jev-recipes 0.2.0 → 0.3.0

package/recipes/action-scope/README.md

@@ -1,49 +0,0 @@
-# Check proposed action scope
-
-Is proposedAction within the work requested in request and constraints? Do not treat a potentially helpful extra task as requested work. An explicit constraint overrides an implied convenience.
-
-```ts
-import { actionScope } from 'jev-recipes/action-scope';
-
-const result = await actionScope({
-  request: 'Explain why this deployment failed.',
-  proposedAction: 'Deploy a replacement release to production.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field            | Accepts                                      |
-| ---------------- | -------------------------------------------- |
-| `request`        | Non-empty text                               |
-| `proposedAction` | Non-empty text                               |
-| `constraints`    | Optional non-empty text                      |
-| `minConfidence`  | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict           | Meaning                                                                                                       |
-| ----------------- | ------------------------------------------------------------------------------------------------------------- |
-| `within_scope`    | The action is explicitly requested or directly necessary under the stated constraints.                        |
-| `additional_work` | The action introduces unrequested work, changes the requested outcome, or violates a stated scope constraint. |
-| `unclear`         | The scope relationship cannot be established from the request.                                                |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Assesses semantic scope only. User authorization and access controls must be enforced by the application.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo action-scope` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe action-scope` shows the input and result schemas.

package/recipes/answer-consistency/README.md

@@ -1,50 +0,0 @@
-# Check answer consistency
-
-Do firstStatement and secondStatement make compatible claims about the same subject and circumstances? Distinguish different conditions from a contradiction.
-
-```ts
-import { answerConsistency } from 'jev-recipes/answer-consistency';
-
-const result = await answerConsistency({
-  firstStatement: 'Guests can export reports.',
-  secondStatement: 'Guests cannot export reports.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field             | Accepts                                      |
-| ----------------- | -------------------------------------------- |
-| `firstStatement`  | Non-empty text                               |
-| `secondStatement` | Non-empty text                               |
-| `context`         | Optional non-empty text                      |
-| `minConfidence`   | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict       | Meaning                                                                                 |
-| ------------- | --------------------------------------------------------------------------------------- |
-| `consistent`  | Both statements concern the same subject and can hold under the supplied circumstances. |
-| `conflicting` | The statements concern the same circumstances and make mutually incompatible claims.    |
-| `unrelated`   | The statements concern different subjects or scopes that should not be compared.        |
-| `unclear`     | The supplied circumstances do not resolve whether the statements conflict.              |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Checks semantic consistency between two statements. Exact numeric and date comparisons belong in code.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo answer-consistency` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe answer-consistency` shows the input and result schemas.

package/recipes/answer-coverage/README.md

@@ -1,60 +0,0 @@
-# Check answer coverage
-
-Check whether a draft answers each supplied question.
-
-```ts
-import { answerCoverage } from 'jev-recipes/answer-coverage';
-
-const result = await answerCoverage({
-  draft: 'Select Forgot password on the sign-in page.',
-  questions: [
-    {
-      id: 'reset',
-      text: 'How do I reset my password?',
-    },
-    {
-      id: 'expiry',
-      text: 'When does the link expire?',
-    },
-  ],
-});
-
-console.log(result.status, result.checks);
-```
-
-## Input
-
-| Field           | Accepts                                      |
-| --------------- | -------------------------------------------- |
-| `draft`         | Non-empty text                               |
-| `questions`     | 1 to 50 `{ id, text }` items with unique IDs |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-`checks` contains the original ID, verdict, status, confidence, and probabilities for each item. `allAnswered` is true only when every check is ready and has verdict `answered`. Any uncertain check makes overall status `review`.
-
-| Verdict    | Meaning                                                  |
-| ---------- | -------------------------------------------------------- |
-| `answered` | Every material part of this question is answered.        |
-| `partial`  | Some material parts are answered but others are missing. |
-| `missing`  | No substantive answer to this question appears.          |
-| `unclear`  | The wording prevents a reliable coverage decision.       |
-
-`unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared item-check helper. This folder owns its question, verdict criteria, and aggregate decision. A live invocation makes one logical Jev request; SDK retries can add transport attempts. All item questions are sent in that request.
-
-## Limits
-
-Checks supplied question IDs only. It does not extract questions or verify the draft against sources.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo answer-coverage` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe answer-coverage` shows the input and result schemas.

