The deck

false friend: UA карта; map = hartă. Missed 3x by SES-004 then recalled — the most drilled word in the ledger. Recovered 2026-08-03 after two misses in SES-003. Plural produced as cârți on 2026-08-09 — â written for ă. The word itself was right. Clean on 2026-08-10 — carte f cărți, the diacritic held this time. Plural produced as cârți on 2026-08-12 — the t to ț alternation fired correctly and the stem vowel took â instead of ă. Same error as 2026-08-09. The î-â position rule is satisfied, so this is a vowel choice, not a spelling-position mistake. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Plural produced as *cârți* again on 2026-08-15 with the singular supplied — the alternations t→ț and the ending are right, the stem vowel is â for ă. **Third sighting of the same choice** after 2026-08-09 and 2026-08-12, so it is stable. Not the input method: the immediately preceding answer in the same set spelled *străzi* with the learner's usual ă substitute, so the â here was a separate deliberate keystroke. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 produced as cartă from English — the -e ending stored as -ă. Definite form also failed the same session as carta. See the -e ending entry in the error log. Definite form asked cold 2026-08-24 and produced as carta, character for character the 2026-08-20 answer after the key. Correct as cartea later the same session once the rule was reframed (ERR-071). The definite cartea was correct on 2026-09-06, cold — this exact item failed on 2026-08-24 as carta. Produced as cartă EN to RO on 2026-09-07 — the third identical production of this wrong form after 2026-08-20 and the definite carta on 2026-08-24. A stored form: ending corrections given minutes earlier the same session carried to nine other words and not to this one.

the target itself; ≈ UA екзамен — true friend Clean EN→RO bare on 2026-08-15. Full package asked 2026-08-24: the word was retrieved, the gender came back masculine and the plural exameni followed from that call. Held at tier 3 per srs.md — retrieval succeeded, so this is a form finding (ERR-074), not evidence against the row.

EN false friend: library = bibliotecă. Learned on the third exposure. Produced as libroară with plural librare on 2026-08-10 — the word itself is no longer stored. Demoted 2 to 1. Retested 2026-08-13, EN→RO bare: produced as libreriă, the Romance shape the false-friend note warns against. The note did not block it. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. The EN library trap did not fire. Produced as librarie on 2026-09-07, cold at 18 days — the internal ă absent and the word otherwise right. The EN library trap did not fire this time. Demoted 2 to 1.

limba română; this repo's namesake Clean RO→EN on 2026-09-06 at 25 days, and limba produced correctly in the article round the same session.

false friend: UA/RU образ = image Clean EN→RO bare on 2026-08-15 — the образ false friend did not fire. Full package asked 2026-08-24: word and gender both right, plural obrazi for obraji — the z to j alternation did not fire. Held at tier 3, form failure only (ERR-074). Correct EN to RO bare on 2026-09-07, cold at 14 days; образ did not fire. Bare only, gate untouched.

Slavic cognate: UA приятель Clean RO→EN on 2026-09-06, held at 28 days — beyond the tier-4 interval, so this is a real retention read. Also produced prietenul correctly in the same session.

NOT ceas (=clock) — UA час trap; hour = oră Clean RO→EN on 2026-09-06, held at 25 days. Correct EN to RO on 2026-09-07; the UA час trap did not fire on this row, though it did on ceas.

false friend: UA/RU труп = corpse; corp is the neutral synonym Clean EN→RO on 2026-08-15 as a LEX-FF probe — труп did not fire. Clean RO→EN on 2026-09-06, held at 22 days. Correct EN to RO on 2026-09-07; труп did not fire, in a round built to make it.

eu iubesc (-i class, -esc- infix); UA любити — Slavic cognate that failed to fire unprompted; not iulie Clean RO→EN on 2026-09-06 at 25 days, but the eu-form failed the same session — asked a iubi eu and answered iub, the -esc- infix omitted. Held at 4 rather than promoted because a recognition pass and a production miss on one day are not a promotion. ERR-093.

eu vorbesc (-i class, -esc- infix); noi vorbim; vorbesc puțin românește; infinitive a vorbi, 1sg vorbesc. Missed 2026-08-03 as a verbe — the EN verb hook fires but the stem is lost. Infinitive produced as a vorbei on 2026-08-09 while the meaning was right — the citation form is not stored as a form. Correct on 2026-08-10 — first clean production of the infinitive on a leak-checked set. Retested 2026-08-13 at the 3-day interval, EN→RO bare: a vorbei. The learner declared it a real error, not a slip. The final vowel is the conjugation class and gates the present tense. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Produced as a vorbi, correct, on 2026-08-19 — EN→RO and cold at 4 days, the first clean production after a vorbei twice. EN→RO is a rung above the RO→EN pass the promotion gate asks for, so the second pass is met and the row moves to 2. Clean EN→RO bare on 2026-08-25 — promoted to 3, so the next asking is the full package. Asked a vorbi voi on 2026-09-06 and answered vorbesci — the -esc- infix added where it must vanish. The row's own note already carries noi vorbim. ERR-093.

