Vocabulon vs VocabLoco · grades 4–6 · 2,191 released ELA test documents

The Word-List Case

Which vocabulary list actually teaches the words state ELA tests test? We pulled every released test we could get, read 225+ real questions verbatim, classified 12,000+ words, and attacked our own numbers before publishing them.

Vocabulon wins every grade — by shrinking margins as grade rises.
Strong at grades 3–4, moderate at grade 5, slight at grade 6 readiness. Every verdict survived hostile statistical recomputation; three of our own claims didn't, and are retracted or corrected in the open. Neither list alone approaches the 90% goal — the add-lists and the new grade-6 list ship with this case.

The briefs

Grade 4

VOCABULON · STRONG · 3–0

44% of the incumbent's slots teach already-known words. Timing-corrected tested-word coverage runs 1.9× (20.5% vs 11.1%). Both lists miss 67% of tested words — a 122-word add-list closes the gap.

Grade 5

VOCABULON · MODERATE · 4–0

121 real questions transcribed. Tested targets 27.5% vs 12.5%; the hostile recount widened it. But ~45% of items are list-proof, and whole programs lean the other way. 139-word add-list included.

Grade 6

READINESS · SLIGHT + A NEW LIST

Neither product has a grade-6 list. Vocabulon's 3–5 words carry slightly better (robust channel: answer choices, 22.2% vs 17.3%). The deliverable: a 285-word evidence-built G6 list — 41 of the top-50 tested words handled vs 5.

The record

2,191

released test documents · 63 assessment programs · grades 3–6

225+

real test items read and transcribed verbatim by 37 reader agents

12,600

words tier/known-grade classified, dual-rated on tested words, triple-audited

269

agents across four workflows, every verdict adversarially attacked

Grade	Panel	Tested-word coverage (U vs L)	Wasted slots (L vs U)	The one honest sentence
3	3–0 strong	16.0% vs 6.7% of targets	65% vs 53%	Incumbent's list is full of sight words, contractions, and math jargon
4	3–0 strong	28.2% vs 16.2% (taught-by-test-day 20.5 vs 11.1)	44% vs 27%	Same productive words (194 vs 195) — the incumbent needs 102 more slots to deliver them
5	4–0 moderate	27.5% vs 12.5%	34% vs 27%	Real edge, but ~45% of items reward context-clue skill no list teaches
6	3–0 slight	choice-channel 22.2% vs 17.3% (robust); tested leg discounted	—	Neither has a G6 list; the proposed 285-word list is the actual answer

Grade-6 proposed list (CSV, 285 words)

Why to trust it

We argued against ourselves, in writing

Every grade verdict was attacked by dedicated adversarial reviewers with full data access, instructed to overturn it. All upheld — and their recomputations are printed, including the ones that found our errors.
Three of our own claims died in audit and are retracted on the pages: the convergence table (flips at high K after family dedup), doc-weighted tested counts ("spectacular ×5" was one shared PARCC item), and the grade-6 list's "every tested word seated" claim (4 missing words, now added and disclosed).
The corpus's known limits are stated, not hidden: no MAP/private-instrument items (argued by proxy), AI-rated difficulty labels (triple-audited at 89–92% agreement), margins to be read as bands, and ~40–45% of vocab points at grades 5–6 are list-proof — won by context-clue and figurative-language instruction, not by any list.
What would strengthen it further is listed on each page: NY item-map full census, a lemma-level recount as signoff condition, a coverage-to-points bridge, and confirming which tests Alpha students actually sit.

Bottom line for the decision: on the evidence, Vocabulon is the better chassis at every grade — provided the grade-4/5 add-lists and the grade-6 list ship with the switch, and context-clue/morphology instruction is funded alongside, because no word list alone reaches the 90% criterion.