The oldest one is the one nobody questions

# Self-authored for the EML case corpus (no external origin). The older a
# workaround is, the more likely its condition is gone and the less likely
# anyone looks at it. Both halves are computed below.
#
# Not looking at old code is a reasonable prior. Old code has survived, it is
# not what changed last week, and the thing that broke this morning is almost
# always something recent. Attention going to the new is how debugging works
# and it is right.
#
# Applied to removal it points the wrong way. A workaround is removable when
# the condition it was written for stopped happening, and that becomes more
# likely as time passes. Review attention becomes less likely over the same
# time, so the two curves cross and never come back.
#
# Both are measured against the same list.

# [id, age in months, the condition still occurs, reviews in the last year]
[["w01", 2, 1, 3], ["w02", 4, 1, 2], ["w03", 6, 1, 2], ["w04", 9, 0, 2], ["w05", 14, 1, 1], ["w06", 18, 0, 1], ["w07", 22, 0, 1], ["w08", 30, 0, 1], ["w09", 38, 0, 0], ["w10", 44, 0, 0], ["w11", 51, 1, 0], ["w12", 60, 0, 0], ["w13", 72, 0, 0], ["w14", 84, 0, 0]] => items

len(items) => n
[12, 36, 999] => bands
["under a year", "one to three years", "over three years"] => band_names

def in_band(age, b):
    0 => lo
    if b > 0:
        bands[b - 1] => lo
    if age < lo:
        return 0
    if age >= bands[b]:
        return 0
    return 1

"workarounds : " + str(n) ^0
"" ^0

"age band              count   condition gone   reviews last year" ^0
for b in [0:2]:
    0 => c
    0 => gone
    0 => revs
    for it in items:
        if in_band(it[1], b) == 1:
            c + 1 => c
            revs + it[3] => revs
            if it[2] == 0:
                gone + 1 => gone
    "  " + band_names[b] + "   " + str(c) + "       " + str(gone) + " of " + str(c) + "           " + str(revs) ^0
"" ^0

0 => young_gone
0 => young_n
0 => old_gone
0 => old_n
0 => young_revs
0 => old_revs
for it in items:
    if it[1] < bands[0]:
        young_n + 1 => young_n
        young_revs + it[3] => young_revs
        if it[2] == 0:
            young_gone + 1 => young_gone
    if it[1] >= bands[1]:
        old_n + 1 => old_n
        old_revs + it[3] => old_revs
        if it[2] == 0:
            old_gone + 1 => old_gone

"the two ends of the list" ^0
"  under a year     : " + str(young_gone) + " of " + str(young_n) + " gone, " + str(young_revs) + " reviews" ^0
"  over three years : " + str(old_gone) + " of " + str(old_n) + " gone, " + str(old_revs) + " reviews" ^0
if old_gone * young_n > young_gone * old_n:
    "  the older group is likelier to be removable" ^0
if old_revs < young_revs:
    "  and it is looked at less, by " + str(young_revs - old_revs) + " reviews" ^0
"" ^0

# ---- what the current order finds ----
#
# Reviews land where attention lands, which is on the newest entries.

5 => sweep
"a review sweep of " + str(sweep) + " entries, newest first" ^0
0 => found_new
0 => seen
for it in items:
    if seen < sweep:
        seen + 1 => seen
        if it[2] == 0:
            found_new + 1 => found_new
"  removable found : " + str(found_new) + " of " + str(sweep) ^0
"" ^0

"the same sweep, oldest first" ^0
0 => found_old
0 => seen2
for i in [0:n - 1]:
    items[n - 1 - i] => it
    if seen2 < sweep:
        seen2 + 1 => seen2
        if it[2] == 0:
            found_old + 1 => found_old
"  removable found : " + str(found_old) + " of " + str(sweep) ^0
if found_old > found_new:
    "  " + str(found_old - found_new) + " more, from the same amount of reading" ^0
"" ^0

# ---- the entry that breaks the rule ----
#
# Age is a correlate, not a fact about any one entry. One old workaround here
# is still load-bearing, and ordering by age is a heuristic rather than a
# licence to delete.

