6",
"registry_label": "Grok 4.6",
"thinking_mode": "none",
"supported_efforts": [
"low",
"medium",
"high",
"xhigh"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party xAI model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev xai reasoning_options effort: low|medium|high|xhigh",
"_reconciliation": "adopt",
"_reconciliation_note": "The exact databricks endpoint is absent from the models.dev databricks catalog, but the first-party xAI entry grok-4.6 advertises effort [low, medium, high, xhigh]. Per the Kimi-K3 precedent, adopt the model-level effort list despite endpoint absence. Databricks serves via MLflow Chat (reasoning_effort only), so thinking_mode stays none and default_effort is null.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.xai.models[\"grok-4.6\"].reasoning_options=[{\"type\":\"effort\",\"values\":[\"low\",\"medium\",\"high\",\"xhigh\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-llama-4-maverick",
"registry_label": "Llama 4 Maverick",
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels",
"_reconciliation": "keep-fallback",
"_reconciliation_note": "Llama 4 Maverick is a non-reasoning model: all 26 model-level records across providers in the pinned payload report reasoning=false with no reasoning_options. There is no effort evidence to adopt. Retain the databricks_v2 concrete_unknown fallback profile.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): every providers.*.models[*] whose id contains \"llama-4-maverick\" (e.g. meta-llama/Llama-4-Maverick-17B-128E-Instruct-FP8) has reasoning=false, reasoning_options=null."
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-meta-llama-3-1-8b-instruct",
"registry_label": "Llama 3.1 8B Instruct",
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels",
"_reconciliation": "keep-fallback",
"_reconciliation_note": "Llama 3.1 8B Instruct is a non-reasoning model: the consensus across model-level records in the pinned payload is reasoning=false with no reasoning_options (a single reseller, pioneer, is an outlier advertising effort [low, medium, high]). There is no first-party or consensus effort evidence to adopt. Retain the databricks_v2 concrete_unknown fallback profile.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.*.models[*] ids containing \"llama-3.1-8b\" report reasoning=false with reasoning_options=null across essentially all providers (huggingface, nvidia, groq, openrouter, kilo, ...); only pioneer/meta-llama/Llama-3.1-8B-Instruct is reasoning=true."
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-meta-llama-3-3-70b-instruct",
"registry_label": "Llama 3.3 70B Instruct",
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels",
"_reconciliation": "keep-fallback",
"_reconciliation_note": "Llama 3.3 70B Instruct is a non-reasoning model: every model-level record across ~50 providers in the pinned payload reports reasoning=false with no effort values (one reseller, berget, sets reasoning=true but with an empty reasoning_options set, i.e. no effort list). There is no effort evidence to adopt. Retain the databricks_v2 concrete_unknown fallback profile.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.*.models[*] ids containing \"llama-3.3-70b\" report reasoning=false, reasoning_options=null across all providers; providers.berget.models[\"meta-llama/Llama-3.3-70B-Instruct\"] is reasoning=true with reasoning_options=[]."
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-qwen3-next-80b-a3b-instruct",
"registry_label": "Qwen3 Next 80B A3B Instruct",
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels",
"_reconciliation": "keep-fallback",
"_reconciliation_note": "The first-party alibaba entry qwen3-next-80b-a3b-instruct reports reasoning=false with no reasoning_options — the -instruct variant is the non-reasoning sibling of qwen3-next-80b-a3b-thinking. There is no effort evidence to adopt for the instruct model. Retain the databricks_v2 concrete_unknown fallback profile.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.alibaba.models[\"qwen3-next-80b-a3b-instruct\"].reasoning=false, reasoning_options=null (contrast qwen3-next-80b-a3b-thinking which is reasoning=true with budget_tokens)."
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-qwen35-122b-a10b",
"registry_label": "Qwen3.5 122B A10B",
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels",
"_reconciliation": "keep-fallback",
"_reconciliation_note": "The first-party alibaba entry qwen3.5-122b-a10b expresses reasoning only as a toggle plus budget_tokens — NOT an effort-level selector. Neither axis is representable on the Databricks MLflow Chat transport (reasoning_effort only), mirroring the claude-opus-4-7 budget_tokens precedent where no effort divergence is reconciled. Resellers that do advertise effort sets conflict sharply (kilo [none, high]; tensorx [none..max]; neon [none, low, medium, high]), so there is no consistent effort list to adopt. Retain the databricks_v2 concrete_unknown fallback profile.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.alibaba.models[\"qwen3.5-122b-a10b\"].reasoning_options=[{\"type\":\"toggle\"},{\"type\":\"budget_tokens\"}]; reseller effort sets diverge (kilo=[none,high], tensorx=[none,minimal,low,medium,high,xhigh,max], neon=[none,low,medium,high])."
