factbag 0.0.0-alpha.1 → 0.0.0-alpha.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -1,5 +1,39 @@
1
1
  # factbag
2
2
 
3
+ ## 0.0.0-alpha.3
4
+
5
+ ### Minor Changes
6
+
7
+ - f8f2712: Add `eq` and `neq` rule-body comparisons for numbers, dates, strings, booleans, and entities. Entities compare by type and id rather than instance; dates compare by instant.
8
+ - 900b104: Add min, max, roundTo, floorTo, and ceilTo numeric arithmetic terms for rule bodies. Rounding supports whole-number decimal places (including negative places), with halves rounded away from zero. Decimal exponent shifting avoids multiplication artifacts at decimal rounding boundaries.
9
+ - 4f040e3: Add `daysBetween(later, earlier)` and `yearsBetween(later, earlier)` as number-valued arithmetic terms over dates. They count signed UTC calendar days and whole anniversary years, ignoring time of day; February 29 anniversaries fall on March 1 in common years. Numeric subtraction on variables now composes with other numeric operators.
10
+
11
+ ### Patch Changes
12
+
13
+ - 01cc338: Speed up queries over large rulesets with linear-time recursive predicate grouping,
14
+ per-query rule and evidence indexes, and judgment reuse during candidate discovery.
15
+ Preserve candidate pruning and distinguish same-named predicate declarations when
16
+ tracking active atoms, avoiding incorrect recursion cutoffs.
17
+ - 80d1e75: Share explanation nodes for reused proof judgments so explanation construction no longer grows with the number of paths through a shared proof. Identical sub-explanations are still listed once, and rendered content is unchanged.
18
+ - a074dab: Judge evidence-free, closed-world recursive components together as a least fixed
19
+ point instead of enumerating recursive proof paths. Doubly recursive transitive
20
+ closures and symmetric compositions no longer require exponential proof search.
21
+ Other components retain the existing search semantics. Variable queries recognize
22
+ a non-recursive disjunctive base case within an exclusive definition.
23
+
24
+ ## 0.0.0-alpha.2
25
+
26
+ ### Major Changes
27
+
28
+ - e212a94: Remove bigint support from predicate schemas, numeric expressions, aggregates, and numeric proof ranges. Numeric values now use JavaScript number exclusively, and an empty sum returns 0. The DSL rejects BigInt schemas and bigint operands rather than coercing them.
29
+
30
+ Migrate BigInt fields to Number and bigint literals to numbers. Division now uses number semantics (fractional results rather than bigint truncation); values outside the safe integer range can lose precision. Date support is unchanged.
31
+
32
+ ### Patch Changes
33
+
34
+ - bc114ed: Add npm alpha version, CI status, and license badges to the README.
35
+ - a3583ab: Publish the full project README on npm, including the document-to-conclusion walkthrough.
36
+
3
37
  ## 0.0.0-alpha.1
4
38
 
5
39
  ### Patch Changes
package/README.md CHANGED
@@ -1,44 +1,306 @@
1
1
  # Factbag
2
2
 
3
- A TypeScript library for combining evidence with explicit rules in a knowledgebase.
4
- Facts can be true, false, conflicted, or unknown. Query results keep track of the
5
- sources and rules behind a conclusion.
3
+ [![npm alpha](https://img.shields.io/npm/v/factbag/alpha)](https://www.npmjs.com/package/factbag)
4
+ [![CI](https://img.shields.io/github/actions/workflow/status/factbag/factbag/ci.yml?branch=main&event=push&label=CI)](https://github.com/factbag/factbag/actions/workflows/ci.yml?query=branch%3Amain)
5
+ [![License: Apache-2.0](https://img.shields.io/badge/license-Apache--2.0-blue)](https://github.com/factbag/factbag/blob/main/LICENSE)
6
6
 