eu învăț (-a class, bare); noi învățăm; keep apart from a înțelege; învăț românește; both directions like UA вчити. Right lexeme on 2026-08-07; produced as a învață — the 3sg form. Infinitive is a învăța, the two vowels swap Produced as a învața on 2026-08-19 — the final class vowel -a correct for the first time on record, the internal ă dropped, and the learner declared it a real error. Held rather than promoted because the verb was on that day's teaching page. Later in the same session it was produced correctly spelled for the wrong meaning, when a înțelege was asked. Asked EN→RO on 2026-09-06 and produced a învăță — the final class vowel written as ă. The learner declared it deliberate: they remembered past corrections about special characters on this word and added one. A stored correction being applied as a rule, not a slip. Demoted from 3. ERR-091. Produced correctly as a învăța EN to RO on 2026-09-07 — the first time on record with the internal ă and the final plain -a both right, one day after a învăță. Held at 2, not promoted: the interval is 1 day against 3, and endings had been corrected earlier in the same session.

eu înțeleg (-e class, no infix); keep apart from a învăța; nu înțeleg — survival phrase №1; same Latin root as intellect. Right lexeme on 2026-08-07 (ERR-011 did not repeat) but produced as a înțeleg — infinitive welded to the eu-form. Given as to know on 2026-08-09 — to know is a ști. Second meaning miss on this row. Produced as a înteleg on 2026-08-10 — the eu-form welded to the marker again, plus the ț without its comma. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. Asked EN→RO on 2026-08-19 and answered a învăța — the neighbour this row's own note warns about, again. Demoted. Asked RO→EN cold on 2026-09-06 and answered to know — a ști. The third recorded meaning miss on this row and the second time the answer is to know, spelled identically. A stored wrong form. ERR-090. Produced as a înțelegi EN to RO on 2026-09-07 — the meaning recovered one day after to know, and the -e class vowel written -i. The lexeme is no longer the problem on this row; the ending is.

a mulțumi + dat., like UA дякувати комусь; keep apart from mă numesc Tier 3 package asked EN to RO on 2026-08-17 and produced as mulțimesc — the unstressed u written as i. The learner had it right and changed it at the last moment, and declared the choice genuine, not a slip. Demoted 3 to 2.

until ~11:00; missed in the U01 graded check — confused with bună ziua. Produced as bună dimeniața on 2026-08-10, the i and e transposed. The learner was asked and confirmed it is a mistake, not a typo. Demoted 2 to 1. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 produced as bună dimineață — the greeting's definite ending dropped. Paired with bună seara the same session.

from ~18:00. Produced as bună saere on 2026-08-09 — the ea of seara written as ae. Produced as bună seară on 2026-08-10 — the ea is now correct and the final -a became -ă instead. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 produced as bună seară — the greeting's definite ending dropped, second sighting the same session after bună dimineața.

formal-neutral default greeting Clean RO→EN on 2026-09-06, held at 25 days.

informal: Ce mai faci; answer bine, mulțumesc. Missed 2026-08-03. Given as whats up on 2026-08-09 — meaning right, register wrong. The -ți makes it formal. ⚠️ The tu form is the single broken item in the whole register grid — produced as facesți and facești in SES-011, both times face- plus an ending borrowed from ești. The stem SHRINKS: fac + i, not face + something. And it sounds nothing like its partner — faci is one syllable, fach, with soft c and the non-syllabic final -i, against three in faceți, fa-CHETS. No sound anchor is why the phrase does not come back whole. Cleared on retest 2 of 2 the same session. Ce mă facei on 2026-08-10. facei is face plus i — the stem still does not shrink to fac-. ERR-022 reproduced with the word faceți printed nowhere in the session, which is the cold confirmation SES-011 could not obtain. Retested 2026-08-15, EN→RO bare at the 3-day interval: Ce mai feceți. The learner declared the stem vowel an attention slip and stated fac- is securely known — dropped from the marking and the tally per the declaration rule, and the row is scored clean. Promoted 2 to 3. The tu-form faci remains the broken half and is untested here. Asked EN→RO on 2026-09-06 and produced Cum ce faceți — the verb right, the frame built from the English how. The chunk is not stored whole. Demoted from 3.

semn = sign, cf. EN signify; Ce înseamnă „obraz"? Demoted in SES-011: produced as însegnia, and placed AFTER the quoted word — Ce obraz însegnia. Word order is Ce înseamnă X, verb second, exactly as English does it. Ce itself was correct Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer.

lit. "with pleasure"; cf. EN pleasure. Given as with pleasure on 2026-08-09 — the literal reading. The use is you are welcome. Produced as ce placere on 2026-08-10 — plăcere right, cu replaced by ce after a run of Ce mai items, and the ă lost. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 written cu placere, diacritic absent. Not the keyboard — înseamnă and stradă were typed correctly in the same round.

