Exercise 3: An 8-Stage Pipeline, Same Bug Position — Possible Solution
====================================================================

SETUP
------------------------------
Extended the chapter's own 4-stage pipeline (validate_cart ->
calculate_subtotal -> apply_tax -> apply_shipping) into 8 stages by
running a second identical pass over the first pass's own output,
using fresh function names (validate_cart_2, calculate_subtotal_2,
apply_tax_2, apply_shipping_2) for stages 5-8. The bug stays exactly
where the chapter placed it: stage 3, apply_tax.

RESULTS
------------------------------
    e2e result: expected $33.73, got $49.28, PASSES: False

    Found it after inspecting 3 of 8 stages (stage 3):
      expected 21.6, got 36.0

Stages checked before finding it: 3 - identical to the chapter's own
4-stage result (3 of 4).

DID THE NUMBER CHANGE? NO - AND HERE'S WHY
------------------------------
Sequential bisection starts at stage 1 and stops the moment it finds a
mismatch. The bug's own position in the pipeline (stage 3) is what
determines how many stages get inspected - not the pipeline's total
length. Doubling the pipeline to 8 stages added 5 more stages
*downstream* of the bug, which bisection never has to reach, because
it already found the fault and stopped at stage 3.

WHY THIS WORKS AS AN ANSWER
------------------------------
This corrects a natural but wrong assumption: that a longer pipeline
automatically means worse fault localization for blind e2e bisection.
The real relationship is between bisection cost and the bug's OWN
position, not overall pipeline length - a bug near the front of a long
pipeline is no harder to find via bisection than the same bug in a
short pipeline, while a bug near the END of a long pipeline (not
tested here) would make the number of stages checked scale with the
pipeline's own length. This is exactly why the chapter's own unit-test
finding matters more as systems grow: unit tests are unaffected by a
bug's position in the pipeline (always 1 targeted check for the
component under test), while blind e2e bisection's cost is only
sometimes small - and depends on exactly where the bug happens to sit.