7
- Factbag includes a document corpus, permission controls, and an integration with
8
- TypeSafe AI’s Jev model.
7
+ <img src="https://raw.githubusercontent.com/factbag/factbag/main/docs/assets/factbag.png" alt="An open canvas bag" align="right" width="180" />
9
8
 
10
- Read [Introducing Factbag](https://daniel.du-pless.is/introducing-factbag) for how and
11
- why to use it, or [Jev and the Art of Truth Maintenance](https://daniel.du-pless.is/jev-and-the-art-of-truth-maintenance)
12
- for the thinking behind it.
9
+ Turn documents into evidence and conclusions with [Jev](https://typesafe.ai/) and TypeScript.
10
+
11
+ Factbag uses TypeSafe AI’s Jev model to read passages and assess specific claims.
12
+ Those assessments become evidence in a knowledge base, alongside facts from your
13
+ application. You write the rules that connect them and query what follows.
14
+
15
+ Each piece of documentary evidence points back to its source passage. Conclusions
16
+ can be **true**, **false**, **conflicted**, or **unknown**. When documents disagree,
17
+ you can inspect the conflicting evidence, correct a particular fact, and query
18
+ the rules again.
19
+
20
+ > **Important:** Factbag is an early alpha release and still pretty rough - APIs will definitely change.
21
+
22
+ ## How it works
23
+
24
+ 1. **Define the claims** you want Jev to assess, with descriptions of what counts
25
+ as evidence.
26
+ 2. **Add documents** to a corpus and choose which predicates Jev may assert or deny.
27
+ 3. **Execute classification jobs.** Jev assesses whether each passage supports,
28
+ contradicts, or is irrelevant to each candidate claim.
29
+ 4. **Query the result.** Factbag combines admitted evidence with your existing
30
+ facts and rules, retaining the sources behind the conclusion.
13
31
 
14
32
  ## Install
15
33
 
16
34
  ```sh
17
- npm install factbag
35
+ npm install factbag@alpha @typesafe-ai/sdk@^0.6.0
18
36
  ```
19
37
 
20
38
  ESM only. Requires Node 22.18+ on the 22.x line, or Node 24 or 26.
21
- This is an early prerelease; APIs may change.
22
39
 
23
- ## Example
40
+ Set your TypeSafe API key in the environment:
41
+
42
+ ```sh
43
+ export TYPESAFE_API_KEY="your-api-key"
44
+ ```
45
+
46
+ ## From a document to a conclusion
47
+
48
+ ### 1. Decide on your ruleset
49
+
50
+ Start with the criteria your application needs to apply. These might be your own
51
+ business rules or requirements set out in legislation or regulations.
52
+
53
+ Here, we're translating the individual-income route in the SEC's
54
+ [accredited-investor definition, 17 CFR § 230.501(a)(6)](<https://www.ecfr.gov/current/title-17/chapter-II/part-230/section-230.501#p-230.501(a)(6)>),
55
+ into three conditions:
56
+
57
+ - Individual income exceeded $200,000 in the first of the two most recent years.
58
+ - Individual income exceeded $200,000 in the second year.
59
+ - The person reasonably expects to reach the same income level in the current year.
60
+
61
+ ### 2. Define the claims
62
+
63
+ Declare the facts we'll combine, and describe the claims Jev will assess.
24
64
 
