Why it matters

Thousands of failing endpoints are not thousands of problems. Endpoints that fail in the same set of views usually share a cause. Grouping them by failing-view signature reduces a list to a handful of work items.

Worked example

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

from collections import defaultdict
views = ["func_ss", "func_ff", "scan_ss"]
slack = {     # endpoint -> slack per view (ns), setup and hold together are not mixed here: setup slack only
    "e1": [-0.04, 0.20, 0.10], "e2": [-0.06, 0.18, 0.12], "e3": [-0.03, -0.02, 0.05],
    "e4": [0.10, 0.15, -0.07], "e5": [-0.05, 0.19, 0.11], "e6": [0.22, 0.30, 0.25],
}
groups = defaultdict(list)
for ep, s in slack.items():
    sig = tuple(v for v, x in zip(views, s) if x < 0)
    if sig:
        groups[sig].append(ep)
assert dict(groups) == {("func_ss",): ["e1", "e2", "e5"], ("func_ss", "func_ff"): ["e3"], ("scan_ss",): ["e4"]}
sizes = sorted(((len(v), k) for k, v in groups.items()), reverse=True)
assert sizes[0] == (3, ("func_ss",))
print("PASS: groups", {k: v for k, v in groups.items()})

Six endpoints become three groups and one clean one. The three that fail only in the slow functional view are one work item. The endpoint failing setup in two functional views needs separate attention because it is already failing in a second corner, which is physically plausible where fixed external delays or dominating clock skew outweigh the faster cells in the fast view. The scan failure has a different owner.

What the review looks for

• A triage table with signature, count, owner and next action, reviewed every day during closure.

• Signatures with one endpoint are not ignored: a singleton can be the most severe.

• Groups are reassigned when a fix changes the signature of its members.

Related topics

• Symptoms to Views

• Next: Scenario Bisect