package/recipes/answer-invalidation/README.md

@@ -1,50 +0,0 @@
-# Check an answer after a source change
-
-Does updatedEvidence still support the entire claim that was based on previousEvidence? Removing necessary support can invalidate the claim even without an explicit contradiction. Do not assume the claim was supported before.
-
-```ts
-import { answerInvalidation } from 'jev-recipes/answer-invalidation';
-
-const result = await answerInvalidation({
-  claim: 'Guests can export reports.',
-  previousEvidence: 'Guests can export reports.',
-  updatedEvidence: 'Exporting reports is restricted to workspace owners. Guests cannot export.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field              | Accepts                                      |
-| ------------------ | -------------------------------------------- |
-| `claim`            | Non-empty text                               |
-| `previousEvidence` | Non-empty text                               |
-| `updatedEvidence`  | Non-empty text                               |
-| `minConfidence`    | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict           | Meaning                                                                                                         |
-| ----------------- | --------------------------------------------------------------------------------------------------------------- |
-| `still_supported` | The updated evidence supports the entire claim.                                                                 |
-| `invalidated`     | The previous evidence supported the claim, but the updated evidence contradicts it or removes required support. |
-| `unclear`         | Previous support or the effect of the new evidence cannot be established.                                       |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Assesses one supplied claim and source revision. It does not monitor changes, compare release dates, or invalidate a cache automatically.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo answer-invalidation` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe answer-invalidation` shows the input and result schemas.

package/recipes/answer-relevance/README.md

@@ -1,49 +0,0 @@
-# Check answer relevance
-
-How directly does draft address request? Judge alignment with the requested subject and task, not factual correctness or completeness.
-
-```ts
-import { answerRelevance } from 'jev-recipes/answer-relevance';
-
-const result = await answerRelevance({
-  request: 'How do I reset my password?',
-  draft: 'You can download invoices from the Billing page.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field           | Accepts                                      |
-| --------------- | -------------------------------------------- |
-| `request`       | Non-empty text                               |
-| `draft`         | Non-empty text                               |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict           | Meaning                                                                                                |
-| ----------------- | ------------------------------------------------------------------------------------------------------ |
-| `relevant`        | The draft directly addresses the requested subject and task.                                           |
-| `partly_relevant` | The draft mixes relevant content with a substantial unrelated tangent or addresses a neighboring task. |
-| `off_topic`       | The draft does not address the requested subject or task.                                              |
-| `unclear`         | The request or draft is too ambiguous to assess alignment.                                             |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Relevance does not establish correctness or complete coverage. Use answer-coverage and verify for those decisions.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo answer-relevance` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe answer-relevance` shows the input and result schemas.

package/recipes/answerability/README.md

@@ -1,54 +0,0 @@
-# Check answerability
-
-Decide whether the supplied evidence can answer an entire question before drafting a response.
-
-```ts
-import { answerability } from 'jev-recipes/answerability';
-
-const result = await answerability({
-  question: 'How do I reset my password, and when does the link expire?',
-  evidence: [
-    {
-      id: 'reset',
-      text: 'Select Forgot password. Reset links expire after 30 minutes.',
-    },
-  ],
-});
-
-console.log(result.canAnswer);
-```
-
-## Input
-
-| Field           | Accepts                                                                      |
-| --------------- | ---------------------------------------------------------------------------- |
-| `question`      | Non-empty text containing the question                                       |
-| `evidence`      | 1 to 50 `{ id, text }` passages with unique non-empty IDs and non-empty text |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8                                 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors. Types are inferred from this folder's Zod 4 schemas.
-
-## Result
-
-| Verdict        | Meaning                                                   |
-| -------------- | --------------------------------------------------------- |
-| `sufficient`   | The evidence answers every material part of the question. |
-| `partial`      | Some material parts can be answered, but others cannot.   |
-| `insufficient` | No material part can be answered.                         |
-| `conflicting`  | Incompatible evidence prevents a consistent answer.       |
-
-`canAnswer` is true only when the verdict is `sufficient` and confidence reaches `minConfidence`. `status` is `ready` at or above that threshold, otherwise `review`. A ready assessment can still describe partial, insufficient, or conflicting evidence. The result includes confidence and all choice probabilities.
-
-The result includes model and token usage. Inspect the outcome as well as its review status.
-
-## Reuse and calls
-
-Uses the shared choice helper. This folder owns the evidence-sufficiency criteria and the can-answer decision. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Handle empty retrieval in application code before calling. This recipe does not establish source truth, draft an answer, or guarantee a later draft uses the evidence correctly. Use verify on drafted claims and answer-coverage on supplied question parts.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo answerability` shows an offline illustration, not an accuracy measurement. Use `npx jev-recipes describe answerability` to inspect the input and result schemas.

