FA-100061 / Preference pair construction / Open access

Preference pairs: replayed response IDs create duplicate pairs · case 01

replayed response IDs create duplicate pairs.

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

ROOT CAUSE

Repeated delivery of one response is treated as a distinct candidate.

THE FAILURE

Repeated delivery of one response is treated as a distinct candidate.

Unsuccessful approach: Deduplicating equal text collapses distinct responses that happen to have identical wording.

Case contract

Candidates are [id,text,score]. Keep first occurrence per ID, then return [highest_id,lowest_id] by score with lexical ID ties, or None if fewer than two unique IDs.

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(candidates):
    return [max(candidates,key=lambda r:r[2])[0],min(candidates,key=lambda r:r[2])[0]] if len(candidates)>1 else None
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
a='a'+str(N); b='b'+str(N)
check('replay same id',solve([[a,'one',N+5],[a,'changed',-N],[b,'two',0]]),[a,b])
check('same text distinct IDs',solve([[a,'same',N],[b,'same',0]]),[a,b])
check('empty',solve([]),None)
check('singleton',solve([[a,'x',N]]),None)
check('unique extrema',solve([[a,'x',N],[b,'y',-N]]),[a,b])
check('first replay retained',solve([[a,'x',N+2],[a,'z',-N],[b,'y',0]]),[a,b])
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
replay same id	`['a1', 'a1']`	`['a1', 'b1']`	Failed
same text distinct IDs	`['a1', 'b1']`	`['a1', 'b1']`	Passed
empty	`None`	`None`	Passed
singleton	`None`	`None`	Passed
unique extrema	`['a1', 'b1']`	`['a1', 'b1']`	Passed
first replay retained	`['a1', 'a1']`	`['a1', 'b1']`	Failed

SHA-256 / 6010e40b5757835a36e0387ddeb9a46013590c3d9aa36cc18f0447daadf62d35

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(candidates):
    unique={r[1]:r for r in candidates}
    u=list(unique.values())
    return [max(u,key=lambda r:r[2])[0],min(u,key=lambda r:r[2])[0]] if len(u)>1 else None
def check(label, actual, expected):
    observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
a='a'+str(N); b='b'+str(N)
check('replay same id',solve([[a,'one',N+5],[a,'changed',-N],[b,'two',0]]),[a,b])
check('same text distinct IDs',solve([[a,'same',N],[b,'same',0]]),[a,b])
check('empty',solve([]),None)
check('singleton',solve([[a,'x',N]]),None)
check('unique extrema',solve([[a,'x',N],[b,'y',-N]]),[a,b])
check('first replay retained',solve([[a,'x',N+2],[a,'z',-N],[b,'y',0]]),[a,b])
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
replay same id	`['a1', 'a1']`	`['a1', 'b1']`	Failed
same text distinct IDs	`None`	`['a1', 'b1']`	Failed
empty	`None`	`None`	Passed
singleton	`None`	`None`	Passed
unique extrema	`['a1', 'b1']`	`['a1', 'b1']`	Passed
first replay retained	`['a1', 'a1']`	`['a1', 'b1']`	Failed

SHA-256 / e7ff824b70da5d1fbd3b21ab62ae730e312f5890ff10c7df44430d357017148e

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 2026-09-30T16:48:31.263088+00:00.

Case digest / 33998944eab64fc9d7da8096e40f1151cc558b6ea6effc2fbe48c479dec9150e