from a se numi ≈ UA називатися; answer Mă numesc…. Produced as Cum vă numesți on 2026-08-07 — formal vă with the tu ending -ești. REG zone. Produced as Mâ te numișt on 2026-08-09 — question and answer still fused, persons mixed. Correct later the same session when embedded in a real question. Clean 4 of 4 in SES-011 — both registers, switched in both directions, and produced whole from English. The stem num- does not change, so there is nothing to assemble wrongly. Clean again on 2026-08-10, this time on a leak-checked set. Retested 2026-08-13 at the 3-day interval: Cum ve numesți. numesți is nume- plus the ending of ești — the assembly strategy of ERR-022, on a chunk taught whole. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off.

neutral-formal; lit. "until the re-seeing" Produced as la revidere on 2026-08-17 — the unstressed e written as i, the same choice as mulțimesc in the same block. Declared genuine. Demoted 3 to 2. 2026-08-20 produced as la revidere — e swapped for i inside a phrase otherwise held.

survival phrase; also mai încet. SES-004: frame recalled, the word rar lost — mai takes a plain a. Correct 2026-08-03 Only vă rog came back on 2026-08-17; mai rar, the half that carries the meaning, was blank. The frame survives without its content word. Demoted 3 to 2. Blank EN to RO cold on 2026-08-24, while its own tail va rog was volunteered for a different prompt in the same round. Demoted from tier 2 (ERR-077).

for getting attention; mă here is a plain object clitic — excuse ME; the -ся analogy belongs to mă numesc and does not apply (corrected 2026-08-10). Produced as me scuzați on 2026-07-31 — the vowel is ă. Correct 2026-08-03 Blank on 2026-08-17, and the learner named the competitor themselves — îmi pare rău was sitting in the slot. Two apology phrases contending for one meaning. Demoted 3 to 2. Asked EN to RO cold on 2026-08-24 and answered va rog, which is please and is the tail of another phrase in the same round. Demoted from tier 2 (ERR-077). Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 14 days.

word order flips — the only greeting that puts bună second; for parting to sleep Produced as noaptă bună on 2026-08-17 — the noun is noapte and noaptă is not a Romanian word. Declared genuine, not a slip. Demoted 3 to 2. 2026-08-20 written noapte buna, diacritic absent, one line after ț was typed correctly.

te rog informal; lit. "I ask you". Blank on 2026-08-09 — no recall at all, the first total miss for this row. SES-011 promoted to 2 on 2 of 2, rolled back to 1 the same day — both items had vă rog printed in a neighbouring prompt. See work/feedback/2026-08-09_leak-audit.md. Correct on 2026-08-10 after a total blank two days earlier, on a leak-checked set. Blank on the meaning 2026-08-20 — produced mă scuzați instead, so the slot retrieved a neighbouring politeness formula rather than nothing. Demoted from 3.

regret, not apology; lit. "it seems bad to me" Clean at the tier 3 gate on 2026-08-17, EN to RO, at an 8-day interval. Promoted 3 to 4.

false friend: UA простий; simple = simplu Clean EN→RO on 2026-08-15 as a LEX-FF probe — простий did not fire. Clean RO→EN on 2026-09-06, held at 22 days. Correct EN to RO on 2026-09-07; простий did not fire, in a round built to make it.

Romance layer: cf. EN benefit, It. bene Clean RO→EN on 2026-09-06 at 25 days, answered good — the adverb sense is the meaning that matters and it surfaced. bun is the adjective.

RU да — the one place Russian is the direct helper; UA так does not help Clean EN→RO at the tier-3 gate on 2026-08-15, at a genuine 8-day interval — the package question degenerates to the bare word on a particle, so this is the whole gate the row can be asked. Promoted 3 to 4. Clean RO→EN on 2026-09-06, held at 22 days.

fără zahăr; not far (=lighthouse, headlight) which UA фара suggests Clean at the tier 3 gate on 2026-08-17, EN to RO, at an 8-day interval. Promoted 3 to 4. Clean EN→RO production on 2026-09-07, cold at 21 days — a rung above the tier-4 sweep, and фара did not fire. Promoted 4 to 5.

also negates verbs, placed before them: nu sunt student Clean RO→EN on 2026-09-06, held at 25 days.

informal, works in both directions Clean RO→EN on 2026-09-06, held at 25 days.

eu sunt / tu ești; ei/ele sunt — same form as eu sunt. Demoted SES-004: estim invented for sunt under load. Recovered to tier 2 on 2026-08-03. Promoted to 4 in SES-011 on 3 of 3 register switching, then rolled back to 3 the same day — 2 of those 3 items printed their own answer in a neighbouring prompt. See the leak audit in work/feedback/2026-08-09_leak-audit.md. The claim that ești and sunteți cannot be assembled from each other still holds; it is the score that does not On 2026-08-19 suntem and sunt were both correct, and Ește was written for este in a different item of the same block. Held rather than promoted. Mixed on 2026-09-06. este appeared in all three sentences that needed it — the copula deletion seen on 2026-09-06 in the U07 check did not repeat. But este was used with a plural subject where sunt belongs. Held at 3. ERR-096.

