The shim is maintained so the migration never finishes

# Self-authored for the EML case corpus (no external origin). A compatibility
# shim translates between an old schema and a new one. One person maintains it
# by hand. What that does to the migration is computed below.
#
# The shim is good work. It has carried every field correctly for two years, it
# handles four edge cases the automated converter never did, and writing it was
# the right call - without it the cutover would have needed every consumer
# migrated in one weekend, which is how the previous attempt failed.
#
# A migration finishes when the old path stops working. The shim's whole
# purpose is to keep it working. So every quarter the shim is maintained is a
# quarter in which no consumer has to move, and the person keeping the old
# path alive is the reason the new one is still optional.
#
# Consumers are counted by which schema they read, each quarter.

# [quarter, consumers on old schema, on new schema, shim edge cases, days he spent on the shim]
[["Q1", 14, 0, 2, 9], ["Q2", 12, 2, 3, 7], ["Q3", 11, 3, 3, 6], ["Q4", 11, 3, 4, 8], ["Q5", 10, 4, 4, 7], ["Q6", 10, 4, 4, 6], ["Q7", 10, 4, 5, 9], ["Q8", 9, 5, 5, 8]] => quarters

len(quarters) => n
quarters[0] => first
quarters[n - 1] => last

"quarter   on old   on new   shim edge cases   his days on the shim" ^0
for q in quarters:
    "  " + q[0] + "        " + str(q[1]) + "        " + str(q[2]) + "        " + str(q[3]) + "                 " + str(q[4]) ^0
"" ^0

0 => shim_days
for q in quarters:
    shim_days + q[4] => shim_days
"consumers moved across " + str(n) + " quarters : " + str(last[2] - first[2]) ^0
"consumers still on the old schema  : " + str(last[1]) ^0
"his days spent on the shim         : " + str(shim_days) ^0
"" ^0

# ---- the rate ----

"migration rate" ^0
"  quarters elapsed : " + str(n) ^0
"  consumers moved  : " + str(last[2] - first[2]) ^0
"  remaining        : " + str(last[1]) ^0
"  quarters to finish at the observed rate : " + str(int(last[1] * n / (last[2] - first[2]))) ^0
"  which is " + str(int(last[1] * n / (last[2] - first[2]) / 4)) + " years" ^0
"" ^0

# ---- what a consumer's incentive looks like ----

"the decision a consumer team makes each quarter" ^0
"  cost of migrating now      : 4 to 12 days of their own work" ^0
"  cost of not migrating now  : 0, the shim carries them" ^0
"  date the old path stops working : not set" ^0
"  every one of those teams is making the correct call with the information" ^0
"  in front of it, and the information is that nothing forces the move" ^0
"" ^0

# ---- the shim is growing ----

"shim edge cases over time" ^0
for q in quarters:
    "  " + q[0] + " : " + str(q[3]) ^0
"  " + str(first[3]) + " -> " + str(last[3]) + ", growing by " + str(last[3] - first[3]) + " across the period" ^0
"  each new edge case is a behaviour the old path has that the new one must" ^0
"  eventually reproduce, so the migration target moves every time the shim" ^0
"  gets better at its job" ^0
"" ^0

# ---- his time, and what it is buying ----

int(shim_days / n) => per_quarter
"his days per quarter : " + str(per_quarter) ^0
"across " + str(n) + " quarters   : " + str(shim_days) + " days" ^0
"consumers migrated in that time : " + str(last[2] - first[2]) ^0
"  his days per consumer migrated : " + str(int(shim_days / (last[2] - first[2]))) ^0
"  and none of those days were spent migrating anybody - they were spent" ^0
"  making it unnecessary" ^0
"" ^0

# ---- what a date would do ----

"the same eight quarters with a shutdown date announced in Q1" ^0
"  consumers on the old schema at Q1 : " + str(first[1]) ^0
"  cost to migrate all of them       : " + str(first[1] * 8) + " days of consumer-team work" ^0
"  his shim days that would not be needed : " + str(shim_days) ^0
"  net : " + str(first[1] * 8 - shim_days) + " days more work in total" ^0
"  so the shim is genuinely cheaper in aggregate, and it is cheaper by" ^0
"  moving work from many teams onto one person indefinitely" ^0
"" ^0

