FA-100081 / Preference pair construction / Open access

Preference pairs: weights are normalized across unrelated prompts · case 01

weights are normalized across unrelated prompts.

Verified by executionVariant 1 · 6 checks per implementationDownload source bundle ↓JSON ↗

ROOT CAUSE

The denominator is computed over the entire dataset, coupling independent prompt groups.

THE FAILURE

The denominator is computed over the entire dataset, coupling independent prompt groups.

Unsuccessful approach: Normalizing each row to one ignores other eligible pairs for the same prompt.

Case contract

Rows are [prompt,id,weight]. Return [prompt,id,numerator,denominator] for positive weights normalized by prompt sum, preserving input order.

Why this case matters

A small deterministic preference-data transform makes pair semantics inspectable without a training framework.

1 / The failure

Exit 1

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

N = 1
observations = []
def solve(rows):
    total=sum(r[2] for r in rows)
    return [[r[0],r[1],r[2],total] for r in rows if r[2]>0]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
p='p'+str(N); q='q'+str(N)
check('separate denominators',solve([[p,'a',N],[p,'b',N],[q,'c',2*N]]),[[p,'a',1,2],[p,'b',1,2],[q,'c',1,1]])
check('zero excluded',solve([[p,'a',0],[p,'b',N]]),[[p,'b',1,1]])
check('empty',solve([]),[])
check('all zero',solve([[p,'a',0]]),[])
check('singleton',solve([[p,'a',N]]),[[p,'a',1,1]])
check('interleaved groups',solve([[p,'a',1],[q,'b',1],[p,'c',3]]),[[p,'a',1,4],[q,'b',1,1],[p,'c',3,4]])
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
separate denominators	`[['p1', 'a', 1, 4], ['p1', 'b', 1, 4], ['q1', 'c', 2, 4]]`	`[['p1', 'a', 1, 2], ['p1', 'b', 1, 2], ['q1', 'c', 1, 1]]`	Failed
zero excluded	`[['p1', 'b', 1, 1]]`	`[['p1', 'b', 1, 1]]`	Passed
empty	`[]`	`[]`	Passed
all zero	`[]`	`[]`	Passed
singleton	`[['p1', 'a', 1, 1]]`	`[['p1', 'a', 1, 1]]`	Passed
interleaved groups	`[['p1', 'a', 1, 5], ['q1', 'b', 1, 5], ['p1', 'c', 3, 5]]`	`[['p1', 'a', 1, 4], ['q1', 'b', 1, 1], ['p1', 'c', 3, 4]]`	Failed

SHA-256 / 0841dc420f610fd0a7b07cc52f306be8fc30ae902911d74d8db1daeac4d5d2b5

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(rows):
    return [[r[0],r[1],1,1] for r in rows if r[2]>0]
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
p='p'+str(N); q='q'+str(N)
check('separate denominators',solve([[p,'a',N],[p,'b',N],[q,'c',2*N]]),[[p,'a',1,2],[p,'b',1,2],[q,'c',1,1]])
check('zero excluded',solve([[p,'a',0],[p,'b',N]]),[[p,'b',1,1]])
check('empty',solve([]),[])
check('all zero',solve([[p,'a',0]]),[])
check('singleton',solve([[p,'a',N]]),[[p,'a',1,1]])
check('interleaved groups',solve([[p,'a',1],[q,'b',1],[p,'c',3]]),[[p,'a',1,4],[q,'b',1,1],[p,'c',3,4]])
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
separate denominators	`[['p1', 'a', 1, 1], ['p1', 'b', 1, 1], ['q1', 'c', 1, 1]]`	`[['p1', 'a', 1, 2], ['p1', 'b', 1, 2], ['q1', 'c', 1, 1]]`	Failed
zero excluded	`[['p1', 'b', 1, 1]]`	`[['p1', 'b', 1, 1]]`	Passed
empty	`[]`	`[]`	Passed
all zero	`[]`	`[]`	Passed
singleton	`[['p1', 'a', 1, 1]]`	`[['p1', 'a', 1, 1]]`	Passed
interleaved groups	`[['p1', 'a', 1, 1], ['q1', 'b', 1, 1], ['p1', 'c', 1, 1]]`	`[['p1', 'a', 1, 4], ['q1', 'b', 1, 1], ['p1', 'c', 3, 4]]`	Failed

SHA-256 / 30ed27929224f3c8954cfff89a41ba948c1f0badc7101d19936fb25312f1a920

HELD IN THE MEMBER ARCHIVE

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

This mechanism has 6 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

These are explicit toy record contracts; they do not claim compatibility with a trainer or improved model behavior. 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 2026-09-30T16:48:31.717323+00:00.

Case digest / 59b307abf159cdd229e1a3c6bddc0d3c9fafe2990069bf0cc6427f24639a3f61