25
65
  ```ts
26
- import { predicate, variable } from "factbag/dsl";
66
+ import { and, enumeration, predicate, variable } from "factbag/dsl";
27
67
  import { KnowledgeBase } from "factbag/kb";
28
68
  import { query } from "factbag/query";
69
+ import { Corpus } from "factbag/corpus";
70
+ import { guard } from "factbag/guards";
71
+ import { classifier } from "factbag/jev/classifier";
72
+
73
+ class Person {
74
+ constructor(
75
+ readonly id: string,
76
+ readonly name: string,
77
+ ) {}
78
+ }
79
+
80
+ const hasIncomeOver200k = predicate(
81
+ "has_income_over_200k",
82
+ { person: Person, year: enumeration("2024", "2025") },
83
+ {
84
+ description:
85
+ "The person's total individual annual income exceeded USD 200,000 in the " +
86
+ "specified year. Joint household income and forecasts do not count.",
87
+ },
88
+ );
89
+
90
+ const expectsSameIncome = predicate(
91
+ "expects_same_income",
92
+ { person: Person },
93
+ {
94
+ description:
95
+ "The person reasonably expects their individual income in 2026 to remain " +
96
+ "at least at the level earned in each of 2024 and 2025.",
97
+ },
98
+ );
99
+
100
+ const isQualifiedInvestor = predicate("is_qualified_investor", {
101
+ person: Person,
102
+ });
103
+ const hasVerifiedIdentity = predicate("has_verified_identity", {
104
+ person: Person,
105
+ });
106
+ const isReadyForReview = predicate("is_ready_for_review", { person: Person });
107
+ ```
29
108
 
30
- const emailVerified = predicate("email_verified", { user: String });
31
- const canSignIn = predicate("can_sign_in", { user: String });
32
- const user = variable("user");
109
+ ### 3. Encode the ruleset
110
+
111
+ All three income conditions must hold for this route to qualification. Our
112
+ application also requires verified identity before sending an investor for review.
113
+
114
+ ```ts
115
+ const person = variable("person");
33
116
  const kb = new KnowledgeBase();
34
117
 
35
- kb.rule(canSignIn({ user })).if(emailVerified({ user }));
36
- kb.asserts(emailVerified({ user: "alice" }), { source: "accounts" });
118
+ kb.rule(isQualifiedInvestor({ person })).if(
119
+ and(
120
+ hasIncomeOver200k({ person, year: "2024" }),
121
+ hasIncomeOver200k({ person, year: "2025" }),
122
+ expectsSameIncome({ person }),
123
+ ),
124
+ );
37
125
 
38
- query(kb, canSignIn({ user: "alice" })).value; // "true"
126
+ kb.rule(isReadyForReview({ person })).if(
127
+ and(isQualifiedInvestor({ person }), hasVerifiedIdentity({ person })),
128
+ );
39
129
  ```
40
130
 
