claude-bot
69c5b788b9
The database was the missing half: extraction and assembly existed, but nothing turned the book's main-content statements into something an agent could actually cite mid-proof, and nothing searched it. Two new scripts. textbook/build_database.py takes the main-content statements and emits both halves of a usable premise library: - an importable Lean file where every entry is a named declaration, so a proof can cite goodman_1_6_20 by name; - a JSONL search index carrying each entry's name, statement, mentioned identifiers and status. An entry is emitted as an `axiom` (asserted from the textbook) until a real lean-bench run proves it; passing --results upgrades those entries to real theorems carrying the exact verified proof. This mirrors how the book itself works -- its exercises are posed assuming its body results, the same way using Mathlib does not mean re-proving Mathlib. The cost is explicit: anything proven through an axiom inherits it, so a garbled extraction asserted as an axiom makes downstream proofs worthless. Guards: --require-checked refuses entries that have not passed check_statements.py (all 21 section 1.6 entries have, and are now marked accordingly), and the verifier's existing #print axioms check surfaces what a finished proof rests on. textbook/search_database.py is the retrieval step the proof loop calls at each line: given the current goal, it ranks premises by identifier and symbol overlap, weighting rare identifiers far above common ones (IsGCDOfList matching means much more than Int matching), preferring a proven premise over an asserted one, and formats the top few for the prompt. Verified against three real goals: the exactly-applicable premise tops each ranking by a wide margin (11.4 vs 1.6 for the gcd goal, 5.1 vs 3.7 for the prime-divides-product goal). It is a ranking heuristic, not a decision procedure -- only Lean can say whether a premise applies. Replacing score_premise() is the whole job of swapping in embeddings. Also: renamed library_candidates/ to database/ to match what it is; generated lib/Goodman/Chapter01Section06.lean (21 premises, 0 proven, 21 asserted) and database/goodman-1.6.index.jsonl; noted a latent bug in the older build_library.py, which attaches each entry's non-import prelude per-theorem and so would emit a shared `def` twice in one file -- build_database.py hoists and dedupes, and has a test pinning that. 13 new tests, all green (33 total, the one failure remains the pre-existing unrelated lean_agent import error in test_dataset.py). Still not done, and worth being clear about: nothing has been proven yet. The bench targets have not been elaboration-checked, no lean-bench run has happened over either file, and the goal-directed loop that would consume this retrieval does not exist -- lean-bench's solve() is still single-shot. Separately, lean-bench cannot currently run at all in this checkout, because lean_bench/cli.py imports lean_agent at module load and lean-agent is not installed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017vbLFFJiHSZqGZHEkjvG5x
59 lines
2.9 KiB
Python
59 lines
2.9 KiB
Python
import unittest
|
|
|
|
from textbook.search_database import format_for_prompt, search, tokenize
|
|
|
|
INDEX = [
|
|
{"name": "goodman_1_6_20", "id": "goodman-1-6-20", "source_label": "Corollary 1.6.20",
|
|
"kind": "corollary", "status": "asserted",
|
|
"statement": "\u2200 (p : \u2115) (as : List \u2124), Nat.Prime p \u2192 (p:\u2124) \u2223 List.prod as \u2192 \u2203 a \u2208 as, (p:\u2124) \u2223 a",
|
|
"natural_language": "", "identifiers": ["Nat.Prime", "List.prod"], "requires": ["Nat.Prime"]},
|
|
{"name": "goodman_1_6_24", "id": "goodman-1-6-24", "source_label": "Proposition 1.6.24",
|
|
"kind": "proposition", "status": "asserted",
|
|
"statement": "\u2200 (as : List \u2124), \u2203 d, IsGCDOfList d as",
|
|
"natural_language": "", "identifiers": ["IsGCDOfList"], "requires": []},
|
|
{"name": "goodman_1_6_1", "id": "goodman-1-6-1", "source_label": "Proposition 1.6.1",
|
|
"kind": "proposition", "status": "asserted",
|
|
"statement": "\u2200 (a : \u2124), a + 0 = a", "natural_language": "",
|
|
"identifiers": [], "requires": []},
|
|
]
|
|
|
|
|
|
class SearchDatabaseTests(unittest.TestCase):
|
|
def test_tokenize_keeps_qualified_names_and_drops_stopwords(self):
|
|
tokens = tokenize('let x := Nat.Prime p')
|
|
self.assertIn('Nat.Prime', tokens)
|
|
self.assertNotIn('let', tokens)
|
|
|
|
def test_exact_premise_ranks_first(self):
|
|
hits = search(INDEX, '\u22a2 IsGCDOfList d as', limit=3)
|
|
self.assertEqual(hits[0]['name'], 'goodman_1_6_24')
|
|
|
|
def test_rare_identifier_outweighs_common_one(self):
|
|
# Both premises mention List; only one mentions Nat.Prime, so the goal about
|
|
# primes must prefer it rather than tying on the common token.
|
|
hits = search(INDEX, '\u22a2 Nat.Prime p \u2192 (p:\u2124) \u2223 List.prod as', limit=3)
|
|
self.assertEqual(hits[0]['name'], 'goodman_1_6_20')
|
|
|
|
def test_irrelevant_premises_are_dropped_not_ranked_last(self):
|
|
hits = search(INDEX, '\u22a2 IsGCDOfList d as', limit=5)
|
|
self.assertNotIn('goodman_1_6_1', [h['name'] for h in hits])
|
|
|
|
def test_proven_premise_is_preferred_over_an_asserted_one(self):
|
|
index = [dict(INDEX[1]), dict(INDEX[1], name='proved_twin', id='proved-twin',
|
|
status='proved', source_label='Proposition 1.6.24b')]
|
|
hits = search(index, '\u22a2 IsGCDOfList d as', limit=2)
|
|
self.assertEqual(hits[0]['name'], 'proved_twin')
|
|
|
|
def test_limit_is_respected_and_empty_result_is_explained(self):
|
|
self.assertEqual(len(search(INDEX, '\u22a2 List.prod as', limit=1)), 1)
|
|
self.assertIn('no database premise', format_for_prompt([]))
|
|
|
|
def test_format_for_prompt_marks_axioms_distinctly(self):
|
|
text = format_for_prompt(search(INDEX, '\u22a2 IsGCDOfList d as', limit=1))
|
|
self.assertIn('asserted from the book (axiom)', text)
|
|
self.assertIn('goodman_1_6_24', text)
|
|
|
|
|
|
if __name__ == '__main__':
|
|
unittest.main()
|