@factbag/jev 0.0.0-bootstrap.0 → 0.1.0-alpha.7

package/CHANGELOG.md

@@ -0,0 +1,60 @@
+# @factbag/jev
+
+## 0.1.0-alpha.7
+
+### Patch Changes
+
+- factbag@0.1.0-alpha.7
+
+## 0.0.0-alpha.6
+
+### Minor Changes
+
+- 6433c6b: Classify open-ended `String`, `Number`, and `Date` fields, and entity fields with no registered entities, over the values each passage offers. Pass `createChunkSymbolExtractor` to `classifier(...)` from `@factbag/jev` or `@factbag/clef` to build an extractor per chunk whose `query` answers candidate values for the open fields of the predicates it classifies. Both packages export the `ChunkSymbolExtractor`, `CreateChunkSymbolExtractor`, and `SymbolAnswer` types.
+
+  Extracted values are candidates only. An extracted entity is registered in `run.kb.entities` when asserted or denied evidence mentions it; entities the model does not admit evidence for are discarded.
+
+### Patch Changes
+
+- Updated dependencies [7225a51]
+- Updated dependencies [7f26b45]
+- Updated dependencies [b7c85ff]
+- Updated dependencies [ed6ec84]
+- Updated dependencies [5114b2f]
+- Updated dependencies [15391f2]
+- Updated dependencies [eab3679]
+- Updated dependencies [ecbd8d9]
+- Updated dependencies [e4b343e]
+  - factbag@0.0.0-alpha.6
+
+## 0.0.0-alpha.5
+
+### Minor Changes
+
+- 19ac508: **Breaking:** Jev moves from `factbag/jev` to its own package, `@factbag/jev`. Install it alongside `factbag` and `@typesafe-ai/sdk`, which is now a required peer dependency at `^0.7.0`, and import `classifier`, `resolver`, `createConfig`, `Config`, and the `DEFAULT_*` constants from `@factbag/jev`. `factbag` no longer has a Jev entry point or an SDK peer dependency. `@factbag/jev` and `@factbag/clef` are released together with `factbag` and require the same `factbag` version.
+
+  **Breaking:** `@factbag/jev` has a single entry point. `factbag/jev/classifier`, `factbag/jev/resolve`, `factbag/jev/config`, and `factbag/jev/errors` have no replacement subpaths.
+
+  Add `batchSize` to Jev's `Config`: how many questions each request holds. It defaults to 64 and has no upper limit. An invalid `batchSize` throws when the config is created or a run is built.
+
+  Jev's `createConfig` now returns `Normalized & Config`, so `client` is no longer guaranteed on the result.
+
+### Patch Changes
+
+- Updated dependencies [19ac508]
+- Updated dependencies [19ac508]
+- Updated dependencies [dd3730b]
+  - factbag@0.0.0-alpha.5
+
+## 0.1.0-next.0
+
+### Minor Changes
+
+- Publish the initial experimental Factbag packages with native Node ESM artifacts and declarations.
+
+### Patch Changes
+
+- Updated dependencies
+  - @factbag/core@0.1.0-next.0
+  - @factbag/corpus@0.1.0-next.0
+  - @factbag/guards@0.1.0-next.0

package/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Daniel du Plessis
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

package/README.md

@@ -1,11 +1,321 @@
 # @factbag/jev
 
-This version is a placeholder. It exists so that npm trusted publishing can be set up
-for this package before its first release.
+`classifier({ corpus, input: kb, config })` returns a lazy iterator of executable
+classification jobs, one per chunk. Constructing and consuming the iterator makes **no remote calls**. The caller
+decides when to execute each job and how much work to run concurrently.
 
