Every quarter deferring the upgrade was correct

# Self-authored for the EML case corpus (no external origin). A dependency
# upgrade was deferred eight quarters running. Each deferral is scored under
# two different rules below, and the accumulated bill is computed.
#
# The reasons were real. A team that deferred once for a launch, once for an
# incident backlog, once for a hiring gap and once for a reorg has deferred
# four times for four good reasons, and nobody here was lazy.
#
# The upgrade's cost is not fixed. It grows with the distance, because each
# skipped version adds its own breaking changes to the pile. Its value, on the
# other hand, is realised in a quarter that is not the quarter you pay for it
# in - the quarter that upgrades ships nothing a user can see.
#
# So the answer depends entirely on which rule is applied. A rule that scores
# a quarter on what shipped in it never looks at the upgrade's cost at all. A
# stricter marginal rule - is this quarter's work worth the extra days that
# waiting one more quarter adds - does look at it, and still says defer for a
# long time. Both are scored below, against the same eight quarters.

# [quarter, versions behind, days to upgrade then, value of the work it displaces, why]
[["Q1", 1, 2, 5, "launch"], ["Q2", 2, 3, 6, "incident backlog"], ["Q3", 3, 5, 6, "hiring gap"], ["Q4", 5, 8, 7, "launch"], ["Q5", 7, 12, 9, "reorg"], ["Q6", 9, 17, 8, "compliance work"], ["Q7", 12, 24, 9, "launch"], ["Q8", 15, 34, 9, "the bill"]] => quarters

len(quarters) => n

"quarter   versions behind   upgrade cost   value shipped if deferred" ^0
for q in quarters:
    "  " + q[0] + "        " + str(q[1]) + "               " + str(q[2]) + " days       " + str(q[3]) + " days (" + q[4] + ")" ^0
"" ^0

# ---- rule A: what shipped this quarter ----
#
# An upgrade delivers nothing a user can see in the quarter it happens, so on
# a rule that scores the quarter on value shipped, the comparison is against
# zero and it is not close.

0 => upgrade_value_in_quarter
0 => defers_a
for q in quarters:
    if q[3] > upgrade_value_in_quarter:
        defers_a + 1 => defers_a
"rule A - score the quarter on value shipped in it" ^0
"  value an upgrade ships in its own quarter : " + str(upgrade_value_in_quarter) ^0
"  quarters where deferring ships more       : " + str(defers_a) + " of " + str(n) ^0
if defers_a == n:
    "  the rule defers every quarter, and the upgrade's cost never enters" ^0
    "  the comparison at all" ^0
"" ^0

# ---- rule B: the marginal cost of one more quarter ----
#
# The stricter rule a careful team actually uses: is this quarter's work worth
# more than the extra days that waiting one more quarter will add?

"rule B - this quarter's value against the cost that waiting adds" ^0
0 => defers_b
0 => scored_b
"" => flip
for i in [0:n - 2]:
    quarters[i + 1][2] - quarters[i][2] => added_by_waiting
    scored_b + 1 => scored_b
    "" => verdict
    if quarters[i][3] > added_by_waiting:
        "defer" => verdict
        defers_b + 1 => defers_b
    else:
        "upgrade" => verdict
        if flip == "":
            quarters[i][0] => flip
    "  " + quarters[i][0] + " : value " + str(quarters[i][3]) + " against " + str(added_by_waiting) + " days added by waiting -> " + verdict ^0
"  quarters where deferring is still correct : " + str(defers_b) + " of " + str(scored_b) ^0
if not (flip == ""):
    "  the first quarter the rule says upgrade : " + flip ^0
    "  so under the stricter rule the team was right " + str(defers_b) + " times running and" ^0
    "  wrong once, at the end" ^0
"" ^0

# ---- what the cost did while the decisions were being made ----

quarters[0][2] => first_cost
quarters[n - 1][2] => last_cost
"the upgrade's cost across the deferrals" ^0
"  at " + quarters[0][0] + " : " + str(first_cost) + " days, " + str(quarters[0][1]) + " version behind" ^0
"  at " + quarters[n - 1][0] + " : " + str(last_cost) + " days, " + str(quarters[n - 1][1]) + " versions behind" ^0
"  multiplied by " + str(int(last_cost / first_cost)) ^0
"" ^0

"cost added per quarter of waiting" ^0
0 => prev
for q in quarters:
    if prev > 0:
        "  " + q[0] + " : " + str(q[2] - prev) + " days added" ^0
    q[2] => prev
"  the increments grow, so waiting is not a flat carry" ^0
"" ^0