41
- For Jev, also install `@typesafe-ai/sdk@^0.6.0`. Other parts of Factbag do not
42
- require it. Live Jev requests need a TypeSafe API key.
131
+ ### 4. Import a known fact and add documents
132
+
133
+ Identity verification comes directly from our onboarding system. Income evidence
134
+ goes into the corpus for Jev to assess.
135
+
136
+ ```ts
137
+ const alice = new Person("alice", "Alice");
138
+ kb.entities.add(alice);
139
+
140
+ kb.asserts(hasVerifiedIdentity({ person: alice }), {
141
+ source: "onboarding:alice",
142
+ });
143
+
144
+ const corpus = new Corpus();
145
+
146
+ // Tax returns provide evidence of past annual income.
147
+ corpus.add({ documentId: "alice-2024-federal-tax-return", text: "…" });
148
+ corpus.add({ documentId: "alice-2025-federal-tax-return", text: "…" });
149
+
150
+ // An employer letter supports the expectation of continued income.
151
+ corpus.add({
152
+ documentId: "alice-2026-employment-confirmation-letter",
153
+ text: "…",
154
+ });
155
+ ```
156
+
157
+ ### 5. Set Jev's permissions
158
+
159
+ The guard starts with no permissions. Permit Jev to assert or deny the two income
160
+ predicates.
161
+
162
+ ```ts
163
+ const incomeEvidence = guard(kb)
164
+ .permit("assert", hasIncomeOver200k)
165
+ .permit("deny", hasIncomeOver200k)
166
+ .permit("assert", expectsSameIncome)
167
+ .permit("deny", expectsSameIncome);
168
+ ```
169
+
170
+ ### 6. Ask Jev to assess the income evidence
171
+
172
+ ```ts
173
+ const run = classifier(corpus, incomeEvidence);
174
+
175
+ for (const job of run) {
176
+ await job.execute(); // Makes a live Jev request.
177
+ }
178
+ ```
179
+
180
+ ### 7. Query the conclusion
181
+
182
+ The output knowledge base combines the imported fact and rules with the admitted
183
+ evidence.
184
+
185
+ ```ts
186
+ const qualification = query(run.kb, isQualifiedInvestor({ person: alice }));
187
+ const reviewReadiness = query(run.kb, isReadyForReview({ person: alice }));
188
+
189
+ // Both are "true" if Jev admits supporting evidence for all three income claims.
190
+ qualification.value; // "true"
191
+ reviewReadiness.value; // "true"
192
+ ```
193
+
194
+ Results can be `true`, `false`, `conflicted`, or
195
+ `unknown`, depending on the evidence admitted.
196
+
197
+ ### 8. Inspect the proof tree
198
+
199
+ Use `explainJudgment()` to follow a conclusion through its rule to the supporting
200
+ evidence and source passages.
201
+
202
+ ```ts
203
+ import { explainJudgment } from "factbag/explain";
204
+
205
+ const proof = explainJudgment(qualification);
206
+ JSON.stringify(proof, null, 2); // JSON shown below.
207
+ ```
208
+
209
+ An abbreviated example with supporting evidence for all three claims:
210
+
211
+ ```json
212
+ {
213
+ "claim": "is_qualified_investor(alice) = true",
214
+ "because": "is_qualified_investor(person) if and(...), with person = alice",
215
+ "given": [
216
+ {
217
+ "claim": "and(...) = true",
218
+ "given": [
219
+ {
220
+ "claim": "has_income_over_200k(alice, 2024) = true",
221
+ "because": {
222
+ "claim": "has_income_over_200k(alice, 2024) = true",
223
+ "source": "alice-2024-federal-tax-return#0-1"
224
+ }
225
+ },
226
+ {
227
+ "claim": "has_income_over_200k(alice, 2025) = true",
228
+ "because": {
229
+ "claim": "has_income_over_200k(alice, 2025) = true",
230
+ "source": "alice-2025-federal-tax-return#0-1"
231
+ }
232
+ },
233
+ {
234
+ "claim": "expects_same_income(alice) = true",
235
+ "because": {
236
+ "claim": "expects_same_income(alice) = true",
237
+ "source": "alice-2026-employment-confirmation-letter#0-1"
238
+ }
239
+ }
240
+ ]
241
+ }
242
+ ]
243
+ }
244
+ ```
245
+
246
+ Source IDs identify document passages; their ranges depend on the document text.
247
+ Use `corpus.resolve(source)` to retrieve the passage behind a piece of evidence.
248
+
249
+ ```ts
250
+ const source = "alice-2024-federal-tax-return#0-1";
251
+ corpus.resolve(source); // Object shown below.
252
+ ```
253
+
254
+ The result includes the exact passage and its containing document:
255
+
256
+ ```json
257
+ {
258
+ "chunk": {
259
+ "id": "alice-2024-federal-tax-return#0-1",
260
+ "documentId": "alice-2024-federal-tax-return",
261
+ "range": { "start": 0, "end": 1 },
262
+ "text": "…"
263
+ },
264
+ "document": {
265
+ "documentId": "alice-2024-federal-tax-return",
266
+ "text": "…",
267
+ "metadata": {}
268
+ }
269
+ }
270
+ ```
271
+
272
+ ### 9. Review conflicting evidence (optional)
273
+
274
+ Suppose an original tax return puts Alice below the income threshold, but an
275
+ amended return puts her above it. Jev admits both accounts, leaving qualification
276
+ `conflicted`. In this example, the other income conditions have supporting evidence.
277
+
278
+ You could pass the conflicting evidence to a human for review, or ask Jev to
279
+ compare the sources using the `resolver`.
280
+
281
+ Evidence judged superseded is retracted from the resolver's output
282
+ knowledge base.
283
+
284
+ ```ts
285
+ import { resolver } from "factbag/jev/resolve";
286
+
287
+ query(run.kb, isQualifiedInvestor({ person: alice })).value; // "conflicted"
288
+
289
+ const review = resolver(corpus, run.kb);
290
+
291
+ for (const job of review) {
292
+ await job.execute();
293
+ }
294
+
295
+ // If the resolver retracts the superseded return's denial:
296
+ query(review.kb, isQualifiedInvestor({ person: alice })).value; // "true"
297
+ ```
298
+
299
+ ## Go further
300
+
301
+ - [Introducing Factbag](https://daniel.du-pless.is/introducing-factbag): how and why
302
+ to use it.
303
+ - [Jev and the Art of Truth Maintenance](https://daniel.du-pless.is/jev-and-the-art-of-truth-maintenance):
304
+ the thinking behind the project.
43
305
 