package/recipes/argument-fit/README.md

@@ -1,51 +0,0 @@
-# Check argument meaning
-
-Does proposedValue for argument express the intended value in request and context? Assess meaning only; syntax and exact identity checks belong in code.
-
-```ts
-import { argumentFit } from 'jev-recipes/argument-fit';
-
-const result = await argumentFit({
-  request: 'Search closed incidents only.',
-  argument: 'Incident status filter',
-  proposedValue: 'open',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field           | Accepts                                      |
-| --------------- | -------------------------------------------- |
-| `request`       | Non-empty text                               |
-| `argument`      | Non-empty text                               |
-| `proposedValue` | Non-empty text                               |
-| `context`       | Optional non-empty text                      |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict     | Meaning                                                         |
-| ----------- | --------------------------------------------------------------- |
-| `fits`      | The proposed value matches the stated intent for this argument. |
-| `conflicts` | The proposed value contradicts the stated intent.               |
-| `unclear`   | The intended value is missing or ambiguous.                     |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Does not validate schemas, compare identifiers, or enforce authorization. Check those deterministically before acting.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo argument-fit` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe argument-fit` shows the input and result schemas.

package/recipes/attempted-step/README.md

@@ -1,48 +0,0 @@
-# Check a previously attempted step
-
-Does conversation establish whether the customer already performed step? Being told to try it, planning to try it, and actually trying it are different. Do not infer not-tried merely from silence.
-
-```ts
-import { attemptedStep } from 'jev-recipes/attempted-step';
-
-const result = await attemptedStep({
-  step: 'Clear the browser cache and try again.',
-  conversation: 'I already cleared my browser cache and tried again. The error is still there.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field           | Accepts                                      |
-| --------------- | -------------------------------------------- |
-| `step`          | Non-empty text                               |
-| `conversation`  | Non-empty text                               |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict     | Meaning                                                                                   |
-| ----------- | ----------------------------------------------------------------------------------------- |
-| `tried`     | The conversation explicitly states or unambiguously reports that this step was performed. |
-| `not_tried` | The conversation explicitly establishes the step has not been performed.                  |
-| `unclear`   | There is no reliable statement of whether the step was performed.                         |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Interprets reports of an attempt. It does not verify that the step was completed correctly or repeat the action.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo attempted-step` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe attempted-step` shows the input and result schemas.

package/recipes/audience-fit/README.md

@@ -1,50 +0,0 @@
-# Check audience fit
-
-Does the level of explanation in document fit the knowledge and needs explicitly described in audience? Do not infer ability from demographic traits.
-
-```ts
-import { audienceFit } from 'jev-recipes/audience-fit';
-
-const result = await audienceFit({
-  document: 'Rotate the OAuth refresh token and invalidate the active session grant.',
-  audience:
-    'A customer unfamiliar with authentication terminology who wants to sign out of all devices.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field           | Accepts                                      |
-| --------------- | -------------------------------------------- |
-| `document`      | Non-empty text                               |
-| `audience`      | Non-empty text                               |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict         | Meaning                                                                                          |
-| --------------- | ------------------------------------------------------------------------------------------------ |
-| `appropriate`   | The terminology and explanation suit the described knowledge and needs.                          |
-| `too_technical` | The document assumes knowledge the supplied audience description says is absent.                 |
-| `too_basic`     | The document spends its explanation below the stated needs without supplying the required depth. |
-| `unclear`       | The audience or explanation is not specified enough to judge.                                    |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Uses only the supplied audience description. It does not infer ability from identity or rewrite the document.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo audience-fit` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe audience-fit` shows the input and result schemas.