# ---- who deferred and who paid ----

# [quarter, engineers on the team who made that call, how many are still here at Q8]
[["Q1", 5, 1], ["Q2", 5, 1], ["Q3", 6, 2], ["Q4", 6, 2], ["Q5", 4, 2], ["Q6", 7, 4], ["Q7", 7, 5]] => rosters
0 => deciders
0 => still
for r in rosters:
    deciders + r[1] => deciders
    still + r[2] => still
"the people" ^0
"  decisions to defer : " + str(len(rosters)) ^0
"  engineer-decisions behind them : " + str(deciders) ^0
"  of those, still on the team when the bill lands : " + str(still) ^0
"  which is " + str(int(still * 100 / deciders)) + "%" ^0
"  the " + str(last_cost) + " days are paid by a team that made " + str(int(still * 100 / deciders)) + "% of the calls" ^0
"" ^0

# ---- what a fixed allowance would have cost ----

2 => allowance
0 => steady_total
for q in quarters:
    steady_total + allowance => steady_total
"upgrading every quarter within a fixed allowance" ^0
"  allowance : " + str(allowance) + " days a quarter" ^0
"  total across " + str(n) + " quarters : " + str(steady_total) + " days" ^0
"  against the single bill at " + quarters[n - 1][0] + " : " + str(last_cost) + " days" ^0
if steady_total < last_cost:
    "  the steady version is " + str(last_cost - steady_total) + " days cheaper in total" ^0
    "  and it never appears on any quarter's list of things that shipped" ^0
"" ^0

# ---- the control: a dependency with no breaking changes ----
#
# Where the versions skipped add nothing to the work, distance costs nothing
# and deferring really is free.

[["Q1", 1, 2], ["Q4", 5, 2], ["Q8", 15, 2]] => flat
"control - a dependency whose upgrade cost does not grow with distance" ^0
for f in flat:
    "  " + f[0] + " : " + str(f[1]) + " versions behind, " + str(f[2]) + " days to upgrade" ^0
flat[0][2] => f_first
flat[len(flat) - 1][2] => f_last
"  cost multiplied by " + str(int(f_last / f_first)) + " across the same span" ^0
"  here every deferral really was free, and the rule that deferred them was" ^0
"  measuring the whole cost" ^0
"" ^0

"The reasons were real and the stricter rule agreed with the team for six" ^0
"quarters running. An upgrade's cost lands in the quarter that does it and" ^0
"its value lands everywhere else, so the quarter is the wrong unit to ask." ^0

quarters = [["Q1", 1, 2, 5, "launch"], ["Q2", 2, 3, 6, "incident backlog"], ["Q3", 3, 5, 6, "hiring gap"], ["Q4", 5, 8, 7, "launch"], ["Q5", 7, 12, 9, "reorg"], ["Q6", 9, 17, 8, "compliance work"], ["Q7", 12, 24, 9, "launch"], ["Q8", 15, 34, 9, "the bill"]]
n = len(quarters)
print("quarter   versions behind   upgrade cost   value shipped if deferred")
for q in quarters:
    print("  " + q[0] + "        " + str(q[1]) + "               " + str(q[2]) + " days       " + str(q[3]) + " days (" + q[4] + ")")
print("")
upgrade_value_in_quarter = 0
defers_a = 0
for q in quarters:
    if q[3] > upgrade_value_in_quarter:
        defers_a = defers_a + 1
print("rule A - score the quarter on value shipped in it")
print("  value an upgrade ships in its own quarter : " + str(upgrade_value_in_quarter))
print("  quarters where deferring ships more       : " + str(defers_a) + " of " + str(n))
if defers_a == n:
    print("  the rule defers every quarter, and the upgrade's cost never enters")
    print("  the comparison at all")
print("")
print("rule B - this quarter's value against the cost that waiting adds")
defers_b = 0
scored_b = 0
flip = ""
for i in range(0, n - 2+1):
    added_by_waiting = quarters[i + 1][2] - quarters[i][2]
    scored_b = scored_b + 1
    verdict = ""
    if quarters[i][3] > added_by_waiting:
        verdict = "defer"
        defers_b = defers_b + 1
    else:
        verdict = "upgrade"
        if flip == "":
            flip = quarters[i][0]
    print("  " + quarters[i][0] + " : value " + str(quarters[i][3]) + " against " + str(added_by_waiting) + " days added by waiting -> " + verdict)
print("  quarters where deferring is still correct : " + str(defers_b) + " of " + str(scored_b))
if not flip == "":
    print("  the first quarter the rule says upgrade : " + flip)
    print("  so under the stricter rule the team was right " + str(defers_b) + " times running and")
    print("  wrong once, at the end")
print("")
first_cost = quarters[0][2]
last_cost = quarters[n - 1][2]
print("the upgrade's cost across the deferrals")
print("  at " + quarters[0][0] + " : " + str(first_cost) + " days, " + str(quarters[0][1]) + " version behind")
print("  at " + quarters[n - 1][0] + " : " + str(last_cost) + " days, " + str(quarters[n - 1][1]) + " versions behind")
print("  multiplied by " + str(int(last_cost / first_cost)))
print("")
print("cost added per quarter of waiting")
prev = 0
for q in quarters:
    if prev > 0:
        print("  " + q[0] + " : " + str(q[2] - prev) + " days added")
    prev = q[2]
print("  the increments grow, so waiting is not a flat carry")
print("")
rosters = [["Q1", 5, 1], ["Q2", 5, 1], ["Q3", 6, 2], ["Q4", 6, 2], ["Q5", 4, 2], ["Q6", 7, 4], ["Q7", 7, 5]]
deciders = 0
still = 0
for r in rosters:
    deciders = deciders + r[1]
    still = still + r[2]
print("the people")
print("  decisions to defer : " + str(len(rosters)))
print("  engineer-decisions behind them : " + str(deciders))
print("  of those, still on the team when the bill lands : " + str(still))
print("  which is " + str(int(still * 100 / deciders)) + "%")
print("  the " + str(last_cost) + " days are paid by a team that made " + str(int(still * 100 / deciders)) + "% of the calls")
print("")
allowance = 2
steady_total = 0
for q in quarters:
    steady_total = steady_total + allowance
print("upgrading every quarter within a fixed allowance")
print("  allowance : " + str(allowance) + " days a quarter")
print("  total across " + str(n) + " quarters : " + str(steady_total) + " days")
print("  against the single bill at " + quarters[n - 1][0] + " : " + str(last_cost) + " days")
if steady_total < last_cost:
    print("  the steady version is " + str(last_cost - steady_total) + " days cheaper in total")
    print("  and it never appears on any quarter's list of things that shipped")
print("")
flat = [["Q1", 1, 2], ["Q4", 5, 2], ["Q8", 15, 2]]
print("control - a dependency whose upgrade cost does not grow with distance")
for f in flat:
    print("  " + f[0] + " : " + str(f[1]) + " versions behind, " + str(f[2]) + " days to upgrade")
f_first = flat[0][2]
f_last = flat[len(flat) - 1][2]
print("  cost multiplied by " + str(int(f_last / f_first)) + " across the same span")
print("  here every deferral really was free, and the rule that deferred them was")
print("  measuring the whole cost")
print("")
print("The reasons were real and the stricter rule agreed with the team for six")
print("quarters running. An upgrade's cost lands in the quarter that does it and")
print("its value lands everywhere else, so the quarter is the wrong unit to ask.")

