FA-100111 / Preference pair construction / Open access

Preference pairs: stale annotation revisions remain active · case 01

stale annotation revisions remain active.

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

ROOT CAUSE

All annotation revisions are pooled, so an old preference survives a correction.

THE FAILURE

All annotation revisions are pooled, so an old preference survives a correction.

Unsuccessful approach: Selecting the latest winning revision ignores a later tie or reversal.

Case contract

Rows are [pair_id,revision,chosen,rejected,valid], with unique revisions per pair. Latest revision wins regardless of validity; emit chosen/rejected only when that latest row is valid, otherwise None. Pair order is first-seen.

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):
    out=[]; seen=set()
    for p,v,c,r,ok in rows:
     if p not in seen: seen.add(p); out.append([p,c,r] if ok else [p,None,None])
    return out
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('later reversal',solve([[p,1,'a','b',True],[p,2,'b','a',True]]),[[p,'b','a']])
check('later invalid',solve([[p,1,'a','b',True],[p,2,'x','y',False]]),[[p,None,None]])
check('older invalid later valid',solve([[p,1,'x','y',False],[p,2,'a','b',True]]),[[p,'a','b']])
check('empty',solve([]),[])
check('independent pair',solve([[q,2,'c','d',True],[p,1,'a','b',True]]),[[q,'c','d'],[p,'a','b']])
check('revision order arbitrary',solve([[p,3,'c','d',True],[p,1,'a','b',True]]),[[p,'c','d']])
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
later reversal	`[['p1', 'a', 'b']]`	`[['p1', 'b', 'a']]`	Failed
later invalid	`[['p1', 'a', 'b']]`	`[['p1', None, None]]`	Failed
older invalid later valid	`[['p1', None, None]]`	`[['p1', 'a', 'b']]`	Failed
empty	`[]`	`[]`	Passed
independent pair	`[['q1', 'c', 'd'], ['p1', 'a', 'b']]`	`[['q1', 'c', 'd'], ['p1', 'a', 'b']]`	Passed
revision order arbitrary	`[['p1', 'c', 'd']]`	`[['p1', 'c', 'd']]`	Passed

SHA-256 / 4a1958282ef89d60e3c7a91a2f0a20afd803c07dab545f7b692b81296657bf5c

2 / The unsuccessful fix

Exit 1

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

N = 1
observations = []
def solve(rows):
    latest={}
    for p,v,c,r,ok in rows:
     if ok and (p not in latest or v>latest[p][0]): latest[p]=(v,c,r)
    return [[p,*latest[p][1:]] for p in latest]
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('later reversal',solve([[p,1,'a','b',True],[p,2,'b','a',True]]),[[p,'b','a']])
check('later invalid',solve([[p,1,'a','b',True],[p,2,'x','y',False]]),[[p,None,None]])
check('older invalid later valid',solve([[p,1,'x','y',False],[p,2,'a','b',True]]),[[p,'a','b']])
check('empty',solve([]),[])
check('independent pair',solve([[q,2,'c','d',True],[p,1,'a','b',True]]),[[q,'c','d'],[p,'a','b']])
check('revision order arbitrary',solve([[p,3,'c','d',True],[p,1,'a','b',True]]),[[p,'c','d']])
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
later reversal	`[['p1', 'b', 'a']]`	`[['p1', 'b', 'a']]`	Passed
later invalid	`[['p1', 'a', 'b']]`	`[['p1', None, None]]`	Failed
older invalid later valid	`[['p1', 'a', 'b']]`	`[['p1', 'a', 'b']]`	Passed
empty	`[]`	`[]`	Passed
independent pair	`[['q1', 'c', 'd'], ['p1', 'a', 'b']]`	`[['q1', 'c', 'd'], ['p1', 'a', 'b']]`	Passed
revision order arbitrary	`[['p1', 'c', 'd']]`	`[['p1', 'c', 'd']]`	Passed

SHA-256 / a94f59352e52702f8b5ebef43206307f646907f1933c655ceec776bd067d1f24

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 2026-09-30T16:48:31.716001+00:00.

Case digest / 34355a1f9d720cf81a1ab384ab9eeb35c01451d837d58ac490e97e39ae36863f