FA-16566 / Floating-point arithmetic / Open access
Quiet NaN propagation strips its diagnostic sign · case 01
Quiet NaN propagation strips its diagnostic sign.
ROOT CAUSE
Quiet NaN propagation strips its diagnostic sign. The faulty expression is return [chosen&((1<<63)-1),False].
THE FAILURE
Quiet NaN propagation strips its diagnostic sign. The faulty expression is return [chosen&((1<<63)-1),False].
Unsuccessful approach: The attempted local correction return [chosen|(1<<63),False] still violates the explicit regression fixtures.
Case contract
Stipulated binary64 addition policy: propagate first signaling NaN, otherwise largest quiet-NaN diagnostic payload (left wins ties); quiet propagated NaNs preserving sign. Opposite infinities produce canonical quiet NaN. Finite values use Python binary64 addition. Return bits and invalid flag.
Why this case matters
An offline floating representation model isolates a reproducible arithmetic fault.
1 / The failure
Exit 1"""Failure Map reference implementation. Python standard library only."""
import json
import math
import struct
N = 1
observations = []
def solve(ab,bb):
mask=(1<<52)-1
an=(ab>>52)&2047==2047 and bool(ab&mask)
bn=(bb>>52)&2047==2047 and bool(bb&mask)
asig=an and not bool(ab&(1<<51))
bsig=bn and not bool(bb&(1<<51))
if asig or bsig:
chosen=ab if asig else bb
return [chosen|(1<<51),True]
if an or bn:
if an and bn:
chosen=ab if (ab&((1<<51)-1)) >= (bb&((1<<51)-1)) else bb
else:
chosen=ab if an else bb
return [chosen&((1<<63)-1),False]
a=struct.unpack('>d',ab.to_bytes(8,'big'))[0]
b=struct.unpack('>d',bb.to_bytes(8,'big'))[0]
if math.isinf(a) and math.isinf(b) and a!=b:
return [0x7ff8000000000000,True]
result=a+b
return [int.from_bytes(struct.pack('>d',result),'big'),False]
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('left signaling', solve(0x7ff0000000000000|N,0x3ff0000000000000), [0x7ff8000000000000|N,True])
check('right signaling', solve(0x3ff0000000000000,0xfff0000000000000|N), [0xfff8000000000000|N,True])
check('signaling precedence', solve(0x7ff8000000000064,0x7ff0000000000000|N), [0x7ff8000000000000|N,True])
check('two signaling', solve(0xfff0000000000000|N,0x7ff0000000000010), [0xfff8000000000000|N,True])
check('larger right payload', solve(0x7ff8000000000000|N,0xfff8000000000064), [0xfff8000000000064,False])
check('larger left payload', solve(0xfff8000000000064,0x7ff8000000000000|N), [0xfff8000000000064,False])
check('equal left wins', solve(0xfff8000000000000|N,0x7ff8000000000000|N), [0xfff8000000000000|N,False])
check('positive quiet', solve(0x7ff8000000000000|N,0x3ff0000000000000), [0x7ff8000000000000|N,False])
check('quiet left only', solve(0xfff8000000000000|N,0x3ff0000000000000), [0xfff8000000000000|N,False])
check('quiet right only', solve(0x3ff0000000000000,0xfff8000000000000|N), [0xfff8000000000000|N,False])
check('opposite infinities', solve(0x7ff0000000000000,0xfff0000000000000), [0x7ff8000000000000,True])
check('same infinity', solve(0xfff0000000000000,0xfff0000000000000), [0xfff0000000000000,False])
check('finite addition', solve(0x3ff0000000000000,0x3ff0000000000000), [0x4000000000000000,False])
check('negative zeros', solve(1<<63,1<<63), [1<<63,False])
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 |
|---|---|---|---|
| left signaling | [9221120237041090561, True] | [9221120237041090561, True] | Passed |
| right signaling | [18444492273895866369, True] | [18444492273895866369, True] | Passed |
| signaling precedence | [9221120237041090561, True] | [9221120237041090561, True] | Passed |
| two signaling | [18444492273895866369, True] | [18444492273895866369, True] | Passed |
| larger right payload | [9221120237041090660, False] | [18444492273895866468, False] | Failed |
| larger left payload | [9221120237041090660, False] | [18444492273895866468, False] | Failed |
| equal left wins | [9221120237041090561, False] | [18444492273895866369, False] | Failed |
| positive quiet | [9221120237041090561, False] | [9221120237041090561, False] | Passed |
| quiet left only | [9221120237041090561, False] | [18444492273895866369, False] | Failed |
| quiet right only | [9221120237041090561, False] | [18444492273895866369, False] | Failed |
| opposite infinities | [9221120237041090560, True] | [9221120237041090560, True] | Passed |
| same infinity | [18442240474082181120, False] | [18442240474082181120, False] | Passed |
| finite addition | [4611686018427387904, False] | [4611686018427387904, False] | Passed |
| negative zeros | [9223372036854775808, False] | [9223372036854775808, False] | Passed |
SHA-256 / f83622f5f52dc3812ed8ad60db85658536de74ef826da442c29d394af6ca2764
2 / The unsuccessful fix
Exit 1"""Failure Map reference implementation. Python standard library only."""
