commit e93692ea1a08ac732c15618f7c5e8fa9383478ef
parent 3b66da8d75aef92493c251a3b2e8039c86ad91fb
Author: Jared Tobin <jared@jtobin.io>
Date: Fri, 3 Jul 2026 14:18:28 -0230
api: log_evalue, log_evalue_sup, and p_value accessors
Every test module now exposes its evidence in calibrated form: the
log e-value (log-wealth normalized so a fresh state sits at zero,
uniform across one-sided and hedged modules), its running supremum,
and the anytime-valid p-value derived from that supremum. decide
remains equivalent to comparing p_value against alpha.
Diffstat:
4 files changed, 159 insertions(+), 0 deletions(-)
diff --git a/lib/Numeric/Eproc/Bernoulli.hs b/lib/Numeric/Eproc/Bernoulli.hs
@@ -73,6 +73,9 @@ module Numeric.Eproc.Bernoulli (
-- * Inspection
, log_wealth
, log_wealth_sup
+ , log_evalue
+ , log_evalue_sup
+ , p_value
, samples
) where
@@ -255,6 +258,47 @@ log_wealth_sup :: State -> Double
log_wealth_sup = st_sup_log_w
{-# INLINE log_wealth_sup #-}
+-- | The current log e-value. For this one-sided test the single
+-- wealth process is itself the e-process (a fresh state already
+-- sits at wealth @1@), so this coincides with 'log_wealth'; the
+-- accessor exists so that e-values read uniformly across test
+-- modules regardless of their internal hedging, e.g. when
+-- convex-combining several e-processes. Not monotone; bounded
+-- above by 'log_evalue_sup'.
+--
+-- >>> log_evalue s0
+-- 0.0
+log_evalue :: State -> Double
+log_evalue = st_log_w
+{-# INLINE log_evalue #-}
+
+-- | The supremum-so-far of the log e-value; coincides with
+-- 'log_wealth_sup' for this one-sided test. Monotone
+-- nondecreasing, starting at @0@; 'decide' rejects exactly when
+-- it crosses @log(1 \/ alpha)@.
+--
+-- >>> log_evalue_sup s0
+-- 0.0
+log_evalue_sup :: State -> Double
+log_evalue_sup = st_sup_log_w
+{-# INLINE log_evalue_sup #-}
+
+-- | The anytime-valid p-value: the reciprocal of the largest
+-- e-value attained so far, @min 1 (exp (negate (log_evalue_sup
+-- s)))@.
+--
+-- Monotone nonincreasing in the sample count, and valid under
+-- optional stopping: under @H_0@,
+-- @P(exists t: p_t <= alpha) <= alpha@ for every @alpha@
+-- simultaneously. 'decide' returns 'Reject' exactly when this
+-- value has reached the configured @alpha@ or below.
+--
+-- >>> p_value s0
+-- 1.0
+p_value :: State -> Double
+p_value s = min 1 (exp (negate (log_evalue_sup s)))
+{-# INLINE p_value #-}
+
-- | The number of samples consumed so far.
--
-- >>> samples s0
diff --git a/lib/Numeric/Eproc/Bernoulli/TwoSided.hs b/lib/Numeric/Eproc/Bernoulli/TwoSided.hs
@@ -55,6 +55,9 @@ module Numeric.Eproc.Bernoulli.TwoSided (
-- * Inspection
, log_wealth
, log_wealth_sup
+ , log_evalue
+ , log_evalue_sup
+ , p_value
, samples
) where
@@ -142,6 +145,38 @@ log_wealth_sup :: State -> Double
log_wealth_sup (State s) = Bounded.log_wealth_sup s
{-# INLINE log_wealth_sup #-}
+-- | The current log e-value of the underlying bounded-mean test:
+-- 'log_wealth' minus @log 2@, normalized so a fresh state sits at
+-- @0@. Not monotone; bounded above by 'log_evalue_sup'.
+--
+-- >>> log_evalue s0
+-- 0.0
+log_evalue :: State -> Double
+log_evalue (State s) = Bounded.log_evalue s
+{-# INLINE log_evalue #-}
+
+-- | The supremum-so-far of the log e-value: 'log_wealth_sup' minus
+-- @log 2@. Monotone nondecreasing, starting at @0@; 'decide'
+-- rejects exactly when it crosses @log(1 \/ alpha)@.
+--
+-- >>> log_evalue_sup s0
+-- 0.0
+log_evalue_sup :: State -> Double
+log_evalue_sup (State s) = Bounded.log_evalue_sup s
+{-# INLINE log_evalue_sup #-}
+
+-- | The anytime-valid p-value: the reciprocal of the largest
+-- e-value attained so far. Monotone nonincreasing; under @H_0@,
+-- @P(exists t: p_t <= alpha) <= alpha@ for every @alpha@
+-- simultaneously. 'decide' returns 'Reject' exactly when this
+-- value has reached the configured @alpha@ or below.