nu plus a vowel contracts and is written so; nu sunt does not. Produced correctly on first use 2026-07-31. Missed in review as N-ama but produced clean as N-am telefon later the same session 2026-08-03. Clean unprompted production N-am telefon on 2026-08-07 Clean on 2026-08-15: *Nu am o carte* → *N-am o carte*, form probe with the frame supplied.

a = EN "to", a separate word (UA uses the suffix -ти); drops when conjugated. Missed 2026-07-31. a vorbei and a înteleg produced on 2026-08-09 — the marker attaches but the form after it is not the infinitive. a vorbi correct on 2026-08-10; a înteleg and a învăță both wrong. The marker attaches reliably and the final vowel of the citation form is still guessed. On 2026-08-15 the marker was correct 3 of 3 (a fi, a ave, a învață) and the class vowel wrong 2 of 3. The row's output is a well-formed infinitive, so it holds at tier 1. Both outputs were well-formed infinitives on 2026-08-19 and a vorbi was correct for the first time, so the row's own criterion is met. Promoted.

nu goes before the VERB, never the noun; UA «не» maps directly

normally omitted, kept for contrast; ei m/mixed vs ele f only; no neuter "it". eu and ea collided, ei missing on 2026-08-03. All eight produced clean on 2026-08-09; dumneavoastră offered as a ninth, which is the formal address form and sits outside the paradigm ele and voi both clean on 2026-08-19; the subject pronoun was also correctly dropped in a production item of the same block.

î at word edges, â inside; one sound, two letters; graded as ORTH-DIA

faci/faceți, ai/aveți, ești/sunteți, te numești/vă numiți. Blended as faciți and aviți on 2026-08-03 — REG zone, exam-critical for S. SES-011 ran a dedicated repair loop and the premise was wrong: register naming 5 of 5, register switching 7 of 8, retest 6 of 6. The ending pair is not the fault. Three of the four verbs keep their stem (av-, num-, and a fi which shares no letters at all) and all three are clean; a face is the only one whose stem shrinks, fac- against face-, and it is the only one that fails. See ERR-022. ROLLED BACK to tier 1 the same day: 14 of that loop's 16 items printed their answers in other prompts, so the 5 of 5 and 7 of 8 measured copying. The a face diagnosis stands on the shape of the wrong answers; the verdict that the ending pair is sound does not. See work/feedback/2026-08-09_leak-audit.md. On 2026-08-10 aveți and vă numiți were correct and faci came out as facei, on a set that printed no answers. Same verdict as ERR-022, now cleanly evidenced. Split result on 2026-08-15. The *faci* half is FIXED — produced correctly, cold, after being the single broken item in the register grid since SES-011. The same *-i* ending then failed on *ești*, produced as *ește*. Demoted on the miss; the note carries the half that works. faci and faceți both correct on 2026-08-19, cold at 4 days, on an English-labelled prompt. The a face stem is the only cell that has ever failed here and it held. Promoted. Asked twice on 2026-08-25 inside other words: the -ți of învățați and the -i of lucri were both the right ending, and both words were wrong for other reasons. Held at 2 rather than promoted on a part of an answer.

no word order change, no auxiliary; UA matches, EN interferes Clean on 2026-08-19 — an English prompt carrying do-support returned a Romanian question with no auxiliary and no inversion.

resembles UA адреса; one d, unlike English. Plural produced as adrease on 2026-08-09 — an ea inserted where the plural is plain adrese. Not produced on 2026-08-10 — stradă came out instead, two minutes after stradă had been keyed. Semantic neighbours inside pe strada X la numărul Y. Clean bare on 2026-08-19, with its final ă. Retrieval and plural both clean on 2026-09-06 EN→RO. Held at 3 because the article half of the tier-3 gate was not asked.

also adresă de e-mail. Clean on 2026-08-10 — promoted to 3. No dexonline entry on the 2026-08-15 factcheck sweep — unverified per docs/mechanics/verification.md. Clean package on 2026-08-19 — un e-mail, e-mailuri. Still unverified against dexonline.

resembles UA інженер; the g is soft, in-ji-ner. Produced as ingener on 2026-08-09 — ge written for gi. EN engineer pulls the e across. ORTH-CG. 3 of 4 on 2026-08-10 — the feminine plural came out ingenere, ORTH-CG again and now in the second syllable. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer.

resembles UA медик; no feminine form, a woman is medic too. Clean twice on 2026-08-09 — meaning and doi medici with the c to ci softening Clean RO→EN on 2026-09-06 at 25 days, and medicul produced correctly in the article round.

