Case 830
The second incident was deduplicated into the first
the_second_incident_was_deduplicated_into_the_first.eml - Every distinct alert fingerprint paged on-call this shift, and the dedup that suppressed the rest removed only exact repeats. What a fingerprint groups together is computed below.
ok: true — round-trip fixpoint reached (python1 == python2)updated 2026-09-13
EML
eml# Self-authored for the EML case corpus (no external origin). Every distinct
# alert fingerprint paged on-call this shift, and the dedup that suppressed the
# rest removed only exact repeats. What a fingerprint groups together is
# computed below.
#
# The deduplication is sensible. It groups alerts by a fingerprint so a flapping
# check does not page a hundred times; it holds the group open only while the
# condition persists; it pages once per new fingerprint; and it is why on-call
# is not drowned in noise.
#
# The fingerprint is the service plus the error class, and two different
# incidents shared one.
4 => dedup_window_hours
2 => distinct_incidents_in_the_window
1 => alerts_paged
1 => incidents_suppressed_as_duplicates
140 => minutes_the_second_incident_went_unpaged
distinct_incidents_in_the_window - alerts_paged => incidents_that_reached_nobody
int(incidents_that_reached_nobody * 10000 / distinct_incidents_in_the_window) => incidents_lost_per_myriad
"dedup window : " + str(dedup_window_hours) + " hours" ^0
"distinct incidents : " + str(distinct_incidents_in_the_window) ^0
" alerts paged : " + str(alerts_paged) ^0
" suppressed as duplicates : " + str(incidents_suppressed_as_duplicates) ^0
" reached nobody : " + str(incidents_that_reached_nobody) ^0
"second incident went unpaged for: " + str(minutes_the_second_incident_went_unpaged) + " minutes" ^0
"incidents lost : " + str(incidents_lost_per_myriad) + " per ten thousand" ^0
"" ^0
# ---- what the dedup verified ----
"the alert dedup" ^0
" groups by : a fingerprint" ^0
" holds a group open : while the condition persists" ^0
" pages : once per new fingerprint" ^0
" purpose : a flapping check does not page a hundred times" ^0
" exact repeats suppressed : as designed" ^0
" verdict : NOISE CONTROLLED" ^0
"" ^0
" holding one page for a persisting condition is the part" ^0
" done right here, and it is why on-call is not drowned" ^0
"" ^0
# ---- what the fingerprint groups ----
"the fingerprint" ^0
" what it is : the service plus the error class" ^0
" what it assumes : the same fingerprint is the same" ^0
" incident, recurring" ^0
" what happened : two incidents, different root causes," ^0
" same service and error class" ^0
" so the second : matched the first's open group" ^0
" and was treated as : a repeat of an incident already" ^0
" being handled" ^0
"" ^0
# ---- what on-call never saw ----
"the second incident" ^0
" its own root cause : different from the first" ^0
" the page it generated : suppressed as a duplicate" ^0
" minutes it ran unattended : " + str(minutes_the_second_incident_went_unpaged) ^0
" is the dedup rule wrong : no; the fingerprints did" ^0
" match" ^0
" is a matching fingerprint the same incident : not" ^0
" always, and here it was not" ^0
"" ^0
# ---- null control ----
# The same two incidents, with the dedup keyed on the fingerprint plus a
# correlation id that differs when the underlying cause differs.
1 => nc_alerts_when_keyed_on_fingerprint_alone
2 => nc_alerts_when_keyed_with_correlation
0 => nc_incidents_lost
"null control - fingerprint plus a correlation id" ^0
" alerts, fingerprint alone : " ^0
" " + str(nc_alerts_when_keyed_on_fingerprint_alone) ^0
" alerts, with a correlation id : " ^0
" " + str(nc_alerts_when_keyed_with_correlation) ^0
" incidents lost : " + str(nc_incidents_lost) ^0
" no alert changed its content; the key stopped collapsing" ^0
" two causes that share a surface" ^0
"" ^0
# ---- the rule ----
"what per-fingerprint dedup guarantees" ^0
" every distinct fingerprint pages once : exactly, held" ^0
" open while the condition persists" ^0
" every incident pages someone : not addressed; alerts" ^0
" dedup by fingerprint, and two distinct incidents shared" ^0
" one - the second was suppressed as a duplicate and went" ^0
" " + str(minutes_the_second_incident_went_unpaged) + " minutes unpaged" ^0
"" ^0
"a fingerprint is a guess that sameness of surface is sameness of cause, and" ^0
"dedup acts on the guess; when two causes wear one face, silencing the repeat" ^0
"silences the second incident" ^0
"" ^0
"It pages once per fingerprint and holds the group while the condition persists" ^0
"- noise controlled. The fingerprint is service plus error class, and two" ^0
"distinct incidents shared it, so the second was suppressed as a duplicate: " ^0
"" + str(minutes_the_second_incident_went_unpaged) + " minutes unpaged, " + str(incidents_lost_per_myriad) + " per ten thousand of incidents reaching nobody." ^0Python (deterministic transpilation)
pythondedup_window_hours = 4
distinct_incidents_in_the_window = 2
alerts_paged = 1
incidents_suppressed_as_duplicates = 1
minutes_the_second_incident_went_unpaged = 140
incidents_that_reached_nobody = distinct_incidents_in_the_window - alerts_paged
incidents_lost_per_myriad = int(incidents_that_reached_nobody * 10000 / distinct_incidents_in_the_window)
print("dedup window : " + str(dedup_window_hours) + " hours")
print("distinct incidents : " + str(distinct_incidents_in_the_window))
print(" alerts paged : " + str(alerts_paged))
print(" suppressed as duplicates : " + str(incidents_suppressed_as_duplicates))