+--
+-- >>> p_value s0
+-- 1.0
+p_value :: State -> Double
+p_value (State s) = Bounded.p_value s
+{-# INLINE p_value #-}
+
-- | The number of samples consumed so far.
--
-- >>> samples s0
diff --git a/lib/Numeric/Eproc/Bounded.hs b/lib/Numeric/Eproc/Bounded.hs
@@ -84,6 +84,9 @@ module Numeric.Eproc.Bounded (
-- * Inspection
, log_wealth
, log_wealth_sup
+ , log_evalue
+ , log_evalue_sup
+ , p_value
, samples
) where
@@ -311,6 +314,47 @@ log_wealth_sup :: State -> Double
log_wealth_sup State{..} = st_sup_log_sum
{-# INLINE log_wealth_sup #-}
+-- | The current log e-value of the convex-hedge e-process: the log
+-- of @(K^+_t + K^-_t) \/ 2@, i.e. 'log_wealth' minus @log 2@.
+--
+-- Unlike 'log_wealth', this is normalized so that a fresh state
+-- sits at @0@ (e-value @1@): it is directly comparable across
+-- test modules regardless of their internal hedging, and is the
+-- form to use when convex-combining several e-processes. Not
+-- monotone; bounded above by 'log_evalue_sup'.
+--
+-- >>> log_evalue s0
+-- 0.0
+log_evalue :: State -> Double
+log_evalue s = log_wealth s - log2_dbl
+{-# INLINE log_evalue #-}
+
+-- | The supremum-so-far of the log e-value: 'log_wealth_sup' minus
+-- @log 2@. Monotone nondecreasing, starting at @0@; 'decide'
+-- rejects exactly when it crosses @log(1 \/ alpha)@.
+--
+-- >>> log_evalue_sup s0
+-- 0.0
+log_evalue_sup :: State -> Double
+log_evalue_sup s = log_wealth_sup s - log2_dbl
+{-# INLINE log_evalue_sup #-}
+
+-- | The anytime-valid p-value: the reciprocal of the largest
+-- e-value attained so far, @min 1 (exp (negate (log_evalue_sup
+-- s)))@.
+--
+-- Monotone nonincreasing in the sample count, and valid under
+-- optional stopping: under @H_0@,
+-- @P(exists t: p_t <= alpha) <= alpha@ for every @alpha@
+-- simultaneously. 'decide' returns 'Reject' exactly when this
+-- value has reached the configured @alpha@ or below.
+--
+-- >>> p_value s0
+-- 1.0
+p_value :: State -> Double
+p_value s = min 1 (exp (negate (log_evalue_sup s)))
+{-# INLINE p_value #-}
+
-- | The number of samples consumed so far.
--
-- >>> samples s0
diff --git a/lib/Numeric/Eproc/Paired.hs b/lib/Numeric/Eproc/Paired.hs
@@ -64,6 +64,9 @@ module Numeric.Eproc.Paired (
-- * Inspection
, log_wealth
, log_wealth_sup
+ , log_evalue
+ , log_evalue_sup
+ , p_value
, samples
) where
@@ -165,6 +168,39 @@ log_wealth_sup :: State -> Double
log_wealth_sup (State s) = Bounded.log_wealth_sup s
{-# INLINE log_wealth_sup #-}
+-- | The current log e-value of the underlying bounded-mean test on
+-- the differences: 'log_wealth' minus @log 2@, normalized so a
+-- fresh state sits at @0@. Not monotone; bounded above by
+-- 'log_evalue_sup'.
+--
+-- >>> log_evalue s0
+-- 0.0
+log_evalue :: State -> Double
+log_evalue (State s) = Bounded.log_evalue s
+{-# INLINE log_evalue #-}
+
+-- | The supremum-so-far of the log e-value: 'log_wealth_sup' minus
+-- @log 2@. Monotone nondecreasing, starting at @0@; 'decide'
+-- rejects exactly when it crosses @log(1 \/ alpha)@.
+--
+-- >>> log_evalue_sup s0
+-- 0.0
+log_evalue_sup :: State -> Double
+log_evalue_sup (State s) = Bounded.log_evalue_sup s
+{-# INLINE log_evalue_sup #-}
+
+-- | The anytime-valid p-value: the reciprocal of the largest
+-- e-value attained so far. Monotone nonincreasing; under @H_0@,
+-- @P(exists t: p_t <= alpha) <= alpha@ for every @alpha@
+-- simultaneously. 'decide' returns 'Reject' exactly when this
+-- value has reached the configured @alpha@ or below.
+--
+-- >>> p_value s0
+-- 1.0
+p_value :: State -> Double
+p_value (State s) = Bounded.p_value s
+{-# INLINE p_value #-}
+
-- | The number of paired observations consumed so far.
--
-- >>> samples s0
