ParamSupportRegistry 14 fields

models/ParamSupportRegistry

version integer > 0

Registry grammar version.

required

refreshedAt string matching /^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d+)?Z$/

UTC timestamp of the refresh that last rewrote this file.

models map of map of map of object

Verdicts nested model id -> endpoint name -> parameter name; all three vocabularies are open by design.

required

verdict one of: supported | unsupported | inconclusive | pending-eval | blocked

Whether the provider accepts and honors this parameter.

required

source one of: doc | eval | observed | prose | schema

Evidence class behind the verdict; absent only while pending eval.

lastChecked string matching /^\d{4}-\d{2}-\d{2}$/

Date the verdict was last confirmed.

required

evidencePath non-empty string

Path or URL of the artifact backing the verdict.

type one of: enum | dimension | integer | string | boolean

Value shape the parameter accepts.

max integer

Maximum accepted integer value when the provider documents one.

toolsCollapseToOne boolean

Whether declared tools make this parameter return only one choice regardless of its requested count.

values non-empty list of non-empty string

Accepted values; required on enum parameters.

valuesComplete boolean

Whether `values` is the exhaustive accepted set rather than a sample.

unsupportedMode one of: rejected | ignored

How an unsupported parameter fails: rejected = provider errors; ignored = accepted with no effect.

note non-empty string

no description declared

Checks

pendingEvalRowsCarryNoSource - pending rows defer their source until evidence exists
settledVerdictsNameTheirSource - settled rows name their source
enumParametersListTheirValues - enum rows list their accepted values
completenessClaimsListTheirValues - exhaustive-value claims include values
unsupportedModeRequiresUnsupportedVerdict - unsupported mode belongs only to unsupported verdicts
dictionaryKeysAreNamed - model, endpoint, and parameter keys are non-empty
refreshedAtIsARealInstant - timestamp shape also names a real instant

Defects

No defects in the 1 record checked.

Where it lives

Bound by	models-catalog
Declared by	lib-models
Declared in	`~/mnt/mdr/skills/lib-models/assets/schemas/param-support.ts`
Bound in	`~/mnt/mdr/skills/models-catalog/assets/schemas/param-support.ts`
Binding	`paramSupportDataStore`
Directory	`~/mnt/mdr/skills/models-catalog/assets`
Files	`param-support.yml`
Referenced as	`param-support-registry`

Specimen record 159 lines

~/mnt/mdr/skills/models-catalog/assets/param-support.yml

1version: 1

2refreshedAt: '2026-11-10T03:32:22.506Z'

3models:

4 openai/gpt-6.1-terra:

5 default:

6 reasoning_effort:

7 verdict: supported

8 source: doc

9 lastChecked: '2026-10-27'

10 evidencePath: https://developers.openai.com/api/docs/models/gpt-6.1-terra

11 type: enum

12 values:

13 - none

14 - low

15 - medium

16 - high

17 - xhigh

18 - max

19 valuesComplete: true

20 note: Provider default is medium.

21 max_completion_tokens:

22 verdict: supported

23 source: doc

24 lastChecked: '2026-10-27'

25 evidencePath: https://developers.openai.com/api/docs/models/gpt-6.1-terra

26 type: integer

27 note: min 16; An upper bound for the number of tokens that can be generated for a response, including visible output tokens and [reasoning tokens](https://developers.openai.com/api/docs/guides/reasoning).; alias of max_output_tokens

28 tools:

29 verdict: supported

30 source: eval

31 lastChecked: '2026-10-27'

32 evidencePath: gen run 2026-10-27/11_51_00_244

33 note: Required declared function call returned report_number with number 323 through the current LiteLLM adapter; provider docs say tool calling requires Responses API.

34 prompt:

35 verdict: supported

36 source: schema

37 lastChecked: '2026-10-27'

38 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

39 note: Reference to a prompt template and its variables.

