diff --git a/session_test.ts b/session_test.ts
new file mode 100644
index 0000000..05bee3e
--- /dev/null
+++ b/session_test.ts
@@ -0,0 +1,397 @@
+/**
+ * Tests for session.ts.
+ */
+// deno-lint-ignore-file no-import-prefix no-unversioned-import
+import { describe, it } from 'jsr:@std/testing/bdd';
+import { expect } from 'jsr:@std/expect';
+import * as fc from 'npm:fast-check';
+
+import {
+  odd_character_wikitext_string,
+  pathological_wikitext_string,
+  spacing_heavy_wikitext_string,
+  wikiish_string,
+} from './_test_utils/arbitraries.ts';
+import { UNICODE_TEXT_FIXTURES } from './_test_utils/unicode_fixtures.ts';
+import { DiagnosticCode } from './events.ts';
+import { createSession } from './session.ts';
+import { events, outlineEvents, parse, parseWithDiagnostics, parseWithRecovery } from './parse.ts';
+import type { TextSource } from './text_source.ts';
+
+const SESSION_FIXTURES = [
+  [
+    '== Lead ==',
+    `A [[Main Page|home]] link with ''italic'' and '''bold'''.`,
+    '',
+    '* Bullet one',
+    '* Bullet two with {{Card|title=Plan|body=value}}',
+    '',
+    '{|',
+    '! Planet !! Notes',
+    '|-',
+    '| Mars || <ref name="m1">Known as the red planet</ref>',
+    '|}',
+  ].join('\n'),
+  [
+    'Paragraph with <nowiki>[[literal]]</nowiki> text and &amp; entity.',
+    'Second line keeps the paragraph open until a blank line.',
+    '',
+    '; Term',
+    ': Description with [https://example.com label]',
+  ].join('\n'),
+  [
+    '[[[[{{{{<!--',
+    '__BROKEN_',
+    '<ref name="n">',
+    "'''''",
+    '&broken',
+    '{|',
+    '| [[Page|{{T|x}}]] || <span class="x">text',
+  ].join('\n'),
+  'Paragraph with <ref name="n"',
+] as const;
+
+type CountedSource = TextSource & {
+  readonly counters: {
+    char_code_at_calls: number;
+    slice_calls: number;
+  };
+};
+
+function createCountedSource(value: string): CountedSource {
+  const counters = {
+    char_code_at_calls: 0,
+    slice_calls: 0,
+  };
+
+  return {
+    get length() {
+      return value.length;
+    },
+
+    charCodeAt(index: number): number {
+      counters.char_code_at_calls++;
+      return value.charCodeAt(index);
+    },
+
+    slice(start: number, end: number): string {
+      counters.slice_calls++;
+      return value.slice(start, end);
+    },
+
+    counters,
+  };
+}
+
+describe('createSession()', () => {
+  it('exposes the original source through session.source', () => {
+    const source = createCountedSource('== Title ==\n\nText');
+    const session = createSession(source);
+
+    expect(session.source).toBe(source);
+  });
+
+  it('returns cached outline and full event streams', () => {
+    const input = '== Title ==\n\nA [[Page|home]] link';
+    const session = createSession(input);
+
+    expect(Array.from(session.outline())).toEqual(Array.from(outlineEvents(input)));
+    expect(Array.from(session.events())).toEqual(Array.from(events(input)));
+  });
+
+  it('reuses the cached outline parse on repeated outline() calls', () => {
+    const source = createCountedSource('== Title ==\n\nA paragraph line');
+    const session = createSession(source);
+
+    const first = Array.from(session.outline());
+    const first_char_calls = source.counters.char_code_at_calls;
+    const second = Array.from(session.outline());
+
+    expect(second).toEqual(first);
+    expect(source.counters.char_code_at_calls).toBe(first_char_calls);
+  });
+
+  it('reuses the cached full event stream on repeated events() calls', () => {
+    const source = createCountedSource('A [[Page|home]] link with {{Template|one}}');
+    const session = createSession(source);
+
+    const first = Array.from(session.events());
+    const first_char_calls = source.counters.char_code_at_calls;
+    const second = Array.from(session.events());
+
+    expect(second).toEqual(first);
