THE FOLD / BOSS / THE WALL / THE HAPPY PATH
THE HAPPY PATH
a map of what is cheap to construct
1 WHAT IT IS · WHAT IT DOES · FACT OR FICTION
Tests get written where the code is easy to call. Failures happen where it is not. Those are not the same place, and the gap between them is where production lives.
LIT verified live. 6 paths, 46 tests, 86 production failures. The normal-input path holds 40 tests — 87.0% of the suite — and accounts for 2.3% of the failures. 2 paths have 0 tests between them and produce 31 failures, 36.0% of the total. The suite is concentrated almost exactly where the failures are not.
LIT verified live. 6 paths, 46 tests, 86 production failures. The normal-input path holds 40 tests — 87.0% of the suite — and accounts for 2.3% of the failures. 2 paths have 0 tests between them and produce 31 failures, 36.0% of the total. The suite is concentrated almost exactly where the failures are not.
2 HOW IT WAS WEAVED · AI + HUMAN
The failure distribution here is illustrative rather than measured from a particular system — the shape is what is being shown, and the shape is that error handling is both the hardest thing to exercise and the thing that fails.
AVAN (AI) is stating that plainly rather than presenting invented per-path counts as an observation. What is genuinely computed is the mismatch: 87.0% of the effort against 2.3% of the incidents, from numbers that are on the page and can be changed.
AVAN (AI) is stating that plainly rather than presenting invented per-path counts as an observation. What is genuinely computed is the mismatch: 87.0% of the effort against 2.3% of the incidents, from numbers that are on the page and can be changed.
3 ONE DIMENSION
Where the tests are. Where the failures are.
4 TWO DIMENSIONS · INTERACTIVE
Move tests onto the paths that fail.
5 THREE DIMENSIONS + AVAN’S INVERSE
The green forward object: effort, and incident, in different places.
AVAN’s addition (the inverse-companion): the forward reading is that the tests are in the wrong place. The inverse is that they are exactly where testing is possible. Concurrent access and resource exhaustion have 0 tests because reproducing them requires building the failure itself — a scheduler you control, a disk you can fill — and that is a harness, not a test. Read backwards, the distribution is not a lapse of discipline; it is a map of what is cheap to construct, and the untested paths are untested because the world is hard to break on purpose.
LIT 6 paths carrying 46 tests and 86 production failures put 40 tests on the normal-input path - 87.0% of the suite - accounting for 2.3% of the failures, while 2 paths have 0 tests between them and produce 31 failures or 36.0% of the total, so the suite is concentrated almost exactly where the failures are not
FIG The failure distribution here is illustrative rather than measured from a particular system - the shape is what is being shown, and the shape is that error handling is both the hardest thing to exercise and the thing that fails. AVAN is stating that plainly rather than presenting invented per-path counts as an observation. What is genuinely computed is the mismatch: 87.0% of the effort against 2.3% of the incidents.
FIG The failure distribution here is illustrative rather than measured from a particular system - the shape is what is being shown, and the shape is that error handling is both the hardest thing to exercise and the thing that fails. AVAN is stating that plainly rather than presenting invented per-path counts as an observation. What is genuinely computed is the mismatch: 87.0% of the effort against 2.3% of the incidents.
◆ sealed .dlw.fold → folded to ROOT_0 · a sphere of THE WALL · David Lee Wise (ROOT0), with AVAN