-The first real release is `0.0.0-alpha.5`, published with `factbag` from
-[factbag/factbag](https://github.com/factbag/factbag/tree/main/providers/jev):
+The package has one entry point, `@factbag/jev`, which exports:
+
+- `classifier({ corpus, input: kb, config? })`
+- `resolver(corpus, kb, config?)`
+- `createConfig`, `Config`, and the `DEFAULT_*` constants
+
+## Install
 
 ```sh
-npm install factbag@alpha @factbag/jev@alpha
+npm install factbag@alpha @factbag/jev@alpha @typesafe-ai/sdk@^0.7.0
 ```
+
+`@factbag/jev` is released with `factbag` and needs the same `factbag` version.
+Set `TYPESAFE_API_KEY`, or pass your own `TypeSafeClient` as `config.client`.
+
+## Usage
+
+```ts
+import { enumeration, predicate } from "factbag/dsl";
+import { KnowledgeBase } from "factbag/kb";
+import { Corpus } from "factbag/corpus";
+import { guard } from "factbag/kb";
+import { classifier } from "@factbag/jev";
+
+class Company {
+  constructor(
+    readonly id: string,
+    readonly tradingName: string,
+  ) {}
+}
+
+const status = predicate(
+  "company_status",
+  {
+    company: Company,
+    status: enumeration("active", "dissolved"),
+  },
+  {
+    description: "The document explicitly establishes this company's legal status.",
+  },
+);
+
+const kb = new KnowledgeBase();
+kb.entities.add(new Company("company-42", "Harbor"));
+const corpus = new Corpus();
+corpus.add({ documentId: "registry", text: "Harbor is an active company." });
+
+// The default TypeSafeClient reads TYPESAFE_API_KEY on the first execute().
+const run = classifier({
+  corpus,
+  input: guard(kb).permit("assert", status).permit("deny", status),
+});
+
+for (const job of run) {
+  await job.execute();
+  // The consumer may await persistence or a rate limiter here.
+}
+
+const enriched = run.kb;
+// kb is unchanged. enriched contains the original knowledge plus new evidence.
+```
+
+## Boundary and registration
+
+Classification and resolution share one optional, flat `Config`. Create it once
+with `createConfig` to share one lazy client across runs:
+
+```ts
+const config = createConfig({ client, minimumProbability: 0.6 });
+const classified = classifier({ corpus, input: kb, config });
+const resolved = resolver(corpus, classified.kb, config);
+```
+
+Admission defaults to `{ minimumProbability: 0.4, ambiguityMargin: 0 }`:
+the strongest ASSERT/DENY outcome must reach 0.4 and outrank IRRELEVANT. An exact
+ASSERT/DENY tie admits both. These are convenience defaults, not calibrated probabilities.
+A plainly worded record often scores in the 0.45-0.65 band, so 0.5 loses evidence a reader
+would call clear; a lower bar writes more, and contradictions surface rather than gaps.
+
+```ts
+classifier({ corpus, input: kb });
+classifier({ corpus, input: kb, config: { client } });
+classifier({ corpus, input: kb, config: { minimumProbability: 0.8, ambiguityMargin: 0.05 } });
+```
+
+- A **guarded KB** retains its exact permit/block policy. `guard(kb)` is default-deny.
+- A **bare KB** permits assert/deny for all predicates referenced in its
+  evidence and rules, including rule bodies. There is no global predicate registry.
+- Query permissions do not make a predicate classifiable.
+- Every exposed predicate is validated synchronously during `classifier(...)`.
+  Unsupported predicates fail even if the corpus is empty; blocked predicates are
+  not registered.
+
+`UnsupportedPredicateError` (from `factbag/exceptions`) identifies the predicate and, where applicable, the
+field that cannot be classified. A nonempty metadata description is required.
+
+Supported field domains:
+
+| Schema field               | Candidate values                                     |
+| -------------------------- | ---------------------------------------------------- |
+| `Boolean`                  | `false`, `true`                                      |
+| `enumeration(...)`         | Explicit string, number, or Date members             |
+| Entity constructor         | Registered instances, plus each chunk's extracted    |
+| `String`, `Number`, `Date` | Each chunk's extracted values; unsupported otherwise |
+
+Without `createChunkSymbolExtractor`, an entity domain must contain registered instances. Existing scalar facts do not
+turn an open-ended schema into an enum. To model a fixed value, declare it explicitly
+with `enumeration(value)` rather than asking the classifier to discover it.
+
+Number and Date enumerations, such as `enumeration(10, 20)` or
+`enumeration(new Date("2026-10-25T00:00:00Z"))`, are classified like string
+enumerations. Dates are sent to the model as ISO strings and retained as `Date`
+values in evidence. The same date in two fields is a valid candidate; only
+repeated entities are excluded.
+
+Entity candidates include their enumerable properties, including `id`, plus `type`
+set to the constructor name. They are JSON-encoded when each request is built:
+nested properties are included, dates become ISO strings, and methods are not sent.
+For example, the company above is sent as
+`{ type: "Company", id: "company-42", tradingName: "Harbor" }`.
+Evidence continues to reference the original registered entity. Related KB facts