+    expect(source.counters.char_code_at_calls).toBe(first_char_calls);
+  });
+
+  it('parse() after events() reuses cached events and only materializes tree', () => {
+    const source = createCountedSource('A [[Page|home]] link and plain text');
+    const session = createSession(source);
+
+    Array.from(session.events());
+    const char_calls_before_parse = source.counters.char_code_at_calls;
+    const slice_calls_before_parse = source.counters.slice_calls;
+
+    const tree = session.parse();
+
+    expect(tree.type).toBe('root');
+    expect(source.counters.char_code_at_calls).toBe(char_calls_before_parse);
+    expect(source.counters.slice_calls).toBeGreaterThan(slice_calls_before_parse);
+  });
+
+  it('caches the parsed tree object', () => {
+    const session = createSession("A '''bold''' word");
+    const first = session.parse();
+    const second = session.parse();
+
+    expect(first).toBe(second);
+    expect(first).toEqual(parse("A '''bold''' word"));
+  });
+
+  it('matches the stateless pipeline across representative mixed inputs', () => {
+    for (const input of SESSION_FIXTURES) {
+      const session = createSession(input);
+
+      const expected_outline = Array.from(outlineEvents(input));
+      const expected_events = Array.from(events(input));
+      const expected_tree = parse(input);
+
+      expect(Array.from(session.outline())).toEqual(expected_outline);
+      expect(Array.from(session.outline())).toEqual(expected_outline);
+      expect(Array.from(session.events())).toEqual(expected_events);
+      expect(Array.from(session.events())).toEqual(expected_events);
+      expect(session.parse()).toEqual(expected_tree);
+      expect(session.parse()).toEqual(expected_tree);
+    }
+  });
+
+  it('matches the stateless diagnostics API on both uncached and cached calls', () => {
+    for (const input of SESSION_FIXTURES) {
+      const session = createSession(input);
+      const expected_diagnostics = parseWithDiagnostics(input);
+
+      expect(session.parseWithDiagnostics()).toEqual(expected_diagnostics);
+      expect(session.parseWithDiagnostics()).toEqual(expected_diagnostics);
+    }
+  });
+
+  it('matches the stateless recovery API on both uncached and cached calls', () => {
+    for (const input of SESSION_FIXTURES) {
+      const session = createSession(input);
+      const expected_recovery = parseWithRecovery(input);
+
+      expect(session.parseWithRecovery()).toEqual(expected_recovery);
+      expect(session.parseWithRecovery()).toEqual(expected_recovery);
+    }
+  });
+
+  it('keeps diagnostics and recovery lanes distinct when queried on the same session', () => {
+    for (const input of SESSION_FIXTURES) {
+      const diagnostics_first = createSession(input);
+      const recovery_first = createSession(input);
+
+      expect(diagnostics_first.parseWithDiagnostics()).toEqual(parseWithDiagnostics(input));
+      expect(diagnostics_first.parseWithRecovery()).toEqual(parseWithRecovery(input));
+
+      expect(recovery_first.parseWithRecovery()).toEqual(parseWithRecovery(input));
+      expect(recovery_first.parseWithDiagnostics()).toEqual(parseWithDiagnostics(input));
+    }
+  });
+
+  it('keeps representative Unicode text classes stable across cached session layers', () => {
+    for (const fixture of UNICODE_TEXT_FIXTURES) {
+      const input = [
+        `== ${fixture.sample} ==`,
+        fixture.sample,
+        `* ${fixture.sample}`,
+        '',
+      ].join('\n');
+      const session = createSession(input);
+      const diagnostic_result = session.parseWithDiagnostics();
+
+      // This mirrors the stateless Unicode matrix, but through the cache
+      // wrapper. The goal is to prove that cached outline/events/tree lanes do
+      // not accidentally normalize, drop, or misclassify unusual payload text.