resembles Latin nomen, EN name; plural identical to singular. Produced as numeștă on 2026-08-07 — the verb a se numi selected for the noun. Third noun-verb blend after a locui a lucra and a învăța a înțelege. Taken as feminine on 2026-08-09 — o nume două numei. It is NEUTER with the plural identical to the singular, and it is the exception that breaks gate 4 of the gender sheet. Taken as feminine again on 2026-08-10, plural given as numi. Second identical miss in two sessions on the gate-4 exception. Taken as feminine a third time on 2026-08-12 in the second drill of the day, and this time the word itself deformed — o numă, where the previous miss had at least kept nume intact. The Latin anchor nomen published hours earlier did not hold. Produced clean bare on 2026-08-19 at the 6-day interval. Taken as feminine again later in the same session when the gender was asked on its own — fourth sighting of the gate-4 exception. Retrieval clean EN→RO on 2026-09-06, but numea produced twice the same session — once for the plural and once for the definite. Held at 3, not demoted: only retrieval failure is evidence against the word. numea is a stored wrong form. ERR-090. Correct EN to RO bare on 2026-09-07 at a 1-day interval, after producing numea twice on 2026-09-06. The stored form did not re-emit once it had been named and shown beside the right form — the repair ERR-090 prescribed, tested at one day. Bare only, so the gate is untouched.

resembles EN number, Latin numerus; la numărul trei. Plural numare on 2026-08-10 — the stem vowel did not move. Demoted 2 to 1. Plural given as namare on 2026-08-10 — neither the stem vowel nor the ending survived. The worst-formed plural of the session and the only one where the singular was not recoverable from the attempt. Retested 2026-08-13, EN→RO bare: numer, ă written as e. Declared a real error. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off.

no hook — from Hungarian város; neither Latin nor Slavic, memorize flat. Missed 2026-08-03 again. Missed again 2026-08-07 as orâf, then as orasi, then replaced by țară — fifth miss across four sessions. No transfer hook at all (Hungarian város); this is retention, not orthography, so it is not coded — daily rotation instead. Sixth miss 2026-08-09 — doi orași for două orașe, gender and plural both. Still no transfer hook. Seventh miss 2026-08-10 — masculine again, plural orași. Still no transfer hook. On 2026-08-12 in the lesson warm-up the article went wrong again — o oraș with the plural orași, so the plural error is downstream of the gender, not separate. Third consecutive session on this one word, and this time feminine rather than masculine. Hours later, in the same session's graded check, the learner produced Un oraș mare with the article correct — the first time un has landed on this word. Asked for the singular on 2026-08-13 and answered orașe — the plural, correctly formed. See the plural-for-singular pattern. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Second clean RO→EN pass on 2026-08-25, cold after ten days — promoted to tier 2 at last.

resembles UA професор; the -oară feminine is the UA -ка instinct Tier 3 gate on 2026-08-17: article and singular correct, plural produced as profesoari. The -or to -oare opening that belongs to the feminine ran on the masculine plural. Retrieval succeeded and the form failed, so the row is held at 3 rather than demoted, per the split probe rule. Tier-3 gate 2026-08-25: un profesor correct, plural produced as profesoari. Retrieval clean, form wrong, held at 3 — see ERR-080.

trap: UA програміст → RO programator, not programist Tier 3 gate on 2026-08-17: article and singular correct, plural produced as programatoari. Same error as profesoari in the same block, one mechanism twice. Held at 3, not demoted — the form failed, not the retrieval. Tier-3 gate 2026-08-25: un programator correct, plural produced as programatoari. Same shape as profesor in the same set, held at 3 — see ERR-080.

â is the medial rule; m/f pair like UA українець and українка Full package produced clean on 2026-08-17 — un român, români — at the tier 3 gate. Promoted 3 to 4.

resembles the UA and EN names. No dexonline entry on the 2026-08-15 factcheck sweep — unverified per docs/mechanics/verification.md. Not produced on 2026-08-19 — Români, the demonym plural, came out instead of the country name. Retrieval failed, so demoted.

Latin strata, EN street; you live pe stradă, not în. Plural produced as strazi on 2026-08-09 — the d to z shift fired, the a to ă did not. Plural strazi again on 2026-08-10 — the d to z shift fires, the a to ă does not. Also produced in place of adresă in the next block, the two sitting inside one taught chunk. Plural străzi produced correctly on 2026-08-10, with both the a → ă shift and the d → z shift. Not a promotion — the answer was published the same morning. Written as strazi again on 2026-08-12, with the ă supplied in the prompt on the same line. Third occurrence of this exact miss. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Plural *străzi* also produced correctly with the singular supplied — both alternations, a→ă and d→z. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. Singular right and the plural produced as strazi on 2026-09-07, cold at 18 days — the d to z alternation fired and the stem a to ă did not.

resembles UA студент Full package produced clean on 2026-08-17 — un student, studenți — at the tier 3 gate. Promoted 3 to 4.