+are not automatically included in the entity payload.
+
+Fields fan out as a Cartesian product. A predicate with no fields has one candidate
+proposition. Every candidate gets an independent `Choice(ASSERT, DENY, IRRELEVANT)`
+question. Assert and deny permissions for the same predicate share one question.
+No enum exclusivity or Boolean equivalence is assumed; those belong to the ruleset.
+
+### Passage symbols
+
+Pass `createChunkSymbolExtractor` to fill open-ended and entity fields from the text
+being classified. The classifier builds one extractor per job over that job's chunk,
+and asks it for the open fields of the predicates it classifies:
+
+```ts
+const run = classifier({
+  corpus,
+  input: access,
+  createChunkSymbolExtractor: (chunks) => {
+    const found = scan(chunks);
+    return {
+      query: (requests) =>
+        new Map(
+          [...requests].map(([predicate, fields]) => [
+            predicate,
+            Object.fromEntries(fields.map((name) => [name, found.for(predicate, name)])),
+          ]),
+        ),
+    };
+  },
+});
+```
+
+- Open fields are `String`, `Number`, `Date`, and entity fields. Boolean and
+  enumeration fields keep their declared domains.
+- `query` receives a map from each predicate to its open field names, and answers with
+  a map from each predicate to the values for each field, possibly asynchronously. How
+  it finds them, and what it caches, is up to the extractor.
+- The answer must cover exactly the requested predicates and fields, with values each
+  field can hold: strings, finite numbers, valid dates, or entities of the field's
+  type. Otherwise the job rejects before anything is classified.
+- Scalar fields take exactly the values answered. Entity fields take them alongside
+  their registered instances. Extracted entities are only candidates: those that
+  asserted or denied evidence mentions are registered in `run.kb.entities`, and the
+  rest are discarded.
+- A predicate with a field left empty is skipped for that chunk.
+
+An entity's type and `id` are its identity, so the extractor decides which mentions
+are the same entity by the ids it mints. `new Person("Alice")` from two chunks is one
+entity.
+
+## Jobs and execution
+
+`run` is a single-cursor `IterableIterator`. Each yielded job classifies one chunk
+and exposes only `execute(): Promise<void>`, sending up to `batchSize` questions per
+request. `batchSize` defaults to 64. TypeSafe sets no upper limit. Re-executing a job makes a fresh attempt; a repeated
+success adds fresh evidence.
+
+The consumer can execute jobs sequentially, in batches, or through a concurrency
+limiter. Jobs contain no scheduler, rate limiter, retry loop, or error handler.
+`config.client` is optional and defaults to `new TypeSafeClient()`, created on the
+first request. Pass a configured client to set the model, retry, timeout or
+transport, or a scripted one in tests.
+
+Request errors propagate unchanged from `execute()`. All answers for a chunk are
+validated before any of that chunk's evidence is written. A failure does not
+terminate the job iterator, remove earlier evidence, or retry automatically.
+The caller can retry the same job or advance to another job.
+
+`run.kb` is a live output store. Concurrent jobs add evidence in completion order.
+Existing evidence, rules and entities retain their identities, but the storage
+containers are independent of the input KB. Candidate domains use the construction-
+time KB snapshot; new evidence does not introduce new candidates during a run.
+
+## Corpus traversal
+
+The classifier reads the whole corpus: every document present at construction
+time. There is no source selection; to classify a subset, pass a
+filtered view such as `corpus.filter((document) => document.metadata["kind"] !== "structured")`.
+Its chunks register in the shared corpus, so evidence resolves from any handle. The adapter always uses a private recursive
+splitter: up to **8,000 Unicode graphemes**, with up to
+**200 graphemes of overlap**, preferring paragraph, line, and word boundaries.
+Short documents remain intact. Jev constructs `chunkers.recursive()` internally;
+source ranges and jobs are produced as iteration advances. The chunker indexes
+the current document's grapheme boundaries when iteration starts.
+
+Overlap supplies neighboring context. There is no configurable chunker or context
+expansion step on the classifier API.
+
+This is a conservative text-sizing policy, not an exact token budget. Jev 1.13's
+[documented limits](https://docs.typesafe.ai/models) are 32k tokens for state plus
+the longest question and 64k for the whole request. Candidate state, metadata and
+questions also consume that budget; this passage bound alone does not guarantee
+that arbitrarily large candidate batches fit.
+
+Requests carry only the passage text. Chunk IDs, document IDs and source ranges are
+kept out of them, so document names and coordinates cannot influence the model.
+
+Each evidence source is the exact consumed chunk ID, resolvable through `corpus.resolve(source)`. The ID encodes the document and