"old entries whose condition still occurs" ^0
for it in items:
    if it[1] >= bands[1]:
        if it[2] == 1:
            "  " + it[0] + " : " + str(it[1]) + " months old and still firing" ^0
"  so the sweep reads them, it does not delete them" ^0
"" ^0

# ---- the total that never shrinks ----

0 => gone_total
for it in items:
    if it[2] == 0:
        gone_total + 1 => gone_total
"removable right now : " + str(gone_total) + " of " + str(n) ^0
0 => zero_review
for it in items:
    if it[3] == 0:
        if it[2] == 0:
            zero_review + 1 => zero_review
"  of those, never reviewed in the last year : " + str(zero_review) ^0
if zero_review > 0:
    "  they will still be here next year, for the same reason they are here now" ^0
"" ^0

# ---- the control: a list where age says nothing ----
#
# Where removability is independent of age, reviewing oldest-first and
# newest-first find the same amount, and the ordering is not the lever.

[["a", 2, 0, 1], ["b", 20, 1, 1], ["c", 40, 0, 1], ["d", 70, 1, 1]] => flat
0 => f_first
0 => f_last
0 => k
for it in flat:
    if k < 2:
        if it[2] == 0:
            f_first + 1 => f_first
    k + 1 => k
0 => k2
for i in [0:len(flat) - 1]:
    flat[len(flat) - 1 - i] => it
    if k2 < 2:
        if it[2] == 0:
            f_last + 1 => f_last
    k2 + 1 => k2
"control - a list where removability does not track age" ^0
"  newest two : " + str(f_first) + " removable, oldest two : " + str(f_last) + " removable" ^0
if f_first == f_last:
    "  the same, so ordering by age would buy nothing here" ^0
"" ^0

"Spending attention on what changed recently is how anything gets debugged." ^0
"Removability accumulates in the other direction, so the two orders are" ^0
"opposite and only one of them is anybody's habit." ^0

items = [["w01", 2, 1, 3], ["w02", 4, 1, 2], ["w03", 6, 1, 2], ["w04", 9, 0, 2], ["w05", 14, 1, 1], ["w06", 18, 0, 1], ["w07", 22, 0, 1], ["w08", 30, 0, 1], ["w09", 38, 0, 0], ["w10", 44, 0, 0], ["w11", 51, 1, 0], ["w12", 60, 0, 0], ["w13", 72, 0, 0], ["w14", 84, 0, 0]]
n = len(items)
bands = [12, 36, 999]
band_names = ["under a year", "one to three years", "over three years"]

def in_band(age, b):
    lo = 0
    if b > 0:
        lo = bands[b - 1]
    if age < lo:
        return 0
    if age >= bands[b]:
        return 0
    return 1