Slavic loan: UA сто and сотня. A noun, not a numeral — hence o sută, două sute Blank RO→EN on 2026-09-06 at 25 days — nothing surfaced. Demoted from 4. The plural sute was correct once supplied.

resembles UA телефон. Produced as telefone in practice 2026-08-07 — pluralised where the singular was wanted. Singular produced as telefone again on 2026-08-09 — second time the plural stem is used for the singular. Clean on 2026-08-10 — singular and plural both right for the first time. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Plural *telefoane* also produced correctly with the singular supplied (form probe). A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 first ask was withdrawn — the item did not name its direction and the learner answered in English. Re-asked with the direction stated and produced correctly. Instrument fault, not a miss.

resembles UA Україна; stress on the a-i — Ucraina. No dexonline entry on the 2026-08-15 factcheck sweep — unverified per docs/mechanics/verification.md. Clean at the tier-3 gate on 2026-08-19, cold at 10 days.

suffix is -ean not -ец. SES-004: produced as ucarini — letters transposed, ending masculine. Ending produced as -ance for -ancă on 2026-08-09 — the feminine suffix is -că. 2 of 4 on 2026-08-10 — both plurals kept the -ea- of the singular: ucraineani, ucraineance. The stem reduction to ucrainen- did not fire in either. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 answered ukrainian; the learner declared autocorrect and named ucrainean. Scored correct, no error recorded.

RU возраст is the direct cognate — same Slavic root, same meaning; the policy's RU clause, found by the 2026-08-10 audit after this row said no hook exists. â medial then final ă — the two vowels this learner conflates. Clean on 2026-08-10 — promoted to 3. Blank RO→EN on 2026-09-06 at 25 days — nothing surfaced. Demoted from 4. The plural vârste was correct minutes later once the word was supplied, so this is retrieval, not form.

resembles Latin terra, EN terrain and territory. Clean in review 2026-08-07; but produced as țari in practice the same day — the a→ă stem shift did not fire. Plural țare on 2026-08-10 — neither the -i ending nor the a to ă shift. Demoted 2 to 1. Plural given as țăre on 2026-08-10 — the stem alternation a → ă ran correctly and the ending -e was then attached from the wrong shelf. The two steps ran side by side instead of in sequence; this is the cleanest instance of the ordering fault the U03 routine page was built for. Clean bare on 2026-08-19, with its final ă. Retrieval clean EN→RO on 2026-09-06; plural produced as țare, the -e rule where -i belongs. Held at 3 — the form failed, not the word. ERR-092.

eu locuiesc (-i class, -esc- infix); cognate locate; Latin locus, EN locate. Keep apart from a lucra to work — blended as Lucuiesc on 2026-08-03. 1sg locuiesc, group -i is U06. Read correctly as to live on 2026-08-09 and then produced for to work minutes later — the link works one way only. Root produced correctly for to live in SES-011, but inside the taught chunk Unde locuiești — in frame only, so the bare English-to-Romanian direction is still untested. Held at tier 1 for that reason. Spelled locuești, without the i. Correct bare from English on 2026-08-10, in the same message as a lucra. ERR-019's open question — the untested EN to RO direction — is now closed, on a leak-checked set. 2026-08-20 the eu form came back lucuesc: the a lucra blend of 2026-08-03 recurring, plus the -i of the stem dropped before -esc. Both halves of ERR-019 live again at a 7-day interval.

eu lucrez (-a class, -ez- infix); cognate lucrative; EN lucrative, Latin lucrare; 1sg lucrez memorized as a form — group I present is U05. Selected as a locui for to work on 2026-08-09 — ERR-008 returns, and only in the EN to RO direction. Correct bare from English on 2026-08-10, in the same message as a locui and with neither blending into the other. Retested 2026-08-13 at the 3-day interval, EN→RO bare: a lucre. The learner declared it a real error, not a slip. The final vowel is the conjugation class and gates the present tense. a locui was clean in the same block, so the old locui/lucra confusion did not fire. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Produced as a lucri EN to RO on 2026-09-07, cold at 18 days — an overcorrection the session caused: four -e verbs had just been flagged for taking -i, and a lucra takes plain -a. Demoted 2 to 1.

= UA що. Pronounced che Clean at the tier 3 gate on 2026-08-17, EN to RO, at an 8-day interval. Promoted 3 to 4.

people only; = UA хто. First sound as in EN church — Romanian writes it bare c before e/i; RO ch is [k], so never spell the sound with ch (ORTH-CG). Given as how in isolation on 2026-08-09 then used correctly in Cine este profesor minutes later — retrievable inside a question, not alone. Correct bare AND in frame the same afternoon in SES-011 Clean bare on 2026-08-19 at the 7-day interval — the first honest test after the same-day repair of 2026-08-09, and the bare entry held.

= UA як; also asks names — Cum te numești, where EN what is the saboteur. Produced as cine for how in isolation on 2026-08-09 then used correctly in Cum vă numiți — same isolation failure as cine. Correct bare AND in frame the same afternoon in SES-011. Correct bare on 2026-08-10, three days after the isolation failure.

