Question balance in Training and the Underworld

Peregrine · 2026-10-13 · counted from the banks and deal code on main, with the Darkroom's new quick-items filter applied

Training's games deal a reading-heavy mix from one shared bank, except the Darkroom, which now deals none. The Underworld splits in two: the Lineup is balanced across reading and writing but light on math, and the Make and Kingpin Showdown are half math. Two pairs of games deal identical pools, and geometry and expression-of-ideas are thin everywhere.

ReadingWritingMath

Bars show each game's eligible pool, which is the set a game can deal from. A single session can differ: on Fresh, the Wager and the Raid deal six questions round-robin (about two of each section), and My Misses deals the student's own misses within the game's pool.

Training

The Wager256 items · median 53 words
Reading 107 (inference and evidence 78, craft and structure 29) · Writing 56 (conventions 41, expression 10, other 5) · Math 93 (algebra 39, data 30, advanced 14, geometry 8, untagged 2). Whole bank, no filter.
The Raid256 items · median 53 words
The same pool as the Wager, dealt the same way.
The Tail reading-paced163 items · median 70 words
Reading 107 · Writing 56 · no math. The longest items in the app, which suits a game paced by reading time.
The Darkroom quick98 items · median 33 words
Writing 51 (conventions 41, expression 10) · Math 47 at 40 words or fewer (algebra 21, advanced 10, data 10, geometry 4, untagged 2) · no reading. Changed today.
Target Practice quick45 claims · median 23 words
Reading 24 (inference and evidence 17, craft 7) · Writing 6 · Math 15. Its own bank of evidence claims, 36 clean and 9 cons.
Beat the Shakedown clocked93 drills · median 42 words
Reading 24 · Writing 21 · Math 48 (12 in each domain). The generated drills, plus the student's open misses from the main bank.

Underworld

The Lineup175 rounds · median 49 words
Reading 71 (inference 55, craft 16) · Writing 68 (expression 34, conventions 32, other 2) · Math 36 (algebra 18, data 14, advanced 2, geometry 2). Its own rounds, 91 cons and 84 clean, six a night.
The Make90 drills · median 42 words
Reading 24 · Writing 21 · Math 45. Generated drills that carry a con.
Kingpin Showdown clockedup to 90 drills · median 42 words
The Make's pool, narrowed to drills whose con has a family and a correct move. The narrowed count was not checked.

By skill, across games

SkillWager / RaidTailDarkroomTarget PracticeShakedownLineupMake / Kingpin
Reading: inference and evidence7878017125512
Reading: craft and structure292907121612
Writing: conventions4141414123212
Writing: expression of ideas10101029349
Math: algebra39—21512189
Math: problem solving and data30—109121412
Math: advanced14—10012212
Math: geometry and trig8—4112212

Wager / Raid, Tail and Darkroom draw on the same 256-item bank, so their columns overlap. Shakedown and Make / Kingpin draw on the same generated drills. Five Wager-bank writing items in other skills and two untagged math items are left out of the table.

What stands out

  1. Training's reading comes from three games on one bank. The Wager, the Raid and the Tail all deal the same 107 reading items. The Darkroom now deals none, which matches its pace, and Target Practice adds 24 short claims.
  2. Two pairs of games deal identical pools. The Wager and the Raid share all 256 items under the same filter. The Make and Kingpin Showdown share the generated drills. A student moving between paired games meets the same questions.
  3. The Underworld splits by game. The Lineup leans on reading and writing (21% math). The Make and Kingpin are half math. Taken together the room is roughly even, but each game alone is lopsided.
  4. Geometry and trig are thin outside the generated drills. The main bank has 8, the Lineup 2 and Target Practice 1. Only the drills (12) give it a fair share.
  5. Expression of ideas is thin in the main bank. It has 10 items next to 41 conventions items. The Lineup has 34, so the skill exists in the content; the main bank simply hasn't been filled.
  6. The main bank can't be split by reading sub-skill. Of its 107 reading items, 103 carry no finer type, so inference, evidence, vocabulary and text structure can't be balanced against each other yet.
  7. The Darkroom's placement differs from the plan. The code places it in Training (GamesView.swift:368). The Darkroom plan and commit c589e77 put it in the Underworld.

Decisions to consider

Should the Raid stop mirroring the Wager?One option gives the Raid its own slice, for example math and writing only, or a pace-matched filter like the Darkroom's, so each Training game trains something distinct.
Where should new bank items go first?The counts point to geometry and trig, expression of ideas, and reading items tagged by sub-skill.
Which room should the Darkroom live in?The code says Training and the plan says the Underworld.
Four development stages again?With quicker questions, a roll would take 12 answers instead of 9. Stage presets would need retuning, and the 98-item pool would last about 8 rolls before repeating, down from about 11. Left undecided, as you asked.

Counts were taken from vigilance-bank.json, generated-vigilance-bank.json, generated-questions.json, target-practice.json and the Lineup rounds, using each game's filter from the deal code. Word counts are passage plus stem. Prime Suspect and the phone games sit outside these two rooms and are not counted. In playtest builds both rooms show only the Lineup.