a9a7a328e6
Mechanical rename produced by scripts/msd-rename.cjs: gsd/Gsd/GSD -> msd/Msd/MSD across contents and paths, upstream package/repo coordinates -> @golem15/msd-core and golem15com/msd-core. Deep links into upstream history, sibling upstream packages, the GSD-2 import feature, CHANGELOG.md and .changeset/ are kept as-is. Hand edits on top: MSD block-letter banner and logos, LICENSE copyright line, package/plugin identity, regenerated lockfile, install-tree fixtures, derived registries and benchmark baseline; migration checksum baseline re-locked (MSD keeps its own install state, so no install had applied the old sums); sort-order and regex-escaped expectations in tests adjusted.
104 lines
4.2 KiB
JavaScript
104 lines
4.2 KiB
JavaScript
'use strict';
|
|
|
|
/**
|
|
* prompt-budget-parity.test.cjs
|
|
*
|
|
* Characterization-parity suite for issue #2929 (epic #1671 Phase 2 —
|
|
* context-composer). `tests/fixtures/prompt-budget-parity/corpus.json` was
|
|
* captured from the PRE-REFACTOR `prompt-budget.applyBudget` implementation
|
|
* by `scripts/gen-prompt-budget-parity-corpus.cjs` (see that script's header
|
|
* for the fixture-provenance rationale). It is the oracle for behavioral
|
|
* parity after the composer-seam refactor: every case's `prompt` and
|
|
* `metadata` must reproduce byte-for-byte / field-for-field.
|
|
*
|
|
* Regenerating this corpus to make a failure here go away DEFEATS ITS
|
|
* PURPOSE — the corpus is frozen precisely so that a disagreement always
|
|
* means the refactor moved observable behavior, never the other way around.
|
|
* If a case fails, fix the refactor, not the fixture.
|
|
*/
|
|
|
|
const { describe, test } = require('node:test');
|
|
const assert = require('node:assert/strict');
|
|
|
|
const { applyBudget, estimateTokens } = require('../msd-core/bin/lib/prompt-budget.cjs');
|
|
const corpus = require('./fixtures/prompt-budget-parity/corpus.json');
|
|
|
|
describe('prompt-budget parity vs pre-refactor corpus', () => {
|
|
test('refactoredOutputIsByteIdenticalToCapturedCorpus', () => {
|
|
for (const c of corpus.cases) {
|
|
const result = applyBudget({
|
|
sections: c.input.sections,
|
|
budget: c.input.budget,
|
|
options: c.input.options,
|
|
});
|
|
assert.equal(
|
|
result.prompt,
|
|
c.expected.prompt,
|
|
`case "${c.name}": prompt mismatch (got ${result.prompt.length} chars, want ${c.expected.prompt.length} chars)`
|
|
);
|
|
}
|
|
});
|
|
|
|
test('refactoredMetadataMatchesCapturedCorpus', () => {
|
|
for (const c of corpus.cases) {
|
|
const result = applyBudget({
|
|
sections: c.input.sections,
|
|
budget: c.input.budget,
|
|
options: c.input.options,
|
|
});
|
|
assert.deepStrictEqual(
|
|
result.metadata,
|
|
c.expected.metadata,
|
|
`case "${c.name}": metadata mismatch`
|
|
);
|
|
}
|
|
});
|
|
|
|
test('parityCorpusCoversEveryParityCriticalRow', () => {
|
|
const cases = corpus.cases;
|
|
assert.ok(cases.length >= 45, `expected >= 45 corpus cases, got ${cases.length}`);
|
|
|
|
const has = (pred, msg) => assert.ok(cases.some(pred), msg);
|
|
|
|
has((c) => c.expected.metadata.hardFailed === true, 'expected >=1 case with hardFailed===true');
|
|
has((c) => c.expected.metadata.hardFailed === false, 'expected >=1 case with hardFailed===false');
|
|
has((c) => c.expected.metadata.noteInjected === true, 'expected >=1 case with noteInjected===true');
|
|
has((c) => c.expected.metadata.noteInjected === false, 'expected >=1 case with noteInjected===false');
|
|
has((c) => c.expected.metadata.omitted.length > 0, 'expected >=1 case with omitted.length > 0');
|
|
has((c) => c.expected.metadata.projectMdShrunk === true, 'expected >=1 case with projectMdShrunk===true');
|
|
has(
|
|
(c) => c.expected.metadata.planTruncationPct > 0,
|
|
'expected >=1 case with planTruncationPct > 0'
|
|
);
|
|
|
|
const a4 = cases.filter((c) => c.name.startsWith('A4-'));
|
|
const a10 = cases.filter((c) => c.name.startsWith('A10-'));
|
|
assert.ok(a4.length >= 8, `expected >= 8 A4- cases, got ${a4.length}`);
|
|
assert.ok(a10.length >= 6, `expected >= 6 A10- cases, got ${a10.length}`);
|
|
|
|
const noTrimOccurred = (c) =>
|
|
c.expected.metadata.omitted.length === 0 &&
|
|
c.expected.metadata.projectMdShrunk === false &&
|
|
c.expected.metadata.planTruncationPct === 0;
|
|
|
|
for (const c of a4) {
|
|
assert.ok(noTrimOccurred(c), `A4 case "${c.name}" must have NO trim of any kind`);
|
|
}
|
|
for (const c of a10) {
|
|
assert.ok(!noTrimOccurred(c), `A10 case "${c.name}" must have at least one trim`);
|
|
}
|
|
});
|
|
|
|
test('estimateTokensRemainsExportedAndUnchanged', () => {
|
|
// src/phase-estimation.cts re-exports this function as `measureTokens`,
|
|
// and CONTEXT.md pins downstream estimate/actual comparisons to this
|
|
// exact chars/4 scale — its math is frozen, not merely conventional.
|
|
assert.equal(typeof estimateTokens, 'function');
|
|
assert.equal(estimateTokens(''), 0);
|
|
assert.equal(estimateTokens(null), 0);
|
|
assert.equal(estimateTokens(undefined), 0);
|
|
assert.equal(estimateTokens('abcd'), 1);
|
|
assert.equal(estimateTokens('abcde'), 2);
|
|
});
|
|
});
|