# ---- what would end it ----

"the quantity that decides when this stops" ^0
"  consumers remaining : " + str(last[1]) + ", falls slowly" ^0
"  shim maintenance cost : " + str(per_quarter) + " days a quarter, flat" ^0
"  shim edge cases : " + str(last[3]) + ", rising" ^0
"  a shutdown date : none" ^0
"  of those four, only the last one can terminate the process, and it is" ^0
"  the only one that is not a measurement" ^0
"" ^0

# ---- the control: a path that was switched off ----
#
# Where the old path was given an end date, the consumers moved and the shim
# was deleted.

[["legacy auth", 9, 2, 0]] => ended
for e in ended:
    "control - " + e[0] + ", shutdown announced two quarters ahead" ^0
    "  consumers at announcement : " + str(e[1]) ^0
    "  quarters to complete      : " + str(e[2]) ^0
    "  shim still maintained     : " + str(e[3]) + " days a quarter" ^0
    "  same shape of problem, same size, and the difference is that the old" ^0
    "  path had a date on which it stopped working" ^0
"" ^0

"The shim carries every field correctly and it prevented a one-weekend" ^0
"cutover that had already failed once. A migration ends when the old path" ^0
"stops working, and " + str(shim_days) + " days of good work have kept it working." ^0

quarters = [["Q1", 14, 0, 2, 9], ["Q2", 12, 2, 3, 7], ["Q3", 11, 3, 3, 6], ["Q4", 11, 3, 4, 8], ["Q5", 10, 4, 4, 7], ["Q6", 10, 4, 4, 6], ["Q7", 10, 4, 5, 9], ["Q8", 9, 5, 5, 8]]
n = len(quarters)
first = quarters[0]
last = quarters[n - 1]
print("quarter   on old   on new   shim edge cases   his days on the shim")
for q in quarters:
    print("  " + q[0] + "        " + str(q[1]) + "        " + str(q[2]) + "        " + str(q[3]) + "                 " + str(q[4]))
print("")
shim_days = 0
for q in quarters:
    shim_days = shim_days + q[4]
print("consumers moved across " + str(n) + " quarters : " + str(last[2] - first[2]))
print("consumers still on the old schema  : " + str(last[1]))
print("his days spent on the shim         : " + str(shim_days))
print("")
print("migration rate")
print("  quarters elapsed : " + str(n))
print("  consumers moved  : " + str(last[2] - first[2]))
print("  remaining        : " + str(last[1]))
print("  quarters to finish at the observed rate : " + str(int(last[1] * n / (last[2] - first[2]))))
print("  which is " + str(int(last[1] * n / (last[2] - first[2]) / 4)) + " years")
print("")
print("the decision a consumer team makes each quarter")
print("  cost of migrating now      : 4 to 12 days of their own work")
print("  cost of not migrating now  : 0, the shim carries them")
print("  date the old path stops working : not set")
print("  every one of those teams is making the correct call with the information")
print("  in front of it, and the information is that nothing forces the move")
print("")
print("shim edge cases over time")
for q in quarters:
    print("  " + q[0] + " : " + str(q[3]))
print("  " + str(first[3]) + " -> " + str(last[3]) + ", growing by " + str(last[3] - first[3]) + " across the period")
print("  each new edge case is a behaviour the old path has that the new one must")
print("  eventually reproduce, so the migration target moves every time the shim")
print("  gets better at its job")
print("")
per_quarter = int(shim_days / n)
print("his days per quarter : " + str(per_quarter))
print("across " + str(n) + " quarters   : " + str(shim_days) + " days")
print("consumers migrated in that time : " + str(last[2] - first[2]))
print("  his days per consumer migrated : " + str(int(shim_days / (last[2] - first[2]))))
print("  and none of those days were spent migrating anybody - they were spent")
print("  making it unnecessary")
print("")
print("the same eight quarters with a shutdown date announced in Q1")
print("  consumers on the old schema at Q1 : " + str(first[1]))
print("  cost to migrate all of them       : " + str(first[1] * 8) + " days of consumer-team work")
print("  his shim days that would not be needed : " + str(shim_days))
print("  net : " + str(first[1] * 8 - shim_days) + " days more work in total")
print("  so the shim is genuinely cheaper in aggregate, and it is cheaper by")
print("  moving work from many teams onto one person indefinitely")
print("")
print("the quantity that decides when this stops")
print("  consumers remaining : " + str(last[1]) + ", falls slowly")
print("  shim maintenance cost : " + str(per_quarter) + " days a quarter, flat")
print("  shim edge cases : " + str(last[3]) + ", rising")
print("  a shutdown date : none")
print("  of those four, only the last one can terminate the process, and it is")
print("  the only one that is not a measurement")
print("")
ended = [["legacy auth", 9, 2, 0]]
for e in ended:
    print("control - " + e[0] + ", shutdown announced two quarters ahead")
    print("  consumers at announcement : " + str(e[1]))
    print("  quarters to complete      : " + str(e[2]))
    print("  shim still maintained     : " + str(e[3]) + " days a quarter")
    print("  same shape of problem, same size, and the difference is that the old")
    print("  path had a date on which it stopped working")
print("")
print("The shim carries every field correctly and it prevented a one-weekend")
print("cutover that had already failed once. A migration ends when the old path")
print("stops working, and " + str(shim_days) + " days of good work have kept it working.")

