Doing some testing with this, and it's not that simple.

While changing the computed font size would fix the font-size for any superscript text (for example), it would break example 6 in particular from the css-inline-spec:

p::first-letter {
  initial-letters: 3;
  color: red;
  width: 5em;
  text-align: right;
  margin-left: -5em;
}

p {
  margin-left: 5em;
}

Here, the user wants "-5em" to be computed against the specified font size.

Based on the discussion, seems like we should continue to say that the computed font-size is unaffected, and take @emilio’s suggestion of maintaining a multiplier, which will affect the used font size and other effects that key off of the used font size (which is a class of things we'll henceforth need to be more clear about).

I think the only things that need to key off the used font size are going to be anything (other than actual <length> values) that keys off the font-size or a character size in the css-text, css-text-decor, and css-inline modules.

Scanning through, looks like that list should include:

• baseline-shift percentage values if line-height is also a percentage (thus referencing the font-size but not via em units)
• letter-spacing and word-spacing percentage values
• tab-size number values
• text-decoration-* percentage values
• text-underline-offset percentage values

Some side effects:

• Need to decouple line-height percentages from em units, as they'll behave differently here. (And should have been different from the start but that ship sailed a couple decades ago.)
• If we need another concept of "used font size", e.g. if we introduce per-font multipliers against font-size to optically equalize glyphs across a fallback list, then we need to be careful not to conflate terminology with what we're doing here. That kind of effect should not be accounted for here.

Slight complication: #2040
Seems that sometimes people want to style the punctuation at the same font size as the body text. Not sure how to make both cases possible.

Invent a "uem" unit for used font-size? Give it a special meaning in initial-letters only?

@faceless2 If we could distinguish inherited font-size vs explicitly set font-size, maybe that would work? Like maybe:

• initial-letter sets some inheritable font-size multiplier which is part of the computed value of font-size (which is now a tuple, a length and a multiplier, with the length as the basis of 1em)
• When font-size inherits, or is set to a percentage, the multiplier inherits also.
• When font-size is set to a <length>, the multiplier is reset to 1.

(To help with terminology, we can say the "computed font size" is 1em, but the "used font size" is 1em×multiplier.)

I had to read that a few times but I think I see what you're aiming at. So you'd essentially redefine font-size, baseline-shift, text-underline-offset etc. to explicitly include the multiplier aspect of the font-size as part of their calculations, but most other properties (eg margin-left) would remain unchanged.

So based on some of the examples linked from #2040, if you wanted "1st as your initial-letter, with st super-scripted and keyed off the used font size, but the leading quote the same size as the paragraph size...

p {
   font-size: 16px;
}
.initial {
    initial-letter: 3 3;
    width: 5em;
    margin-left: -5em;
}
.initial .quote {
    font-size: 16px;
    vertical-align: text-top;
}
sup {
    baseline-shift: 0.4em;
    font-size: 0.5em;
}

<p>
 <span class="initial">
  <span class="quote">&lquo;</span>
  1<sup>st</sup>
 </span>
</p>

The units used in sup are multiplied by our new multiplier, because they're font-relative. But the units in .quote are not, because they're absolute. And the width and margin-left properties are unaffected by this multiplier.

Have I understood this correctly?

Not quite. font-size: 0.5em would reset the multiplier to 1, because it's a <length> value. But font-size: 50% would maintain the multiplier. This avoids any need to be careful about font-relative lengths generally. :) Only situations where font-relativity is expressed in percentages (line-height, font-size, and various text-* properties) would end up relative to the used font size.

The CSS Working Group just discussed initial-letter used vs computed font-size.