44
306
  Apache-2.0 licensed.
@@ -13,7 +13,10 @@ var PrimitiveArithmeticProgrammerError = class extends ProgrammerError {
13
13
  /** The arithmetic goal being evaluated when the failure occurred. */
14
14
  goal;
15
15
  constructor(operation, operands, goal) {
16
- super(`${operation} expected numeric operands`);
16
+ const rounding = operation === "roundTo" || operation === "floorTo" || operation === "ceilTo";
17
+ super(
18
+ `${operation} expected ${operation === "daysBetween" || operation === "yearsBetween" ? "date" : "numeric"} operands${rounding ? " and a whole number of decimal places" : ""}`
19
+ );
17
20
  this.name = "PrimitiveArithmeticProgrammerError";
18
21
  this.operation = operation;
19
22
  this.operands = operands;
@@ -50,7 +50,7 @@ function fieldIdentityToken(value) {
50
50
  return ["entity", value.constructor.name, value.id];
51
51
  }
52
52
  if (typeof value === "bigint") {
53
- return ["bigint", value.toString()];
53
+ throw new TypeError("bigint is not supported; use number values");
54
54
  }
55
55
  return value;
56
56
  }
@@ -82,5 +82,6 @@ export {
82
82
  isAtomForPredicate,
83
83
  atomEquals,
84
84
  fieldEquals,
85
- fieldIdentityToken
85
+ fieldIdentityToken,
86
+ isEntityValue
86
87
  };
@@ -0,0 +1,2 @@
1
+ /** Reject removed scalar types before building rules or atoms. */
2
+ export declare function assertSupportedValue(value: unknown): void;
@@ -6,9 +6,9 @@
6
6
  */
7
7
  export { and, exists, letIn, not, or } from "./formula.js";
8
8
  export { count, sum } from "./operators/aggregate.js";
9
- export type { AggregateBuilder, ComparableTerm, DateTerm, NumericTerm } from "./types.js";
10
- export { add, div, mul, sub } from "./operators/arithmetic.js";
11
- export { gt, gte, lte } from "./operators/comparison.js";
9
+ export type { AggregateBuilder, ComparableTerm, DateTerm, EqualityTerm, NumericTerm, } from "./types.js";
10
+ export { add, ceilTo, daysBetween, div, floorTo, max, min, mul, roundTo, sub, yearsBetween, } from "./operators/arithmetic.js";
11
+ export { eq, gt, gte, lte, neq } from "./operators/comparison.js";
12
12
  export { closedPredicate, predicate } from "./predicate.js";
13
13
  export { enumeration } from "./enumeration.js";
14
14
  export type { ClosedWorldPredicate, OpenWorldPredicate, Predicate, PredicateBase, PredicateDeclaration, PredicateMetadata, PredicateWorld, } from "../types/predicate.js";
@@ -1,11 +1,18 @@
1
1
  import {
2
2
  assertAtomLike,
3
3
  assertGroundAtom
4
- } from "../chunk-4IP7CXNV.js";
4
+ } from "../chunk-RMQ3YX7V.js";
5
5
  import {
6
6
  Var
7
7
  } from "../chunk-UVOCOH2I.js";
8
8
 