+range, and document text is immutable.
+
+## Admission and results
+
+- When `IRRELEVANT` is at least as probable as either polarity, admit neither.
+- Otherwise, a polarity must meet `minimumProbability` and fall within
+  `ambiguityMargin` of the stronger polarity.
+- Only permitted polarities are inserted. Forbidden outcomes are not relabeled or
+  renormalized into permitted ones.
+- Both polarities can be admitted, preserving evidence that needs review.
+
+New literals carry source, outcome probability,
+candidate IDs, selected choice, SDK confidence, and the full distribution. SDK confidence remains separate from outcome probability. These
+thresholds are application policy, not calibrated defaults.
+
+## Optional relevance filtering
+
+Set `config: { routing: true }` to reduce the candidate facts sent for classification
+by filtering predicates and values for relevance to each passage. Each chunk job
+first asks which predicates the passage may bear on. For predicates that make claims
+about supplied entities or values, it can also ask which of those symbols the passage
+involves. For example, before classifying ownership claims, it can narrow the list
+of companies to those relevant to the passage. Only combinations of the retained
+symbols are classified.
+
+Symbol filtering runs when a retained predicate has symbols to filter and its
+candidate combinations exceed `candidateProductCutoff` (default 16). Each distinct
+symbol is checked once across all surviving predicates and roles. Boolean field
+values are always retained. `predicateThreshold` and `symbolThreshold`
+(default 0.3) are inclusive.
+
+Filtering is **off by default**: a false negative in either pass silently drops
+evidence, including denials. Filtering adds requests and does not guarantee cost savings. Run
+`pnpm --filter @factbag/jev demo:staged` to compare both modes on a scripted fixture.
+
+## Resolving contradictions
+
+Each chunk is classified on its own, so a 2012 removal notice and a 2022
+redesignation can leave the same fact asserted and denied. `resolver` reads those
+passages together and asks, for each one, whether its account still stands. It works
+like `classifier`: construction makes no remote calls, and each job is one request.
+
+```ts
+import { resolver } from "@factbag/jev";
+
+const resolution = resolver(corpus, run.kb);
+for (const job of resolution) {
+  await job.execute();
+}
+const resolved = resolution.kb;
+```
+
+- There is one job per atom with evidence of both signs. Its `execute()` sends
+  every distinct source passage in one request with a KEEP/RETRACT
+  question per passage: does another passage supersede this one for the fact as
+  described, including any date it is stated as of? Disagreement alone is not
+  supersession, so two documents that merely conflict both stand.
+- The model judges documents, not the fact. Evidence from a retracted passage is removed
+  from `resolution.kb`; the rest stays, and the solver decides what the surviving
+  evidence means. Retracting every passage leaves the fact `unknown`, which is right:
+  the evidence was withdrawn, not disproved.
+- A passage is retracted only when RETRACT reaches `minimumProbability` and leads KEEP by
+  more than `ambiguityMargin` (defaults 0.4 and 0). With two outcomes that means RETRACT must
+  exceed 0.5 in practice, whatever the minimum. Keeping the evidence is the default,
+  so an uncertain answer changes nothing.
+- Every piece of conflicting evidence must have a `source` that resolves in the corpus.
+  Construction throws for evidence without a source, naming the atom, and propagates
+  corpus errors such as `Invalid chunk ID`, so no job exists until all are readable.
+  To resolve a KB with stipulated or tool-sourced evidence, decide those contradictions
+  first or resolve a fork without them.
+- `resolution.kb` is an independent copy of the input, which is unchanged.
+  Contradictions are independent, so jobs can run in any order or concurrently.
+- A job settles once. Request errors propagate unchanged and leave the job retryable;
+  executing it again after it succeeds throws. `config.client` defaults to
+  `new TypeSafeClient()`, as for the classifier.
+
+## Internal structure
+
+This package only connects Jev to the shared workflows in `@factbag/system-one`:
+the TypeSafe client, Jev's batch limits, and the public signatures. Candidate
+selection, admission, prompts, batching, and answer validation live there and are
+bundled into the published package. The same code serves `@factbag/clef`.
+
+## Examples and checks
+
+Read [Fix one fact, not the whole answer](https://github.com/factbag/factbag/blob/main/providers/jev/docs/README.md) for the narrative, or the
+[example notes](https://github.com/factbag/factbag/blob/main/providers/jev/examples/README.md) for the runnable scenarios.
+
+```sh
+pnpm --filter @factbag/jev demo
+pnpm --filter @factbag/jev test
+pnpm --filter @factbag/jev typecheck
+```
+
+Demos use the real SDK with MSW HTTP interception; focused unit tests also use
+injected clients. Neither needs credentials or live inference.