Seven Karakas or Eight?

The disagreement is in the source, not in the arithmetic

If another site has given you a different atmakaraka, the arithmetic is almost certainly identical in both. What differs is the list of planets they were willing to rank.

One school ranks the Sun, Moon, Mars, Mercury, Jupiter, Venus and Saturn. The other adds Rahu to that list. The office goes to whichever candidate stands at the highest degree. So adding a candidate can take it away from the planet that held it.

Where each reading comes from

The fork is textual and it is old. Jaimini's own sutra at 1.1.10 reads saptanam astanam va, which translates as "of the classical planets, or of that set plus Rahu". The word va is the Sanskrit "or". The founding text of the system states the ambiguity rather than resolving it.

The Brihat Parashara Hora Shastra recites several positions at 32.1 to 32.2 and adjudicates none of them. Among modern teachers the split holds: K. N. Rao and Iranganti Rangacharya teach seven always. P. V. R. Narasimha Rao wrote the Jagannatha Hora software. He teaches a conditional eight. It engages only when a pair of those planets share the same whole degree as each other.

So nobody is repeating a corruption. Both readings are defensible, which is exactly why calculators disagree without any of them being broken.

Rahu's extra complication

Rahu is a computed point rather than a body, and it moves backwards through the zodiac. Schools that include it therefore do not use its degree as printed. They subtract it from 30 first. The reasoning is given at BPHS 32.5. The stated cause there is vilomagamanat, meaning "because of its reverse motion".

That introduces a second question underneath the first. Rahu's position depends on whether you use its mean node or its true node. The mean and the true node differ by up to about a degree and a half. Ranking eight means answering both questions. Ranking seven means answering neither.

What we compute, and why

We rank seven, and we do it at full floating-point precision rather than on rounded degrees.

The conditional-eight rule is the interesting case to test, because its trigger depends entirely on what you accept as "the same degree". Measured across real charts, how often that rule fires depends on what counts as equal. At the same whole degree it fires on about 51.6 percent. At the arcminute it fires on 0.9 percent. At the arcsecond it fires on 0.05 percent. A rule can swing by a factor of a thousand depending on how precisely you look. That is a poor thing to hang a reading on, however well attested it is. At full precision it effectively never fires, so we do not use it.

Seven is also the modal teaching position in India, and it avoids the mean-versus-true node question entirely.

How much it actually changes

Less than the argument suggests, and not evenly.

The choice moves the atmakaraka itself in about 12 percent of charts. It moves the amatyakaraka, the second rung and the one a vocational reading leans on, in about 25 percent. So the fork matters roughly twice as often for career reading as it does for the self. Check which scheme produced yours.

If your calculators disagree, you are most likely in that 12 percent. Check which scheme each one used before deciding either is wrong.