"Wishbone Is Only for Hobby Projects"

This claim is harder to dismantle than the previous two, because "hobby" is not a technical property of an interface at all. There is no port, no timing parameter and no rule that differs.

So the useful work is not defending the protocol. It is separating five variables that the word silently merges.

1. Five Variables, One Word

Only the last row is about Wishbone, and it is the row that carries none of the meaning people intend. A disciplined AXI peripheral and a careless AXI peripheral differ on rows three and four; so do two Wishbone peripherals. The bus name is constant across the entire quality range.

2. The Experiment That Settles It

Hold the protocol fixed. Vary the implementation, and vary the discipline of the test pointed at it. Four rigs, one interface.

First, what a casual bring-up test looks at — a scoreboard and the protocol checks:

  WHAT A CASUAL TEST LOOKS AT - scoreboard and protocol only
  implementation   scoreboard  protocol P0  protocol P2  verdict
  weak                      0            0            0  looks fine
  disciplined               0            0            0  looks fine

Both pass. One of those implementations has an incomplete address decode and a byte-lane merge that ignores SEL.

Now the same two implementations under a test that seeds non-zero data, probes the window boundaries rather than the interior, and runs against a slave with wait states:

  WHAT A DISCIPLINED TEST LOOKS AT - and it is the SAME RTL
  implementation   owner errors  scoreboard  delivered mask  verdict
  weak                        6           2               f        FAIL
  disciplined                 0           0               2  looks fine

  THE WEAK IMPLEMENTATION APPEARS IN BOTH TABLES.
  The casual test reports 0 mismatches and 0 violations.
  The disciplined test reports 6 misrouted addresses and 2
  data mismatches, on identical RTL.

Six addresses reaching a target the documented map does not assign them to, and a partial write delivered with mask f where 2 was asked for — with zero protocol violations in every column of both tables.

  WHAT THE DISCIPLINED TEST DID DIFFERENTLY
    seeded non-zero data before the partial write;
    probed the window boundaries, not just the interior;
    ran against a slave with wait states.

Three decisions. None is about the protocol; all three are about the engineer.

3. Reading The Result Correctly

What this supports. For this implementation pair and these two stimuli, the defects were invisible to a scoreboard-and-protocol-checks test and visible to one that seeded data, swept boundaries and inserted wait states.

What this does not support. Anything about any project's quality, anyone's competence, or whether Wishbone designs are on average better or worse than designs using other interfaces. That last question would need a survey nobody has run, and this experiment is not it.

The transferable claim is narrower and stronger:

Protocol name does not confer engineering rigor. What a review can read is the decode, the error handling, the parameter checking, and whether the tests could have failed.

4. What Evidence About A Project Actually Supports

Module 29 inspected five repositories at recorded commits. The discipline there applies exactly here, and it is worth restating because this is the chapter where people reach for the wrong conclusion.

None of those rows supports "industry standard", and none supports "hobby" either. Both labels are claims about a population, and five repositories are not a population.

5. What A Reviewer Can Actually Read

If the label is empty, what replaces it? The six chapters of Module 30, applied to the specific artifact in front of you:

Six questions, all answerable from the artifact, none answerable from the protocol name. An unfamiliar block that answers all six is a good block regardless of which bus it speaks; one that answers none is not rescued by speaking a fashionable one.

6. The Design Consequence

Rejecting a block for its bus. A peripheral with a complete decode, tested at two latencies, with negative controls, is a lower integration risk than an untested one, whichever interface each of them speaks. The name is not the evidence, and treating it as the evidence means declining evidence you already have.

Accepting a block for its bus. The symmetric and more expensive error. The four rigs above are the demonstration: the weak implementation passes a casual test with zero violations on every column, and "it uses a serious protocol" adds exactly nothing to that result.

7. The Replacement Statement

8. What To Carry Forward

• Five variables, one word. Only one of them is the protocol, and it is the one that carries no meaning.
• The same RTL passed and failed depending on the test pointed at it — six misrouted addresses, zero protocol violations, identical design.
• A casual test that reports nothing has told you about the test.
• Evidence about a repository supports that repository at that commit. Not a population, in either direction.
• The six review questions are answerable from the artifact. Use them instead of the label.

Chapter 31.4 takes the word that makes protocol comparisons useless.