You are right that there are many permutations and possibilities.

Just introducing a guardrail like that doesn’t fix things enough, though. If you had guardrails like “only use margin once the portfolio falls below 70% and repay the margin once it is back to 80%” then the sequence looks like:

You don’t use margin until 1974, after the crash in September of that year.
By May 1975 (just a few months later), markets have recovered, your portfolio is back above 80%, and you repay the margin.
A similar thing happens in 1982. You tap into margin for a few months but then repay it quickly.
The real problems set in by 1990. Your portfolio drops below 70% and never recovers. That means you continue to use margin. By 1994, you’re using 50% margin.
Technically, you last the entire 30 years and die with a (net) $166,000 left in your portfolio. That’s an improvement over failing before 30 years. But also have over $500,000 in margin debt at age 95 (and 77% margin loan to value).

I argue that’s just implausible that a retiree would actually be okay accumulating that much debt.

The main problem is that the guardrails break down when you get to an advanced age. There’s no need to maintain 70% portfolio value when you are, say, 85 years old. The guardrails are just too high. So do we get rid of the whole guardrails/margin approach after the first 10–15 years of retirement? Or change the guardrails over time?

A bigger problem is that — in the quick test I did — the guardrails are based on nominal amounts. Since, realistically, that’s what people look at. No one actually looks at inflation adjusted figures of portfolios. But that means the portfolio is actually way below 70% of its inflation-adjusted starting value.

If we set the guardrail to 70% inflation-adjusted then the debt spiral of the OP comes back.

It probably seems like I am coming across as overly negative on this idea. Don’t get me wrong, I think it has some appeal, especially for people that aren’t always worried about “worst case in history” scenarios. But I’m very cautious that we’d end up just data mining a set of guardrails that turns out not to be well-suited to future scenarios.