import json
import math
import struct
N = 1
observations = []
def solve(ab,bb):
mask=(1<<52)-1
an=(ab>>52)&2047==2047 and bool(ab&mask)
bn=(bb>>52)&2047==2047 and bool(bb&mask)
asig=an and not bool(ab&(1<<51))
bsig=bn and not bool(bb&(1<<51))
if asig or bsig:
chosen=ab if asig else bb
return [chosen|(1<<51),True]
if an or bn:
if an and bn:
chosen=ab if (ab&((1<<51)-1)) >= (bb&((1<<51)-1)) else bb
else:
chosen=ab if an else bb
return [chosen|(1<<63),False]
a=struct.unpack('>d',ab.to_bytes(8,'big'))[0]
b=struct.unpack('>d',bb.to_bytes(8,'big'))[0]
if math.isinf(a) and math.isinf(b) and a!=b:
return [0x7ff8000000000000,True]
result=a+b
return [int.from_bytes(struct.pack('>d',result),'big'),False]
def check(label, actual, expected):
observations.append({"check": label, "actual": actual, "expected": expected, "passed": actual == expected})
check('left signaling', solve(0x7ff0000000000000|N,0x3ff0000000000000), [0x7ff8000000000000|N,True])
check('right signaling', solve(0x3ff0000000000000,0xfff0000000000000|N), [0xfff8000000000000|N,True])
check('signaling precedence', solve(0x7ff8000000000064,0x7ff0000000000000|N), [0x7ff8000000000000|N,True])
check('two signaling', solve(0xfff0000000000000|N,0x7ff0000000000010), [0xfff8000000000000|N,True])
check('larger right payload', solve(0x7ff8000000000000|N,0xfff8000000000064), [0xfff8000000000064,False])
check('larger left payload', solve(0xfff8000000000064,0x7ff8000000000000|N), [0xfff8000000000064,False])
check('equal left wins', solve(0xfff8000000000000|N,0x7ff8000000000000|N), [0xfff8000000000000|N,False])
check('positive quiet', solve(0x7ff8000000000000|N,0x3ff0000000000000), [0x7ff8000000000000|N,False])
check('quiet left only', solve(0xfff8000000000000|N,0x3ff0000000000000), [0xfff8000000000000|N,False])
check('quiet right only', solve(0x3ff0000000000000,0xfff8000000000000|N), [0xfff8000000000000|N,False])
check('opposite infinities', solve(0x7ff0000000000000,0xfff0000000000000), [0x7ff8000000000000,True])
check('same infinity', solve(0xfff0000000000000,0xfff0000000000000), [0xfff0000000000000,False])
check('finite addition', solve(0x3ff0000000000000,0x3ff0000000000000), [0x4000000000000000,False])
check('negative zeros', solve(1<<63,1<<63), [1<<63,False])
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 |
|---|---|---|---|
| left signaling | [9221120237041090561, True] | [9221120237041090561, True] | Passed |
| right signaling | [18444492273895866369, True] | [18444492273895866369, True] | Passed |
| signaling precedence | [9221120237041090561, True] | [9221120237041090561, True] | Passed |
| two signaling | [18444492273895866369, True] | [18444492273895866369, True] | Passed |
| larger right payload | [18444492273895866468, False] | [18444492273895866468, False] | Passed |
| larger left payload | [18444492273895866468, False] | [18444492273895866468, False] | Passed |
| equal left wins | [18444492273895866369, False] | [18444492273895866369, False] | Passed |
| positive quiet | [18444492273895866369, False] | [9221120237041090561, False] | Failed |
| quiet left only | [18444492273895866369, False] | [18444492273895866369, False] | Passed |
| quiet right only | [18444492273895866369, False] | [18444492273895866369, False] | Passed |
| opposite infinities | [9221120237041090560, True] | [9221120237041090560, True] | Passed |
| same infinity | [18442240474082181120, False] | [18442240474082181120, False] | Passed |
| finite addition | [4611686018427387904, False] | [4611686018427387904, False] | Passed |
| negative zeros | [9223372036854775808, False] | [9223372036854775808, False] | Passed |
SHA-256 / a4fdaed2cef6b1bc475a36e0f539813b7b3a74d5b279a6ebd3bcf95e4c2d7b5d
HELD IN THE MEMBER ARCHIVE
The verified repair and its recorded checks are member-only.
This mechanism has 14 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.
Sign in to the archive ↗Verification & scope
Controlled binary64 or explicitly stipulated miniature format; no hardware exception flags or platform floating environment are modeled. 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 2026-09-29T14:39:37.678459+00:00.
Case digest / 901ff164cc94dcaa572635e3a3a3bd0a03c47f36798e87af6a9383a35d29ea9b