Latin quantus, EN quantity; Câți ani ai stays welded as one block before the verb. Produced as câți cate on 2026-08-09 — â right on the first form and missing on the second. Meaning correct bare in SES-011, but written Câti in Câti ani ai — the ț lost its comma. Second diacritic slip on this one word in one day; the â is now reliable and the ț is not. Held at tier 1. Both forms correct with both diacritics on 2026-08-10 — the ț that was lost twice on 2026-08-09 held. No dexonline entry on the 2026-08-15 factcheck sweep — unverified per docs/mechanics/verification.md.

Unde locuiești; = UA де. Given as from in isolation on 2026-08-09 then used correctly in Unde locuiești — UA де is nearly the same word and still did not fire alone. Correct bare AND in frame the same afternoon in SES-011. Correct bare on 2026-08-10, three days after the isolation failure.

Sunt din Ucraina; UA «з» Produced as unde on 2026-08-17 — where instead of from. A function word from the same family arrived in its place. Demoted 3 to 2. Asked EN to RO cold on 2026-08-24 and answered unde, which is where. Demoted from tier 2 (ERR-077).

UA «у мене є» has no verb — EN is the model. 2nd most frequent verb; also the perfect-compus auxiliary from U09. Au correct on 2026-08-09 but avesti and aviți produced for ai and aveți the same session. SES-011 read this as repaired and promoted to 2; rolled back to 1 the same day because both switch items printed their own answer in the neighbouring prompt. The true state of ai and aveți is UNKNOWN — wrong 2026-08-03, wrong the morning of 2026-08-09, and not validly tested since. Test it on English prompts only. See work/feedback/2026-08-09_leak-audit.md. Tested cold on 2026-08-10 with an English-only prompt, as ERR-023 demanded. aveți correct, the tu form given as au. So the dumneavoastră form is solid and ai is not — and au is a real slot of the paradigm, not an assembled string, which makes this a different failure from ERR-008. ai and aveți both correct on 2026-08-19, cold, on an English-only prompt as ERR-023 demanded. This closes the slot that had been unknown since 2026-08-03. Clean twice on 2026-09-06 — avem asked bare and are inside a sentence, cold at 18 days.

no de below 20 — trei ani but douăzeci de ani; heavily tested in R. Correct 3 of 3 on first day including the negative case nouăsprezece ani. Clean both directions again on 2026-08-09 — nouăsprezece ani without de and treizeci de ani with it Clean both directions on 2026-08-19 — trei ani without de, cincizeci de ani with it. cincizeci de ani produced correctly inside a sentence on 2026-09-06, with de intact.

digit + spre + zece for teens ≈ UA один-на-дцять; digit + zeci for tens ≈ UA два-дцять; joined by și. Three contractions only: paisprezece, șaisprezece, șaizeci — patru contracts in the teens alone. Overgeneralized to șasisprezece and paizeci on 2026-08-03. șasizeci for șaizeci and douâzeci for douăzeci on 2026-08-09 — the three contractions now failing in the opposite direction from 2026-08-03, and â written for ă. On 2026-08-10 both contractions fired — paisprezece and șaisprezece — and douăzeci came back with its ă after douâzeci the day before. Only the ș of șaisprezece was lost. On 2026-08-15: *doizeci* for douăzeci — the tens take the feminine *două*, not *doi*; *paizeci* for patruzeci, real in speech but not in writing; *șapto* for șapte, a verbatim repeat of ERR-037. 100 was clean. 2 of 3 on 2026-08-19 — paisprezece and douăzeci both right, including the ă that failed on 2026-08-09 and 2026-08-15. șasizeci for șaizeci is a verbatim repeat of the 2026-08-09 error. Scored 3.5 of 8 cold on 2026-09-06. Seven of eight failures share one fact: the tens were written with -zece instead of -zeci. The teens ending -sprezece was right both times and the joining și was right twice. ERR-092.

cine people-only ≈ UA хто, ce ≈ що, cum ≈ як; front position, no inversion. Cum te numești not Ce — UA agrees, EN is the saboteur. Scrambled in isolation on 2026-08-09 — cine and cum swapped both ways, unde lost — then all three correct inside full questions ten minutes later. Retested bare the same afternoon in SES-011: 5 of 5 alone and 5 of 5 in frame. Same-day repair only — the 3-day interval is the first honest test of whether the bare entries held cine, cum and ce all clean on 2026-08-19 at the 7-day interval. This is the honest test the 2026-08-12 note was waiting for, and the bare entries held.

EN uses be, UA uses no verb at all (мені 30 років); RO uses a real transitive verb. Third instance of Romanian keeping a verb Ukrainian deletes. Clean 3 of 3 on 2026-08-03 Clean on 2026-08-19 — Am treizeci și unu de ani, whole and unprompted. Ea are cincizeci de ani produced whole and correct on 2026-09-06.

Latin balneum, the root behind Italian bagno. Two things at once in the plural: a → ă, and the -ie collapses into the -i. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off.