print("workarounds : " + str(n))
print("")
print("age band              count   condition gone   reviews last year")
for b in range(0, 3):
    c = 0
    gone = 0
    revs = 0
    for it in items:
        if in_band(it[1], b) == 1:
            c = c + 1
            revs = revs + it[3]
            if it[2] == 0:
                gone = gone + 1
    print("  " + band_names[b] + "   " + str(c) + "       " + str(gone) + " of " + str(c) + "           " + str(revs))
print("")
young_gone = 0
young_n = 0
old_gone = 0
old_n = 0
young_revs = 0
old_revs = 0
for it in items:
    if it[1] < bands[0]:
        young_n = young_n + 1
        young_revs = young_revs + it[3]
        if it[2] == 0:
            young_gone = young_gone + 1
    if it[1] >= bands[1]:
        old_n = old_n + 1
        old_revs = old_revs + it[3]
        if it[2] == 0:
            old_gone = old_gone + 1
print("the two ends of the list")
print("  under a year     : " + str(young_gone) + " of " + str(young_n) + " gone, " + str(young_revs) + " reviews")
print("  over three years : " + str(old_gone) + " of " + str(old_n) + " gone, " + str(old_revs) + " reviews")
if old_gone * young_n > young_gone * old_n:
    print("  the older group is likelier to be removable")
if old_revs < young_revs:
    print("  and it is looked at less, by " + str(young_revs - old_revs) + " reviews")
print("")
sweep = 5
print("a review sweep of " + str(sweep) + " entries, newest first")
found_new = 0
seen = 0
for it in items:
    if seen < sweep:
        seen = seen + 1
        if it[2] == 0:
            found_new = found_new + 1
print("  removable found : " + str(found_new) + " of " + str(sweep))
print("")
print("the same sweep, oldest first")
found_old = 0
seen2 = 0
for i in range(0, n):
    it = items[n - 1 - i]
    if seen2 < sweep:
        seen2 = seen2 + 1
        if it[2] == 0:
            found_old = found_old + 1
print("  removable found : " + str(found_old) + " of " + str(sweep))
if found_old > found_new:
    print("  " + str(found_old - found_new) + " more, from the same amount of reading")
print("")
print("old entries whose condition still occurs")
for it in items:
    if it[1] >= bands[1]:
        if it[2] == 1:
            print("  " + it[0] + " : " + str(it[1]) + " months old and still firing")
print("  so the sweep reads them, it does not delete them")
print("")
gone_total = 0
for it in items:
    if it[2] == 0:
        gone_total = gone_total + 1
print("removable right now : " + str(gone_total) + " of " + str(n))
zero_review = 0
for it in items:
    if it[3] == 0:
        if it[2] == 0:
            zero_review = zero_review + 1
print("  of those, never reviewed in the last year : " + str(zero_review))
if zero_review > 0:
    print("  they will still be here next year, for the same reason they are here now")
print("")
flat = [["a", 2, 0, 1], ["b", 20, 1, 1], ["c", 40, 0, 1], ["d", 70, 1, 1]]
f_first = 0
f_last = 0
k = 0
for it in flat:
    if k < 2:
        if it[2] == 0:
            f_first = f_first + 1
    k = k + 1
k2 = 0
for i in range(0, len(flat)):
    it = flat[len(flat) - 1 - i]
    if k2 < 2:
        if it[2] == 0:
            f_last = f_last + 1
    k2 = k2 + 1
print("control - a list where removability does not track age")
print("  newest two : " + str(f_first) + " removable, oldest two : " + str(f_last) + " removable")
if f_first == f_last:
    print("  the same, so ordering by age would buy nothing here")
print("")
print("Spending attention on what changed recently is how anything gets debugged.")
print("Removability accumulates in the other direction, so the two orders are")
print("opposite and only one of them is anybody's habit.")

workarounds : 14

age band              count   condition gone   reviews last year
  under a year   4       1 of 4           9
  one to three years   4       3 of 4           4
  over three years   6       5 of 6           0

the two ends of the list
  under a year     : 1 of 4 gone, 9 reviews
  over three years : 5 of 6 gone, 0 reviews
  the older group is likelier to be removable
  and it is looked at less, by 9 reviews

a review sweep of 5 entries, newest first
  removable found : 1 of 5

the same sweep, oldest first
  removable found : 4 of 5
  3 more, from the same amount of reading

old entries whose condition still occurs
  w11 : 51 months old and still firing
  so the sweep reads them, it does not delete them

removable right now : 9 of 14
  of those, never reviewed in the last year : 5
  they will still be here next year, for the same reason they are here now

control - a list where removability does not track age
  newest two : 1 removable, oldest two : 1 removable
  the same, so ordering by age would buy nothing here

Spending attention on what changed recently is how anything gets debugged.
Removability accumulates in the other direction, so the two orders are
opposite and only one of them is anybody's habit.

Trace event types