quarter   on old   on new   shim edge cases   his days on the shim
  Q1        14        0        2                 9
  Q2        12        2        3                 7
  Q3        11        3        3                 6
  Q4        11        3        4                 8
  Q5        10        4        4                 7
  Q6        10        4        4                 6
  Q7        10        4        5                 9
  Q8        9        5        5                 8

consumers moved across 8 quarters : 5
consumers still on the old schema  : 9
his days spent on the shim         : 60

migration rate
  quarters elapsed : 8
  consumers moved  : 5
  remaining        : 9
  quarters to finish at the observed rate : 14
  which is 3 years

the decision a consumer team makes each quarter
  cost of migrating now      : 4 to 12 days of their own work
  cost of not migrating now  : 0, the shim carries them
  date the old path stops working : not set
  every one of those teams is making the correct call with the information
  in front of it, and the information is that nothing forces the move

shim edge cases over time
  Q1 : 2
  Q2 : 3
  Q3 : 3
  Q4 : 4
  Q5 : 4
  Q6 : 4
  Q7 : 5
  Q8 : 5
  2 -> 5, growing by 3 across the period
  each new edge case is a behaviour the old path has that the new one must
  eventually reproduce, so the migration target moves every time the shim
  gets better at its job

his days per quarter : 7
across 8 quarters   : 60 days
consumers migrated in that time : 5
  his days per consumer migrated : 12
  and none of those days were spent migrating anybody - they were spent
  making it unnecessary

the same eight quarters with a shutdown date announced in Q1
  consumers on the old schema at Q1 : 14
  cost to migrate all of them       : 112 days of consumer-team work
  his shim days that would not be needed : 60
  net : 52 days more work in total
  so the shim is genuinely cheaper in aggregate, and it is cheaper by
  moving work from many teams onto one person indefinitely

the quantity that decides when this stops
  consumers remaining : 9, falls slowly
  shim maintenance cost : 7 days a quarter, flat
  shim edge cases : 5, rising
  a shutdown date : none
  of those four, only the last one can terminate the process, and it is
  the only one that is not a measurement

control - legacy auth, shutdown announced two quarters ahead
  consumers at announcement : 9
  quarters to complete      : 2
  shim still maintained     : 0 days a quarter
  same shape of problem, same size, and the difference is that the old
  path had a date on which it stopped working

The shim carries every field correctly and it prevented a one-weekend
cutover that had already failed once. A migration ends when the old path
stops working, and 60 days of good work have kept it working.

Trace event types