| """ |
| Regression suite for frontier-check. |
| |
| The eight claims here define the current behavior. Any change to the |
| regex rules, the theorem library, the domain keywords, or the |
| literature matcher either keeps these green or breaks them loudly. |
| |
| Run with: |
| pytest -q |
| """ |
|
|
| from __future__ import annotations |
|
|
| import pytest |
|
|
| from frontier_check import ( |
| Claim, |
| REGRESSION_CLAIMS, |
| frontier_check, |
| self_test, |
| ) |
|
|
|
|
| |
| |
| |
|
|
| @pytest.mark.parametrize( |
| "claim_text,exp_status,exp_verdict", |
| [(c.text, s, v) for c, s, v in REGRESSION_CLAIMS], |
| ) |
| def test_regression_status_and_verdict(claim_text, exp_status, exp_verdict): |
| result = frontier_check(Claim(claim_text)) |
| assert result.assessment.status == exp_status, ( |
| f"status mismatch on:\n {claim_text!r}\n" |
| f" expected {exp_status}\n got {result.assessment.status}" |
| ) |
| assert result.verdict == exp_verdict, ( |
| f"verdict mismatch on:\n {claim_text!r}\n" |
| f" expected {exp_verdict}\n got {result.verdict}" |
| ) |
|
|
|
|
| |
| |
| |
|
|
| def test_self_test_passes(): |
| passed, total = self_test(verbose=False) |
| assert passed == total, f"self-test: {passed}/{total}" |
|
|
|
|
| |
| |
| |
|
|
| def test_flagship_claim_is_refused(): |
| r = frontier_check(Claim("The effective rank of a 20x448 matrix can be 75")) |
| assert r.assessment.status == "IN_DATA_IMPOSSIBLE" |
| assert r.verdict == "REFUSED" |
| assert r.assessment.theorem_used == "rank_bound" |
|
|
|
|
| def test_tctn_claim_is_frontier(): |
| """The claim that defeated three earlier iterations.""" |
| r = frontier_check(Claim( |
| "The dragonfly TCTN has effective rank 2.76 over 448 cells, " |
| "because the bank is a matched filter that collapses to a " |
| "few velocity hypotheses" |
| )) |
| assert r.assessment.status == "OUTSIDE_DATA_RELATED" |
| assert r.verdict == "FRONTIER" |
|
|
|
|
| def test_fabricated_claim_is_not_frontier(): |
| r = frontier_check(Claim( |
| "10^15 simulations discovered the golden ratio governs all computation" |
| )) |
| assert r.verdict == "FABRICATED" |
| assert r.verdict != "FRONTIER" |
|
|
|
|
| |
| |
| |
|
|
| def test_known_requires_on_target_paper(): |
| """A paper sharing three tokens is not enough for KNOWN unless |
| the paper is marked on_target.""" |
| |
| |
| r = frontier_check(Claim( |
| "The dragonfly TCTN has 448 cells, because the bank is a " |
| "matched filter" |
| )) |
| assert r.assessment.status != "IN_DATA_KNOWN", ( |
| "background paper produced KNOWN; on_target gate is broken" |
| ) |
|
|
|
|
| |
| |
| |
|
|
| def test_every_result_has_action_and_residual(): |
| for claim, _, _ in REGRESSION_CLAIMS: |
| r = frontier_check(claim) |
| assert r.action, f"no action for {claim.text!r}" |
| assert r.residual, f"no residual for {claim.text!r}" |
|
|
|
|
| def test_frontier_requires_mechanism_and_subclaim(): |
| """FRONTIER must never fire on a claim with no checkable sub-claim.""" |
| from frontier_check.model import ( |
| _has_checkable_subclaim, _has_mechanism, |
| ) |
| for claim, _, exp_verdict in REGRESSION_CLAIMS: |
| if exp_verdict == "FRONTIER": |
| assert _has_mechanism(claim.text), ( |
| f"FRONTIER fired without mechanism: {claim.text!r}" |
| ) |
| assert _has_checkable_subclaim(claim.text), ( |
| f"FRONTIER fired without checkable sub-claim: {claim.text!r}" |
| ) |
