Observed (found by the dn-mathematical-foundations audit, verified against HEAD 68d8d39): core/graph/conductance.py's module docstring states the χ_s design law as "any ≼-chain confined to a stratum within a window has ≤ N_s(W)" (conductance.py:57-58), but the implementation's numerator is not confined: _longest_chain (conductance.py:357-372) calls stratum_spine.proper_time(a, b) over pairs of window ids, and proper_time (core/temporal/spine.py:829) counts the max chain through the causal interval in the FULL stratum. chi_s (conductance.py:375-390) divides by n_s = len(window ids).
Concrete counterexample (the suite's own fixture): a 3-event total-order stratum (_one_doc_spine(versions=3)) with window=[first_id, last_id] gives longest = 3, N_s = 2, χ_s = 1.5 ∉ (0,1].
Why it is latent, not live: the only wired call is chi_s_all (conductance.py:393), which passes window=None — and for the whole-stratum window the law holds. The named falsifier (test_conductance.py:314) never exercises window, so it cannot catch this.
Where it is recorded: dn-mathematical-foundations (the foundations design note, PR #47) states the (0,1] range law for window=None as the CND-6 row's hypothesis (§2.5), and MF-3 (the cut/conductance/curvature falsifier batch) carries the windowed-law falsifier as a named deliverable.
Resolution shape (builder's call at MF-3): either confine the numerator to chains within the window (matching the docstring's law) or pin the docstring/row to the window=None hypothesis explicitly in code; per the note's non-goal 10, MF-3 files the failing test + this issue's disposition, never a drive-by fix. Closes when the windowed falsifier lands and the law/implementation pair is coherent one way or the other.
Observed (found by the dn-mathematical-foundations audit, verified against HEAD
68d8d39):core/graph/conductance.py's module docstring states the χ_s design law as "any ≼-chain confined to a stratum within a window has ≤ N_s(W)" (conductance.py:57-58), but the implementation's numerator is not confined:_longest_chain(conductance.py:357-372) callsstratum_spine.proper_time(a, b)over pairs of window ids, andproper_time(core/temporal/spine.py:829) counts the max chain through the causal interval in the FULL stratum.chi_s(conductance.py:375-390) divides byn_s = len(window ids).Concrete counterexample (the suite's own fixture): a 3-event total-order stratum (
_one_doc_spine(versions=3)) withwindow=[first_id, last_id]gives longest = 3, N_s = 2, χ_s = 1.5 ∉ (0,1].Why it is latent, not live: the only wired call is
chi_s_all(conductance.py:393), which passeswindow=None— and for the whole-stratum window the law holds. The named falsifier (test_conductance.py:314) never exerciseswindow, so it cannot catch this.Where it is recorded: dn-mathematical-foundations (the foundations design note, PR #47) states the (0,1] range law for window=None as the CND-6 row's hypothesis (§2.5), and MF-3 (the cut/conductance/curvature falsifier batch) carries the windowed-law falsifier as a named deliverable.
Resolution shape (builder's call at MF-3): either confine the numerator to chains within the window (matching the docstring's law) or pin the docstring/row to the window=None hypothesis explicitly in code; per the note's non-goal 10, MF-3 files the failing test + this issue's disposition, never a drive-by fix. Closes when the windowed falsifier lands and the law/implementation pair is coherent one way or the other.