Reference: Virtual Doubling vs. Compensatory Lengthening (II-Guttural Piel/Pual)

In the Piel (and Pual), R2 must carry a Dagesh Forte. When R2 is one of the gutturals (א ה ח ע) — or ר, which behaves the same way despite not being a true guttural — that dagesh cannot be written. Two different repair strategies are attested:

• Virtual doubling — no Dagesh is written, but the preceding vowel stays short, as if the doubling were still phonologically "felt."
• Compensatory lengthening — no Dagesh is written, and the preceding vowel lengthens to compensate (Piel: Hireq → Tsere; Pual: Qibbuts → Holem).

The rule, by which letter occupies R2

R2 letter	Typical outcome	Why
ר	Compensatory lengthening — essentially exceptionless	Not a true guttural at all (a liquid/sonorant); Hebrew's resistance to a doubled ר is strong enough that the preceding vowel must lengthen to preserve the syllable's weight.
ח, ע	Virtual doubling — the default, though not exceptionless	True pharyngeal fricatives with enough articulatory substance that the syllable can still be "felt" as closed without a written dagesh, so the vowel doesn't need to lengthen.
ה	Genuinely mixed	No reliable majority direction — must be checked per root.
א	Genuinely mixed, though compensatory lengthening is somewhat more common	By the Masoretic period א had lost most of its consonantal force; its behavior is closer to a frozen, per-root convention than a live rule.

Corpus check (every 3ms Perfect Piel/Pual token in the Hebrew Bible with a guttural or ר in R2)

R2	Roots showing virtual doubling	Roots showing compensatory lengthening
ר	(none)	ברך, חרף, שרש — 14/14 distinct roots
ח	כחד, כחש, נחל, נחם, רחם, רחק, שחת — 7/7 distinct roots	(none, once confounded cases are excluded)
ע	בער	ערה, רעה
ה	כהן, מהר, נהג	כהה, קהה
א	נאץ, נאר	באר, מאן, ארש

Every ר-medial root and every ח-medial root in this sample lands exactly where the rule predicts, with zero exceptions. ע, ה, and א each show real roots going both directions.

Practical takeaway

• ר → always compensatory lengthening. ח → always virtual doubling. These can be predicted reliably.
• ע → usually virtual doubling, but confirm exceptions like רעה.
• ה and א → no safe prediction. These are lexically fixed per root — check the actual attested pointing (BDB/HALOT) rather than deriving it.

So the pattern is not simply "memorize each verb" (there's real phonetic logic behind ר and ח), but it is also not a fully general phonological rule — for ה- and א-medial roots, the Masoretic vocalization preserves what amounts to a frozen, per-root convention.