Why it matters

The scenario matrix is large. Most endpoints are limited by a few scenarios, which is the dominant-corner idea applied to regression. A fast loop runs only those and tells you in minutes whether a batch helped; the full matrix runs overnight and checks the rest.

Worked example

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

owned = {"s1": 420, "s2": 310, "s3": 180, "s4": 40, "s5": 25, "s6": 15, "s7": 6, "s8": 4}     # endpoints whose worst slack is in each scenario
total = sum(owned.values())
assert total == 1000
order = sorted(owned, key=lambda s: -owned[s])
coverage = {}
running = 0
for i, s in enumerate(order, start=1):
    running += owned[s]
    coverage[i] = running / total
assert [round(coverage[i], 3) for i in (1, 2, 3, 4)] == [0.42, 0.73, 0.91, 0.95]
fast_set = order[:3]
assert fast_set == ["s1", "s2", "s3"]
uncovered = total - sum(owned[s] for s in fast_set)
assert uncovered == 90
print("PASS: fast set", fast_set, "covers", round(100 * coverage[3], 1), "% of owned endpoints; uncovered", uncovered)

Three of eight scenarios own 910 of 1000 endpoints, so a fast loop of three scenarios covers 91% of endpoints by worst-slack ownership. The remaining 90 endpoints are owned by five other scenarios and are seen only in the full run. Coverage grows slowly after the third scenario, which is why the fast set stops there. Ownership changes as fixes land (a fixed scenario hands its endpoints to the next worst), so the table is refreshed regularly.

Two loops

What the review looks for

• The fast set is derived from current ownership data and re-derived after large changes.

• The uncovered fraction is stated, so nobody mistakes a fast result for a signoff one.

• Failures that appear only in the full run are analysed: they show where the fast set is blind.

Related topics

• What to Rerun

• Dominant Corner Analysis

• Next: Avoiding Fix Breaks Fix