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-gemma-3-12b",
"registry_label": "Gemma 3 12B",
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"_provenance": "all axes materialized from provider fallback (databricks_v2/concrete_unknown)",
"_source": "registry_labels",
"_reconciliation": "keep-fallback",
"_reconciliation_note": "Gemma 3 12B is a non-reasoning model: every model-level record across providers in the pinned payload reports reasoning=false with no reasoning_options. There is no effort evidence to adopt. Retain the databricks_v2 concrete_unknown fallback profile.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.*.models[*] ids containing \"gemma-3-12b\" (e.g. google/gemma-3-12b-it) all report reasoning=false, reasoning_options=null."
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-inkling",
"registry_label": "Inkling",
"thinking_mode": "none",
"supported_efforts": [
"low",
"medium",
"high",
"xhigh",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"_provenance": "supported_efforts+thinking_mode+default_effort+normalization_policy: adopted from models.dev first-party Thinking Machines model-level evidence (Kimi-K3 precedent); databricks_v2_wire_route+label: registry_labels",
"source": "models.dev thinkingmachines reasoning_options effort: low|medium|high|xhigh|max",
"_reconciliation": "adopt",
"_reconciliation_note": "The exact databricks endpoint is absent from the models.dev databricks catalog, but the first-party Thinking Machines entry Inkling advertises a reasoning toggle plus effort [low, medium, high, xhigh, max]. Per the Kimi-K3 precedent, adopt the model-level effort list despite endpoint absence. The toggle is not representable on the MLflow Chat request schema (reasoning_effort only), so thinking_mode stays none and default_effort is null — mirroring the kimi-k3 record.",
"_reconciliation_doc": "https://models.dev/api.json (pinned SHA-256 9266003029c4ea265e923637826211f597db08e8f0f40123aad8547411029238): providers.thinkingmachines.models[\"thinkingmachines/Inkling\"].reasoning_options=[{\"type\":\"toggle\"},{\"type\":\"effort\",\"values\":[\"low\",\"medium\",\"high\",\"xhigh\",\"max\"]}]"
},
{
"provider": "databricks_v2",
"raw_model_id": "databricks-gpt-5",

437
scripts/normative-corpus.json
View File

@@ -2266,5 +2266,442 @@
"normalization_policy": "openai-standard",
"registry_label": null
}
},
{
"_group": "Databricks UC model-family humanization probes (#6918 follow-up)",
"_note": "Exact-record and UC-FQN strip probes for the Gemini/DeepSeek/GLM/Grok/Llama/Qwen/Gemma/Inkling families surfaced by UC discovery."
},
{
"id": "dbv2-gemini-3-1-flash-image-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-1-flash-image",
"_note": "Probes the Gemini 3.1 Flash Image endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": "Gemini 3.1 Flash Image"
}
},
{
"id": "dbv2-gemini-3-5-flash-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-5-flash",
"_note": "Probes the Gemini 3.5 Flash endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"minimal",
"low",
"medium",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "Gemini 3.5 Flash"
}
},
{
"id": "dbv2-gemini-3-5-flash-lite-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-5-flash-lite",
"_note": "Probes the Gemini 3.5 Flash Lite endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"minimal",
"low",
"medium",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "Gemini 3.5 Flash Lite"
}
},
{
"id": "dbv2-gemini-3-6-flash-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-6-flash",
"_note": "Probes the Gemini 3.6 Flash endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"minimal",
"low",
"medium",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "Gemini 3.6 Flash"
}
},
{
"id": "dbv2-gemini-3-pro-image-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-gemini-3-pro-image",
"_note": "Probes the Gemini 3 Pro Image endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"low",
"high"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "Gemini 3 Pro Image"
}
},
{
"id": "dbv2-deepseek-v4-flash-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-deepseek-v4-flash-0731",
"_note": "Probes the DeepSeek V4 Flash endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"low",
"high",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "DeepSeek V4 Flash"
}
},
{
"id": "dbv2-deepseek-v4-pro-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-deepseek-v4-pro-0813",
"_note": "Probes the DeepSeek V4 Pro endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"high",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "DeepSeek V4 Pro"
}
},
{
"id": "dbv2-glm-5-3-flash-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-glm-5-3-flash",
"_note": "Probes the GLM-5.3 Flash endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"low",
"high",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "GLM-5.3 Flash"