print(" reached nobody : " + str(incidents_that_reached_nobody))
print("second incident went unpaged for: " + str(minutes_the_second_incident_went_unpaged) + " minutes")
print("incidents lost : " + str(incidents_lost_per_myriad) + " per ten thousand")
print("")
print("the alert dedup")
print(" groups by : a fingerprint")
print(" holds a group open : while the condition persists")
print(" pages : once per new fingerprint")
print(" purpose : a flapping check does not page a hundred times")
print(" exact repeats suppressed : as designed")
print(" verdict : NOISE CONTROLLED")
print("")
print(" holding one page for a persisting condition is the part")
print(" done right here, and it is why on-call is not drowned")
print("")
print("the fingerprint")
print(" what it is : the service plus the error class")
print(" what it assumes : the same fingerprint is the same")
print(" incident, recurring")
print(" what happened : two incidents, different root causes,")
print(" same service and error class")
print(" so the second : matched the first's open group")
print(" and was treated as : a repeat of an incident already")
print(" being handled")
print("")
print("the second incident")
print(" its own root cause : different from the first")
print(" the page it generated : suppressed as a duplicate")
print(" minutes it ran unattended : " + str(minutes_the_second_incident_went_unpaged))
print(" is the dedup rule wrong : no; the fingerprints did")
print(" match")
print(" is a matching fingerprint the same incident : not")
print(" always, and here it was not")
print("")
nc_alerts_when_keyed_on_fingerprint_alone = 1
nc_alerts_when_keyed_with_correlation = 2
nc_incidents_lost = 0
print("null control - fingerprint plus a correlation id")
print(" alerts, fingerprint alone : ")
print(" " + str(nc_alerts_when_keyed_on_fingerprint_alone))
print(" alerts, with a correlation id : ")
print(" " + str(nc_alerts_when_keyed_with_correlation))
print(" incidents lost : " + str(nc_incidents_lost))
print(" no alert changed its content; the key stopped collapsing")
print(" two causes that share a surface")
print("")
print("what per-fingerprint dedup guarantees")
print(" every distinct fingerprint pages once : exactly, held")
print(" open while the condition persists")
print(" every incident pages someone : not addressed; alerts")
print(" dedup by fingerprint, and two distinct incidents shared")
print(" one - the second was suppressed as a duplicate and went")
print(" " + str(minutes_the_second_incident_went_unpaged) + " minutes unpaged")
print("")
print("a fingerprint is a guess that sameness of surface is sameness of cause, and")
print("dedup acts on the guess; when two causes wear one face, silencing the repeat")
print("silences the second incident")
print("")
print("It pages once per fingerprint and holds the group while the condition persists")
print("- noise controlled. The fingerprint is service plus error class, and two")
print("distinct incidents shared it, so the second was suppressed as a duplicate: ")
print("" + str(minutes_the_second_incident_went_unpaged) + " minutes unpaged, " + str(incidents_lost_per_myriad) + " per ten thousand of incidents reaching nobody.")stdout (executed)
textdedup window : 4 hours
distinct incidents : 2
alerts paged : 1
suppressed as duplicates : 1
reached nobody : 1
second incident went unpaged for: 140 minutes
incidents lost : 5000 per ten thousand
the alert dedup
groups by : a fingerprint
holds a group open : while the condition persists
pages : once per new fingerprint
purpose : a flapping check does not page a hundred times
exact repeats suppressed : as designed
verdict : NOISE CONTROLLED
holding one page for a persisting condition is the part
done right here, and it is why on-call is not drowned
the fingerprint
what it is : the service plus the error class
what it assumes : the same fingerprint is the same
incident, recurring
what happened : two incidents, different root causes,
same service and error class
so the second : matched the first's open group
and was treated as : a repeat of an incident already
being handled
the second incident
its own root cause : different from the first
the page it generated : suppressed as a duplicate
minutes it ran unattended : 140
is the dedup rule wrong : no; the fingerprints did
match
is a matching fingerprint the same incident : not
always, and here it was not
null control - fingerprint plus a correlation id
alerts, fingerprint alone :
1
alerts, with a correlation id :
2
incidents lost : 0
no alert changed its content; the key stopped collapsing
two causes that share a surface
what per-fingerprint dedup guarantees
every distinct fingerprint pages once : exactly, held
open while the condition persists
every incident pages someone : not addressed; alerts
dedup by fingerprint, and two distinct incidents shared
one - the second was suppressed as a duplicate and went
140 minutes unpaged
a fingerprint is a guess that sameness of surface is sameness of cause, and
dedup acts on the guess; when two causes wear one face, silencing the repeat
silences the second incident
It pages once per fingerprint and holds the group while the condition persists
- noise controlled. The fingerprint is service plus error class, and two
distinct incidents shared it, so the second was suppressed as a duplicate:
140 minutes unpaged, 5000 per ten thousand of incidents reaching nobody.Trace event types
eml:run:starteml:assigneml:outputeml:run:done