+      expect(Array.from(session.outline())).toEqual(Array.from(outlineEvents(input)));
+      expect(Array.from(session.events())).toEqual(Array.from(events(input)));
+      expect(session.parse()).toEqual(parse(input));
+      expect(diagnostic_result).toEqual(parseWithDiagnostics(input));
+      expect(diagnostic_result.diagnostics).toEqual([]);
+    }
+  });
+
+  it('keeps representative Unicode text classes stable in repeated session stress inputs', () => {
+    for (const fixture of UNICODE_TEXT_FIXTURES) {
+      const input = `${fixture.sample}\n`.repeat(512);
+      const session = createSession(input);
+
+      // Repeated plain-text lines are a useful session check because they keep
+      // the block structure simple while forcing the caches to replay a lot of
+      // line tracking and paragraph continuation work.
+      expect(session.parseWithDiagnostics().diagnostics).toEqual([]);
+      expect(session.parse().type).toBe('root');
+      expect(Array.from(session.events()).some((event) => event.kind === 'text')).toBe(true);
+    }
+  });
+
+  it('keeps cached results stable across different consumer access orders', () => {
+    const input = SESSION_FIXTURES[0];
+    const source = createCountedSource(input);
+    const session = createSession(source);
+
+    const first_tree = session.parse();
+    const char_calls_after_parse = source.counters.char_code_at_calls;
+    const outline = Array.from(session.outline());
+    const full_events = Array.from(session.events());
+
+    expect(first_tree).toEqual(parse(input));
+    expect(outline).toEqual(Array.from(outlineEvents(input)));
+    expect(full_events).toEqual(Array.from(events(input)));
+    expect(source.counters.char_code_at_calls).toBeGreaterThan(char_calls_after_parse);
+  });
+
+  it('keeps the cheap parse lane separate from diagnostics-enabled event caches', () => {
+    const input = 'Paragraph with <ref name="cite-1">note';
+    const source = createCountedSource(input);
+    const session = createSession(source);
+
+    session.parse();
+    const char_calls_after_parse = source.counters.char_code_at_calls;
+
+    Array.from(session.events());
+    const char_calls_after_first_events = source.counters.char_code_at_calls;
+    Array.from(session.events());
+
+    expect(char_calls_after_first_events).toBeGreaterThan(char_calls_after_parse);
+    expect(source.counters.char_code_at_calls).toBe(char_calls_after_first_events);
+  });
+
+  it('never throws and stays equivalent to stateless APIs on syntax-heavy fuzz input', () => {
+    fc.assert(
+      fc.property(
+        wikiish_string(),
+        spacing_heavy_wikitext_string(),
+        pathological_wikitext_string(),
+        (a: string, b: string, c: string) => {
+          const input = `${a}\n${b}\n${c}`;
+          const session = createSession(input);
+
+          expect(Array.from(session.outline())).toEqual(Array.from(outlineEvents(input)));
+          expect(Array.from(session.events())).toEqual(Array.from(events(input)));
+          expect(session.parse()).toEqual(parse(input));
+        },
+      ),
+    );
+  });
+
+  it('keeps session.parse() usable on mixed Unicode and recovery-heavy inputs', () => {
+    fc.assert(
+      fc.property(
+        odd_character_wikitext_string(),
+        pathological_wikitext_string(),
+        (odd: string, hostile: string) => {
+          const input = `${odd}\n${hostile}`;
+          const session = createSession(input);
+
+          expect(session.parse().type).toBe('root');
+        },
+      ),
+    );
+  });
+
+  it('parseWithDiagnostics() matches the stateless diagnostic API', () => {
+    for (const input of SESSION_FIXTURES) {
+      const session = createSession(input);
+      expect(session.parseWithDiagnostics()).toEqual(parseWithDiagnostics(input));
+    }
+  });
+
+  it('reuses the cached diagnostics result on repeated parseWithDiagnostics() calls', () => {