}
},
{
"id": "dbv2-grok-4-6-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-grok-4-6",
"_note": "Probes the Grok 4.6 endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"low",
"medium",
"high",
"xhigh"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "Grok 4.6"
}
},
{
"id": "dbv2-llama-4-maverick-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-llama-4-maverick",
"_note": "Probes the Llama 4 Maverick endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": "Llama 4 Maverick"
}
},
{
"id": "dbv2-meta-llama-3-1-8b-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-meta-llama-3-1-8b-instruct",
"_note": "Probes the meta-llama record; the llama- token strips the meta- prefix identically for record and query.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": "Llama 3.1 8B Instruct"
}
},
{
"id": "dbv2-meta-llama-3-3-70b-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-meta-llama-3-3-70b-instruct",
"_note": "Probes the meta-llama 3.3 70B record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": "Llama 3.3 70B Instruct"
}
},
{
"id": "dbv2-qwen3-next-80b-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-qwen3-next-80b-a3b-instruct",
"_note": "Probes the Qwen3 Next 80B record; the bare qwen token strips on a hyphen boundary.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": "Qwen3 Next 80B A3B Instruct"
}
},
{
"id": "dbv2-qwen35-122b-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-qwen35-122b-a10b",
"_note": "Probes the Qwen3.5 122B record; the bare qwen token strips a qwen35 stem with no separator.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": "Qwen3.5 122B A10B"
}
},
{
"id": "dbv2-gemma-3-12b-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-gemma-3-12b",
"_note": "Probes the Gemma 3 12B endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": "Gemma 3 12B"
}
},
{
"id": "dbv2-inkling-exact-record-probe",
"provider": "databricks_v2",
"raw_model_id": "databricks-inkling",
"_note": "Probes the Inkling endpoint record and label.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"low",
"medium",
"high",
"xhigh",
"max"
],
"default_effort": null,
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-standard",
"registry_label": "Inkling"
}
},
{
"id": "dbv2-uc-fqn-gemini-3-5-flash-strip-probe",
"provider": "databricks_v2",
"raw_model_id": "system.ai.gemini-3-5-flash",
"_note": "Probes strip parity on a system.ai. UC FQN carrying the gemini- token (resolve carries no label; the alias label path is unit-tested).",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": null
}
},
{
"id": "dbv2-uc-fqn-meta-llama-strip-probe",
"provider": "databricks_v2",
"raw_model_id": "system.ai.meta-llama-3-3-70b-instruct",
"_note": "Probes strip parity on a UC FQN where the llama- token strips through meta-.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": null
}
},
{
"id": "dbv2-uc-goose-deepseek-strip-probe",
"provider": "databricks_v2",
"raw_model_id": "data_workflow_tools.goose.goose-deepseek-v4-pro-0813",
"_note": "Probes strip parity on a goose- prefixed UC FQN carrying the deepseek- token.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": null
}
},
{
"id": "dbv2-uc-fqn-inkling-strip-probe",
"provider": "databricks_v2",
"raw_model_id": "system.ai.inkling",
"_note": "Probes strip parity on a UC FQN carrying the bare inkling token.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"none",
"minimal",
"low",
"medium",
"high",
"xhigh"
],
"default_effort": "medium",
"databricks_v2_wire_route": "mlflow-chat",
"normalization_policy": "openai-clamp-max-to-xhigh",
"registry_label": null
}
},
{
"_group": "Label/capability token isolation probes (#6955 review pass 1)",
"_note": "Pins that label_family_tokens (the UC-humanization superset) never leaks into capability resolve(): capability stripping still uses only claude-/gpt-/kimi-, so a label token appearing before a gpt- marker must NOT displace the gpt-5-pro exact profile."
},
{
"id": "isolation-openai-gemini-gpt-5-pro-probe",
"provider": "openai",
"raw_model_id": "tenant-gemini-gpt-5-pro",
"_note": "The gemini- label token must not strip here; capability resolve keeps the gpt-5-pro high-only profile.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"high"
],
"default_effort": "high",
"databricks_v2_wire_route": "not-applicable",
"normalization_policy": "openai-standard",
"registry_label": null
}
},
{
"id": "isolation-openai-qwenchanted-gpt-5-pro-probe",
"provider": "openai",
"raw_model_id": "tenant-qwenchanted-gpt-5-pro",
"_note": "The bare qwen label token must not fire mid-segment; capability resolve keeps the gpt-5-pro high-only profile.",
"expect": {
"thinking_mode": "none",
"supported_efforts": [
"high"
],
"default_effort": "high",
"databricks_v2_wire_route": "not-applicable",
"normalization_policy": "openai-standard",
"registry_label": null
}
}
]