quarter   versions behind   upgrade cost   value shipped if deferred
  Q1        1               2 days       5 days (launch)
  Q2        2               3 days       6 days (incident backlog)
  Q3        3               5 days       6 days (hiring gap)
  Q4        5               8 days       7 days (launch)
  Q5        7               12 days       9 days (reorg)
  Q6        9               17 days       8 days (compliance work)
  Q7        12               24 days       9 days (launch)
  Q8        15               34 days       9 days (the bill)

rule A - score the quarter on value shipped in it
  value an upgrade ships in its own quarter : 0
  quarters where deferring ships more       : 8 of 8
  the rule defers every quarter, and the upgrade's cost never enters
  the comparison at all

rule B - this quarter's value against the cost that waiting adds
  Q1 : value 5 against 1 days added by waiting -> defer
  Q2 : value 6 against 2 days added by waiting -> defer
  Q3 : value 6 against 3 days added by waiting -> defer
  Q4 : value 7 against 4 days added by waiting -> defer
  Q5 : value 9 against 5 days added by waiting -> defer
  Q6 : value 8 against 7 days added by waiting -> defer
  Q7 : value 9 against 10 days added by waiting -> upgrade
  quarters where deferring is still correct : 6 of 7
  the first quarter the rule says upgrade : Q7
  so under the stricter rule the team was right 6 times running and
  wrong once, at the end

the upgrade's cost across the deferrals
  at Q1 : 2 days, 1 version behind
  at Q8 : 34 days, 15 versions behind
  multiplied by 17

cost added per quarter of waiting
  Q2 : 1 days added
  Q3 : 2 days added
  Q4 : 3 days added
  Q5 : 4 days added
  Q6 : 5 days added
  Q7 : 7 days added
  Q8 : 10 days added
  the increments grow, so waiting is not a flat carry

the people
  decisions to defer : 7
  engineer-decisions behind them : 40
  of those, still on the team when the bill lands : 17
  which is 42%
  the 34 days are paid by a team that made 42% of the calls

upgrading every quarter within a fixed allowance
  allowance : 2 days a quarter
  total across 8 quarters : 16 days
  against the single bill at Q8 : 34 days
  the steady version is 18 days cheaper in total
  and it never appears on any quarter's list of things that shipped

control - a dependency whose upgrade cost does not grow with distance
  Q1 : 1 versions behind, 2 days to upgrade
  Q4 : 5 versions behind, 2 days to upgrade
  Q8 : 15 versions behind, 2 days to upgrade
  cost multiplied by 1 across the same span
  here every deferral really was free, and the rule that deferred them was
  measuring the whole cost

The reasons were real and the stricter rule agreed with the team for six
quarters running. An upgrade's cost lands in the quarter that does it and
its value lands everywhere else, so the quarter is the wrong unit to ask.

Trace event types