+    const input = '{|\n| Cell';
+    const source = createCountedSource(input);
+    const session = createSession(source);
+
+    const first = session.parseWithDiagnostics();
+    const first_char_calls = source.counters.char_code_at_calls;
+    const second = session.parseWithDiagnostics();
+
+    expect(second).toBe(first);
+    expect(second).toEqual(parseWithDiagnostics(input));
+    expect(source.counters.char_code_at_calls).toBe(first_char_calls);
+  });
+
+  it('parseWithDiagnostics() does not expose the recovery summary field', () => {
+    const session = createSession(SESSION_FIXTURES[0]);
+
+    expect(Object.hasOwn(session.parseWithDiagnostics(), 'recovered')).toBe(false);
+  });
+
+  it('parseWithRecovery() matches the stateless recovery-aware API', () => {
+    for (const input of SESSION_FIXTURES) {
+      const session = createSession(input);
+      expect(session.parseWithRecovery()).toEqual(parseWithRecovery(input));
+    }
+  });
+
+  it('reuses the cached recovery result on repeated parseWithRecovery() calls', () => {
+    const input = '{|\n| Cell';
+    const source = createCountedSource(input);
+    const session = createSession(source);
+
+    const first = session.parseWithRecovery();
+    const first_char_calls = source.counters.char_code_at_calls;
+    const second = session.parseWithRecovery();
+
+    expect(second).toBe(first);
+    expect(second).toEqual(parseWithRecovery(input));
+    expect(source.counters.char_code_at_calls).toBe(first_char_calls);
+  });
+
+  it('parseWithDiagnostics() preserves real parser diagnostics from recovery input', () => {
+    const input = '{|\n| Cell';
+    const session = createSession(input);
+    const result = session.parseWithDiagnostics();
+
+    expect(result.tree.type).toBe('root');
+    expect(result.diagnostics.length).toBeGreaterThan(0);
+    expect(result.diagnostics[0].code).toBe(DiagnosticCode.UNCLOSED_TABLE);
+    expect(result.diagnostics[0].source).toBe('block');
+    expect(result.diagnostics[0].anchor).toEqual({
+      kind: 'tree-path',
+      path: [0],
+      node_type: 'text',
+    });
+    expect(result.diagnostics[0].recoverable).toBe(true);
+    expect(result.diagnostics[0].severity).toBe('warning');
+  });
+
+  it('does not reuse the loose tree identity when diagnostics strip recovered nodes', () => {
+    const session = createSession('{|\n| Cell');
+    const tree = session.parse();
+    const result = session.parseWithDiagnostics();
+
+    expect(result.tree).not.toBe(tree);
+  });
+
+  it('parseWithRecovery() keeps the loose tree while parseWithDiagnostics() stays strict', () => {
+    const session = createSession('{|\n| Cell');
+    const diagnostics_result = session.parseWithDiagnostics();
+    const recovery_result = session.parseWithRecovery();
+
+    expect(recovery_result.diagnostics).toHaveLength(diagnostics_result.diagnostics.length);
+    expect(recovery_result.diagnostics[0]?.code).toBe(diagnostics_result.diagnostics[0]?.code);
+    expect(recovery_result.diagnostics[0]?.message).toBe(diagnostics_result.diagnostics[0]?.message);
+    expect(recovery_result.diagnostics[0]?.source).toBe(diagnostics_result.diagnostics[0]?.source);
+    expect(recovery_result.diagnostics[0]?.severity).toBe(diagnostics_result.diagnostics[0]?.severity);
+    expect(recovery_result.diagnostics[0]?.recoverable).toBe(
+      diagnostics_result.diagnostics[0]?.recoverable,
+    );
+    expect(recovery_result.diagnostics[0]?.position).toEqual(
+      diagnostics_result.diagnostics[0]?.position,
+    );
+    expect(recovery_result.diagnostics[0]?.anchor.node_type).toBe('table');
+    expect(diagnostics_result.diagnostics[0]?.anchor.node_type).toBe('text');
+    expect(recovery_result.recovered).toBe(true);
+    expect(recovery_result.tree.children[0]?.type).toBe('table');
+    expect(diagnostics_result.tree.children[0]?.type).toBe('text');
+  });
+});
\ No newline at end of file