9
+ // src/dsl/assert-supported-value.ts
10
+ function assertSupportedValue(value) {
11
+ if (typeof value === "bigint" || value === BigInt) {
12
+ throw new TypeError("bigint is not supported; use number values and Number fields");
13
+ }
14
+ }
15
+
9
16
  // src/dsl/formula.ts
10
17
  function not(atom) {
11
18
  return { sign: "-", atom };
@@ -20,6 +27,7 @@ function exists(...args) {
20
27
  return { kind: "exists", args };
21
28
  }
22
29
  function letIn(variable2, value, body) {
30
+ assertSupportedValue(value);
23
31
  return { kind: "letIn", args: [variable2, value, body] };
24
32
  }
25
33
 
@@ -31,6 +39,7 @@ function count(value) {
31
39
  return aggregateBuilder("count", value);
32
40
  }
33
41
  function aggregateBuilder(kind, value) {
42
+ assertSupportedValue(value);
34
43
  const where = (body, options) => ({
35
44
  kind,
36
45
  args: options === void 0 ? [value, body] : [value, body, options]
@@ -40,37 +49,98 @@ function aggregateBuilder(kind, value) {
40
49
 
41
50
  // src/dsl/operators/arithmetic.ts
42
51
  function add(left, right) {
52
+ assertSupportedValue(left);
53
+ assertSupportedValue(right);
43
54
  return { kind: "add", args: [left, right] };
44
55
  }
45
56
  function sub(left, right) {
57
+ assertSupportedValue(left);
58
+ assertSupportedValue(right);
46
59
  return { kind: "sub", args: [left, right] };
47
60
  }
48
61
  function mul(left, right) {
62
+ assertSupportedValue(left);
63
+ assertSupportedValue(right);
49
64
  return { kind: "mul", args: [left, right] };
50
65
  }
51
66
  function div(left, right) {
67
+ assertSupportedValue(left);
68
+ assertSupportedValue(right);
52
69
  return { kind: "div", args: [left, right] };
53
70
  }
71
+ function daysBetween(later, earlier) {
72
+ assertSupportedValue(later);
73
+ assertSupportedValue(earlier);
74
+ return { kind: "daysBetween", args: [later, earlier] };
75
+ }
76
+ function yearsBetween(later, earlier) {
77
+ assertSupportedValue(later);
78
+ assertSupportedValue(earlier);
79
+ return { kind: "yearsBetween", args: [later, earlier] };
80
+ }
81
+ function min(left, right) {
82
+ assertSupportedValue(left);
83
+ assertSupportedValue(right);
84
+ return { kind: "min", args: [left, right] };
85
+ }
86
+ function max(left, right) {
87
+ assertSupportedValue(left);
88
+ assertSupportedValue(right);
89
+ return { kind: "max", args: [left, right] };
90
+ }
91
+ function roundTo(value, places) {
92
+ assertSupportedValue(value);
93
+ assertSupportedValue(places);
94
+ return { kind: "roundTo", args: [value, places] };
95
+ }
96
+ function floorTo(value, places) {
97
+ assertSupportedValue(value);
98
+ assertSupportedValue(places);
99
+ return { kind: "floorTo", args: [value, places] };
100
+ }
101
+ function ceilTo(value, places) {
102
+ assertSupportedValue(value);
103
+ assertSupportedValue(places);
104
+ return { kind: "ceilTo", args: [value, places] };
105
+ }
54
106
 
