How the engine works.

Most tools track one thing: right or wrong, maybe by broad topic. Tachyon tracks two.

First, the sub-skill. Not “Transitions” but TRN_CONTRAST versus TRN_CAUSE_EFFECT. Not “linear functions” but LINF_READ_GRAPH. There are 44 of these in Reading and Writing and over a hundred in Math. That axis decides what you practice next.

Second, the error tag. Every wrong answer choice in the bank is labeled with the specific reason a student picks it. That axis decides which misconception to fix.

In Reading and Writing, a wrong choice might be a REVERSAL (you flipped cause and effect), OUT_OF_SCOPE (true, but not what the passage is about), or TOO_EXTREME (the right idea pushed past what the text supports).

In Math, it might be GRAPH_INTERCEPT (you read the y-intercept when the question wanted the x), PROC_SLOPE_CALC (a slip in the slope arithmetic), or TRAP_SWITCHED_VARIABLES (you solved for the wrong variable). Math is tagged even more finely than Reading and Writing.

None of this is scraped. The bank is original, hand-verified, and difficulty-calibrated, and the tags are built in when the question is written.

One miss is noise. A pattern is a signal. When the same error tag shows up a third time, the engine stops the drill for a 90-second lesson on exactly the trap you keep falling for.

Then it hands you fresh questions built to spring that same trap, to check the fix held. Like a tutor leaning over your shoulder, except it’s 2am and it never gets bored.

As a weakness closes, its questions stop showing up as often and the next weakness moves to the front. You’re always spending your limited time on what you haven’t fixed yet.

That’s the whole idea. Practice aimed at where you’re actually losing points, re-aimed after every answer.