package/recipes/cache-match/README.md

@@ -1,51 +0,0 @@
-# Check a cached answer
-
-Does cachedAnswer address question with the same relevant meaning and conditions as originalQuestion? Do not assume freshness, access, tenant, or product compatibility beyond the supplied facts.
-
-```ts
-import { cacheMatch } from 'jev-recipes/cache-match';
-
-const result = await cacheMatch({
-  question: 'Where can I get my invoices?',
-  originalQuestion: 'How do I download invoices?',
-  cachedAnswer: 'Open Billing and select Download invoice.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field              | Accepts                                      |
-| ------------------ | -------------------------------------------- |
-| `question`         | Non-empty text                               |
-| `originalQuestion` | Non-empty text                               |
-| `cachedAnswer`     | Non-empty text                               |
-| `context`          | Optional non-empty text                      |
-| `minConfidence`    | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict      | Meaning                                                                                                   |
-| ------------ | --------------------------------------------------------------------------------------------------------- |
-| `reusable`   | The cached answer addresses the new question without a material mismatch in meaning or stated conditions. |
-| `unsuitable` | The cached answer leaves a material part unanswered or applies to a different stated condition.           |
-| `unclear`    | Missing or ambiguous conditions prevent establishing a match.                                             |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Enforce tenant, permissions, version, and freshness checks in code before calling. A semantic match does not verify the cached answer.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo cache-match` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe cache-match` shows the input and result schemas.

package/recipes/cancellation-check/README.md

@@ -1,50 +0,0 @@
-# Check cancellation intent
-
-Does message ask to cancel, pause, or continue task? Resolve which task the instruction concerns. Pausing means stop temporarily; cancellation abandons the task.
-
-```ts
-import { cancellationCheck } from 'jev-recipes/cancellation-check';
-
-const result = await cancellationCheck({
-  task: 'Draft a response to the customer.',
-  message: 'Hold off until I send you the updated policy.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field           | Accepts                                      |
-| --------------- | -------------------------------------------- |
-| `task`          | Non-empty text                               |
-| `message`       | Non-empty text                               |
-| `context`       | Optional non-empty text                      |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict    | Meaning                                                                         |
-| ---------- | ------------------------------------------------------------------------------- |
-| `cancel`   | The message clearly asks to abandon this task.                                  |
-| `pause`    | The message clearly asks to suspend this task temporarily.                      |
-| `continue` | The message clearly asks to proceed with this task.                             |
-| `unclear`  | The message does not clearly establish one of these instructions for this task. |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Does not stop running work or infer permission to continue from silence. The caller applies the decision.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo cancellation-check` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe cancellation-check` shows the input and result schemas.

package/recipes/certainty-match/README.md

@@ -1,49 +0,0 @@
-# Check certainty wording
-
-Does the certainty expressed in draft match assessment? Treat assessment as the supplied evidence assessment and do not independently re-evaluate its truth.
-
-```ts
-import { certaintyMatch } from 'jev-recipes/certainty-match';
-
-const result = await certaintyMatch({
-  draft: 'The outage was definitely caused by the deploy.',
-  assessment: 'The deploy is one possible cause; the cause has not been established.',
-});
-
-console.log(result.status, result.verdict);
-```
-
-## Input
-
-| Field           | Accepts                                      |
-| --------------- | -------------------------------------------- |
-| `draft`         | Non-empty text                               |
-| `assessment`    | Non-empty text                               |
-| `minConfidence` | Optional number from 0 to 1; defaults to 0.8 |
-
-See [shared options and behavior](../README.md#shared-options-and-behavior) for client configuration, validation, and errors.
-
-## Result
-
-| Verdict       | Meaning                                                                       |
-| ------------- | ----------------------------------------------------------------------------- |
-| `overstated`  | The draft expresses stronger certainty than the assessment supports.          |
-| `appropriate` | The draft expresses a degree of certainty consistent with the assessment.     |
-| `understated` | The draft expresses materially weaker certainty than the assessment supports. |
-| `unclear`     | The certainty levels cannot be compared from the supplied wording.            |
-
-The result includes `verdict`, `confidence`, and all choice `probabilities`. `unclear` always requires review, even at high confidence. Other verdicts can be ready, including negative assessments.
-
-Results include model and token usage. Decisions below `minConfidence` require review. Inspect the outcome as well as the status.
-
-## Reuse and calls
-
-Uses the shared choice helper for the Jev call and response parsing. This folder owns its question, verdict criteria, and review policy. A live invocation makes one logical Jev request; SDK retries can add transport attempts.
-
-## Limits
-
-Compares wording with a supplied assessment. It does not calibrate probabilities or establish the assessment itself.
-
-## Example input
-
-[demo.json](demo.json) contains editable input and a hand-authored response. After installing `jev-recipes`, `npx jev-recipes demo certainty-match` shows the result offline. The fixture is an illustration, not an accuracy measurement. `npx jev-recipes describe certainty-match` shows the input and result schemas.