55
107
  // src/dsl/operators/comparison.ts
108
+ function eq(left, right) {
109
+ assertSupportedValue(left);
110
+ assertSupportedValue(right);
111
+ return { kind: "eq", args: [left, right] };
112
+ }
113
+ function neq(left, right) {
114
+ assertSupportedValue(left);
115
+ assertSupportedValue(right);
116
+ return { kind: "neq", args: [left, right] };
117
+ }
56
118
  function gt(left, right) {
119
+ assertSupportedValue(left);
120
+ assertSupportedValue(right);
57
121
  return { kind: "gt", args: [left, right] };
58
122
  }
59
123
  function gte(left, right) {
124
+ assertSupportedValue(left);
125
+ assertSupportedValue(right);
60
126
  return { kind: "gte", args: [left, right] };
61
127
  }
62
128
  function lte(left, right) {
129
+ assertSupportedValue(left);
130
+ assertSupportedValue(right);
63
131
  return { kind: "lte", args: [left, right] };
64
132
  }
65
133
 
66
134
  // src/dsl/predicate.ts
67
135
  function makePredicate(name, params, world, metadata) {
136
+ Object.values(params).forEach(assertSupportedValue);
68
137
  const holder = {};
69
138
  const construct = (fields) => {
70
139
  const current = holder.current;
71
140
  if (current === void 0) {
72
141
  throw new Error(`Predicate ${name} was called before initialization`);
73
142
  }
143
+ Object.values(fields).forEach(assertSupportedValue);
74
144
  return { predicate: current, fields };
75
145
  };
76
146
  const declared = construct;
@@ -113,20 +183,29 @@ export {
113
183
  and,
114
184
  assertAtomLike,
115
185
  assertGroundAtom,
186
+ ceilTo,
116
187
  closedPredicate,
117
188
  count,
189
+ daysBetween,
118
190
  div,
119
191
  enumeration,
192
+ eq,
120
193
  exists,
194
+ floorTo,
121
195
  gt,
122
196
  gte,
123
197
  letIn,
124
198
  lte,
199
+ max,
200
+ min,
125
201
  mul,
202
+ neq,
126
203
  not,
127
204
  or,
128
205
  predicate,
206
+ roundTo,
129
207
  sub,
130
208
  sum,
131
- variable
209
+ variable,
210
+ yearsBetween
132
211
  };
@@ -21,7 +21,7 @@ import type { AggregateBuilder, NumericTerm, Term } from "../types.js";
21
21
  * );
22
22
  * ```
23
23
  */
24
- export declare function sum(value: NumericTerm): AggregateBuilder<number | bigint>;
24
+ export declare function sum(value: NumericTerm): AggregateBuilder;
25
25
  /**
26
26
  * Count the rows matched by the aggregate body. Runs through the same
27
27
  * aggregate machinery as {@link sum}, but thresholds the number of matching
@@ -1,6 +1,8 @@
1
1
  import type { DateTerm, Formula, NumericTerm } from "../types.js";
2
2
  /** Add two numeric terms, producing a numeric rule-body expression. */
3
- export declare function add(left: NumericTerm, right: NumericTerm): Formula<number | bigint, readonly [NumericTerm, NumericTerm]>;
3
+ export declare function add(left: NumericTerm, right: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
4
+ /** Subtract one numeric term from another. */
5
+ export declare function sub(left: NumericTerm, right: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
4
6
  /**
5
7
  * Subtract a numeric duration in milliseconds from a date term, producing a
6
8
  * date expression.
@@ -19,9 +21,74 @@ export declare function add(left: NumericTerm, right: NumericTerm): Formula<numb
19
21
  * ```
20
22
  */
21
23
  export declare function sub(left: DateTerm, right: NumericTerm): Formula<Date, readonly [DateTerm, NumericTerm]>;
22
- /** Subtract one numeric term from another. */
23
- export declare function sub(left: NumericTerm, right: NumericTerm): Formula<number | bigint, readonly [NumericTerm, NumericTerm]>;
24
24
  /** Multiply two numeric terms, producing a numeric rule-body expression. */
25
