FA-100421 / Feature preprocessing fit scope / Open access

Quantile bin assignment uses held-out rows while fitting · case 01

Quantile bin assignment uses held-out rows while fitting.

Verified by executionVariant 1 · 8 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

Fitting reads values outside the designated training rows.

THE FAILURE

Fitting reads values outside the designated training rows.

Unsuccessful approach: Filtering rows after computing the statistic still leaves held-out influence.

Case contract

Train is nonempty, bins is positive. Select empirical cut values at floor(i*n/bins), capped to the final training index. Count strict greater-than cut comparisons, so equality stays in the lower bin.

Why this case matters

A small deterministic preprocessing contract makes the fit boundary and transform behavior directly inspectable.

1 / The failure

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(train, values, bins):
    v=sorted(train+values); edges=[v[(i*len(v))//bins] for i in range(1,bins)]; return [sum(x>e for e in edges) for x in values]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('three equal-width ranks',solve([0,1,2,3,4,5],[0,2,5],3),[0,0,2])
check('nondivisible rank boundary',solve([0,1,2,3,4],[3],2),[1])
check('heldout cannot set cuts',solve([0,10,20],[8,12,18],3),[0,1,1])
check('boundary equality stays lower',solve([0,2,4,6],[2,3,4],2),[0,0,0])
check('repeated training values',solve([1,1,1,9],[1,9],2),[0,1])
check('one bin',solve([0,2,4],[N,100],1),[0,0])
check('empty values',solve([0,2],[],2),[])
check('upper extreme',solve([0,10],[10],2),[0])
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
three equal-width ranks	`[0, 0, 2]`	`[0, 0, 2]`	Passed
nondivisible rank boundary	`[0]`	`[1]`	Failed
heldout cannot set cuts	`[0, 1, 1]`	`[0, 1, 1]`	Passed
boundary equality stays lower	`[0, 0, 1]`	`[0, 0, 0]`	Failed
repeated training values	`[0, 1]`	`[0, 1]`	Passed
one bin	`[0, 0]`	`[0, 0]`	Passed
empty values	`[]`	`[]`	Passed
upper extreme	`[0]`	`[0]`	Passed

SHA-256 / 0a09914c76a9d1a71ca5b16c488d1d59d3e8980eb16305a925ab7e58f0bc9c23

2 / The unsuccessful fix

Exit 1

"""Failure Map reference implementation. Python standard library only."""
import json

N = 1
observations = []
def solve(train, values, bins):
    v=sorted(train); edges=[v[min(len(v)-1,((i*len(v)+bins-1)//bins))] for i in range(1,bins)]; return [sum(x>e for e in edges) for x in values]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('three equal-width ranks',solve([0,1,2,3,4,5],[0,2,5],3),[0,0,2])
check('nondivisible rank boundary',solve([0,1,2,3,4],[3],2),[1])
check('heldout cannot set cuts',solve([0,10,20],[8,12,18],3),[0,1,1])
check('boundary equality stays lower',solve([0,2,4,6],[2,3,4],2),[0,0,0])
check('repeated training values',solve([1,1,1,9],[1,9],2),[0,1])
check('one bin',solve([0,2,4],[N,100],1),[0,0])
check('empty values',solve([0,2],[],2),[])
check('upper extreme',solve([0,10],[10],2),[0])
print(json.dumps({"observations": observations, "passed": all(x["passed"] for x in observations)}, ensure_ascii=False))
raise SystemExit(0 if all(x["passed"] for x in observations) else 1)

Boundary fixture	Actual	Expected	Outcome
three equal-width ranks	`[0, 0, 2]`	`[0, 0, 2]`	Passed
nondivisible rank boundary	`[0]`	`[1]`	Failed
heldout cannot set cuts	`[0, 1, 1]`	`[0, 1, 1]`	Passed
boundary equality stays lower	`[0, 0, 0]`	`[0, 0, 0]`	Passed
repeated training values	`[0, 1]`	`[0, 1]`	Passed
one bin	`[0, 0]`	`[0, 0]`	Passed
empty values	`[]`	`[]`	Passed
upper extreme	`[0]`	`[0]`	Passed

SHA-256 / ec3fe8b5402f086c7e2e736a2dc22cd2783698dcd846f6bad081e9a2ae46ce1c

HELD IN THE MEMBER ARCHIVE

The verified repair and its recorded checks are member-only.

This mechanism has 8 recorded checks per implementation. The open-access tier publishes the failure and the unsuccessful fix; the repaired source that passes every check, and the observations that prove it, are available to members.

Every case sharing this mechanism uses the same contract and the same repair, so this one record is held back for all of them.

Member access is invitation-based. Sign in with your invited account to inspect the repair.

Verification & scope

Toy empirical rank binning with duplicate cuts permitted and equality assigned to the lower bin. This reproducer isolates one failure mechanism. Results cover the supplied fixtures. Variants within a family share a test contract and should remain grouped when constructing evaluation splits. Related mechanisms with a shared evaluation_group must also remain together; these controlled models are not independent production incidents.

Observations recorded using Python 3.12.14 at 2026-09-30T16:48:35.362781+00:00.

Case digest / 59c3b5ce29a0878e5dd40cbfe3b54b24135327f015aca553da07d0d409140a91