40 metadata:

41 verdict: supported

42 source: schema

43 lastChecked: '2026-10-27'

44 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

45 note: Set of 16 key-value pairs that can be attached to an object. This can be

46 top_logprobs:

47 verdict: supported

48 source: schema

49 lastChecked: '2026-10-27'

50 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

51 type: integer

52 note: min 0; max 20; An integer between 0 and 20 specifying the maximum number of most likely

53 temperature:

54 verdict: unsupported

55 source: doc

56 lastChecked: '2026-11-10'

57 evidencePath: ~/mnt/mdr/skills/models-litellm/assets/config.yml

58 unsupportedMode: ignored

59 note: Prior schema evidence records supported (https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml). min 0; max 2; default 1; What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic. LiteLLM route openai/gpt-6.1-terra/default drops temperature via additional_drop_params; provider support does not apply to this route.

60 top_p:

61 verdict: supported

62 source: schema

63 lastChecked: '2026-10-27'

64 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

65 note: min 0; max 1; default 1; An alternative to sampling with temperature, called nucleus sampling,

66 user:

67 verdict: supported

68 source: schema

69 lastChecked: '2026-10-27'

70 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

71 note: This field is being replaced by `safety_identifier` and `prompt_cache_key`. Use `prompt_cache_key` instead to maintain caching optimizations.

72 safety_identifier:

73 verdict: supported

74 source: schema

75 lastChecked: '2026-10-27'

76 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

77 type: string

78 note: A stable identifier used to help detect users of your application that may be violating OpenAI's usage policies.

79 prompt_cache_key:

80 verdict: supported

81 source: schema

82 lastChecked: '2026-10-27'

83 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

84 type: string

85 note: Used by OpenAI to cache responses for similar requests to optimize your cache hit rates. Replaces the `user` field. [Learn more](https://developers.openai.com/api/docs/guides/prompt-caching).

86 prompt_cache_retention:

87 verdict: supported

88 source: schema

89 lastChecked: '2026-10-27'

90 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

91 type: enum

92 values:

93 - in_memory

94 - 24h

95 valuesComplete: true

96 note: Deprecated. Use `prompt_cache_options.ttl` instead.

97 prompt_cache_options:

98 verdict: supported

99 source: schema

100 lastChecked: '2026-10-27'

101 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

102 note: Options for prompt caching. Supported for `gpt-5.6` and later models. By default, OpenAI automatically chooses one implicit cache breakpoint. You can add explicit breakpoints to content blocks with `prompt_cache_breakpoint`. Each request can write up to four breakpoints. For cache matching, OpenAI considers up to the latest 80 breakpoints in the conversation, without a content-block lookback limit. Set `mode` to `explicit` to disable the implicit breakpoint. The `ttl` defaults to `30m`, which is currently the only supported value. See the [prompt caching guide](https://developers.openai.com/api/docs/guides/prompt-caching) for current details.

103 previous_response_id:

104 verdict: supported

105 source: schema

106 lastChecked: '2026-10-27'

107 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

108 type: string

109 note: The unique ID of the previous response to the model. Use this to

110 model:

111 verdict: supported

112 source: schema

113 lastChecked: '2026-10-27'

114 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

115 type: string

116 background:

117 verdict: supported

118 source: schema

119 lastChecked: '2026-10-27'

120 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

121 type: boolean

122 note: default false; Whether to run the model response in the background.

123 max_tool_calls:

124 verdict: supported

125 source: schema

126 lastChecked: '2026-10-27'

127 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

128 type: integer

129 note: The maximum number of total calls to built-in tools that can be processed in a response. This maximum number applies across all built-in tool calls, not per individual tool. Any further attempts to call a tool by the model will be ignored.

130 text:

131 verdict: supported

132 source: schema

133 lastChecked: '2026-10-27'

134 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

135 note: Configuration options for a text response from the model. Can be plain

136 tool_choice:

137 verdict: supported

138 source: schema

139 lastChecked: '2026-10-27'

140 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

141 type: enum

142 values:

143 - none

144 - auto

145 - required

146 valuesComplete: true

147 note: Controls which (if any) tool is called by the model.

148 access_programs:

149 verdict: supported

150 source: schema

151 lastChecked: '2026-10-27'

152 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

153 note: Domain-specific access programs to use for this request.

154 service_tier:

155 verdict: supported

156 source: schema

157 lastChecked: '2026-10-27'

158 evidencePath: https://raw.githubusercontent.com/openai/openai-openapi/master/openapi.yaml

159 type