- export declare function mul(left: NumericTerm, right: NumericTerm): Formula<number | bigint, readonly [NumericTerm, NumericTerm]>;
25
+ export declare function mul(left: NumericTerm, right: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
26
26
  /** Divide one numeric term by another, producing a numeric rule-body expression. */
27
- export declare function div(left: NumericTerm, right: NumericTerm): Formula<number | bigint, readonly [NumericTerm, NumericTerm]>;
27
+ export declare function div(left: NumericTerm, right: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
28
+ /**
29
+ * Count signed UTC calendar days from `earlier` to `later`, ignoring time of day.
30
+ * Add one to count both the first and last day of a stay.
31
+ */
32
+ export declare function daysBetween(later: DateTerm, earlier: DateTerm): Formula<number, readonly [DateTerm, DateTerm]>;
33
+ /**
34
+ * Count signed whole years from `earlier` to `later` by UTC calendar anniversaries,
35
+ * ignoring time of day. February 29 anniversaries fall on March 1 in common years.
36
+ * A person's age on a day is `yearsBetween(day, birthday)`.
37
+ */
38
+ export declare function yearsBetween(later: DateTerm, earlier: DateTerm): Formula<number, readonly [DateTerm, DateTerm]>;
39
+ /**
40
+ * The lesser of two numeric terms.
41
+ *
42
+ * @example
43
+ * ```ts
44
+ * // Deduction reduction cannot exceed 80% of the deductions.
45
+ * letIn(R, min(mul(0.03, Excess), mul(0.8, Deductions)), gte(R, 0))
46
+ * ```
47
+ */
48
+ export declare function min(left: NumericTerm, right: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
49
+ /**
50
+ * The greater of two numeric terms.
51
+ *
52
+ * @example
53
+ * ```ts
54
+ * // The remaining allowance cannot be negative.
55
+ * letIn(A, max(sub(2000, Reduction), 0), gte(A, 0))
56
+ * ```
57
+ */
58
+ export declare function max(left: NumericTerm, right: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
59
+ /**
60
+ * Round to a whole number of decimal places, with halves away from zero.
61
+ * Negative places round to tens, hundreds, and so on. Decimal exponent shifting
62
+ * avoids multiplication artifacts such as `1.005 * 100`, so `roundTo(1.005, 2)`
63
+ * yields `1.01`. Inputs and results still have JavaScript number precision;
64
+ * this does not recover precision already lost in an earlier expression.
65
+ *
66
+ * @example
67
+ * ```ts
68
+ * // Tax is 15% of taxable income, in whole dollars.
69
+ * letIn(Tax, roundTo(mul(Income, 0.15), 0), gte(Tax, 0))
70
+ * ```
71
+ */
72
+ export declare function roundTo(value: NumericTerm, places: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
73
+ /**
74
+ * Round toward negative infinity at a whole number of decimal places.
75
+ * Negative places and decimal exponent shifting work as in {@link roundTo}.
76
+ *
77
+ * @example
78
+ * ```ts
79
+ * // Only complete thousands of eligible expenses count.
80
+ * letIn(Eligible, floorTo(Expenses, -3), gte(Eligible, 0))
81
+ * ```
82
+ */
83
+ export declare function floorTo(value: NumericTerm, places: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;
84
+ /**
85
+ * Round toward positive infinity at a whole number of decimal places.
86
+ * Negative places and decimal exponent shifting work as in {@link roundTo}.
87
+ *
88
+ * @example
89
+ * ```ts
90
+ * // Two percentage points for each $2,500, or fraction thereof, of the excess.
91
+ * letIn(Points, mul(2, ceilTo(div(Excess, 2500), 0)), lte(Points, 100))
92
+ * ```
93
+ */
94
+ export declare function ceilTo(value: NumericTerm, places: NumericTerm): Formula<number, readonly [NumericTerm, NumericTerm]>;