French bureau, matches EN bureau. Ends in a vowel, so the plural is -uri, not -e. Produced as o buro on 2026-08-10 — a feminine article on a neuter, and the word itself not yet formed. Still not formed on 2026-08-12 in the second drill — o buiro, after o bură earlier the same day. The bureau anchor has now failed twice in one day. Correct spelling picked out of three close variants on 2026-08-19, in the unscored probe that separated stored letters from written letters. All six items were right. A probe moves no tier and no interval; the reading is about the output path, not about birou. Produced as berou on 2026-08-19 — hours after this exact spelling was picked correctly out of three close variants that same morning. Fifth failed production of this word. Demoted. See the ERR entry.

From bucate, victuals, ← Latin buccata. The -ie → -ii plural is regular for every feminine in -ie. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Produced as bucăterea on 2026-09-07, cold at 18 days — stem right, -ie written -ea. Demoted 2 to 1.

EN decoy: this means computer, not a pocket device; o → oa in the plural. Taken as masculine on 2026-08-07 — doi calculatori. Corrected 2026-08-07, right on 2026-08-09 — the o to oa shift fired Clean package on 2026-08-19 — un calculator, calculatoare, the o to oa shift fired again.

EN false friend: a camera is un aparat foto; UA камера partially helps as chamber Clean EN→RO bare on 2026-08-15 as a LEX-FF probe — EN camera did not fire. Bare, not the tier-3 package, so no promotion. Tier-3 gate 2026-08-25: o cameră correct, plural produced as camerel. Retrieval clean, form wrong, so held at 3 rather than demoted — see ERR-079. Singular and plural both right on 2026-09-07, cold at 13 days. The article half was not asked, so the tier-3 gate stays open.

French canapé; UA канапа is the same borrowing. Stressed -ea class, plural -ele. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off.

maps directly to Latin casa; the stem a does NOT shift — the contrast that proves masă mese must be stored, not derived On 2026-08-19 o casă was right and the plural came out casa, the definite singular taught that morning. Held at 3 per the split-probe rule — retrieval succeeded and only the form failed. Retrieval and plural both clean on 2026-09-06 EN→RO. Held at 3 because the article half of the tier-3 gate was not asked.

false friend: UA час means time — time is timp. Written as chasi on 2026-08-07, EN spelling for the ce sound. Taken as masculine on 2026-08-09 — doi cheasi. Two errors in one word: consonant-final inanimate is neuter, and ch is k so cheasi reads kyasi. Second ORTH-CG miss on this row. Taken as masculine again on 2026-08-10, plural ceași. Second identical gender miss; the false-friend meaning is solid, the gender is not. Plural given as cași on 2026-08-10 — the -uri ending replaced by -i and the frozen diphthong ea dropped. ceas is the page's worked example of a diphthong that must NOT alternate, precisely because the singular already violates the position rule. Clean EN→RO bare on 2026-08-15 — the час false friend did not fire. Full package asked 2026-08-24: the word was retrieved, the gender came back masculine and the plural cesi followed. Held at tier 3, form failure only (ERR-074). Produced as chas on 2026-09-07, cold at 14 days — the meaning right and the spelling Ukrainian час in Latin letters. Romanian ce is the ch sound and ch is a hard k, so chas would read kas. Retrieval succeeded, so the row holds; this is an orthography finding.

Latin clavis; RO turns cl- into ch-, as in a chema from clamare. Taken as masculine on 2026-08-09 — un cheie doi chei. It ends in -e and is not alive, so gate 4 makes it feminine. Clean on 2026-08-10 — gate 4 routed it to feminine correctly. Asked for the singular on 2026-08-13 and answered un chei — the plural, correctly formed, with a masculine article. See the plural-for-singular pattern. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Recognition clean 2026-08-20 and promoted; the definite form failed the same session as cheiele. The word is held, the article rule is not — see G-0002. Produced as chaie on 2026-09-07, cold at 18 days — the e and a transposed, the lexeme right. Demoted 2 to 1.

Slavic, and RU is the closer match here — ковёр is the same word; UA standard is килим. Neuter in -or, so the plural is -oare and the o opens to oa before it. Correct spelling picked out of three close variants on 2026-08-19, in the unscored probe that separated stored letters from written letters. All six items were right. A probe moves no tier and no interval; the reading is about the output path, not about covor. Produced clean bare on 2026-08-19 and held rather than promoted: the spelling was on screen that morning in the discrimination probe.

French crayon, matches EN crayon; o → oa in the plural. Taken as masculine on 2026-08-09 — doi creioni. Consonant-final inanimate is neuter and the plural is creioane. Produced as creaon on 2026-08-10 with plural creone. The learner was asked and confirmed the vowel swap is a mistake, not a typo. Plural given as creione on 2026-08-10 — the ending -e was right and the stem was never revisited, so o did not open to oa. Second consecutive session on this word, and the failure moved: creone on 2026-08-09 lost the i as well. Produced as *creaon* on 2026-08-15, EN→RO bare — the medial *i* became *a*. Demoted on partial recall.

Latin cultellus, the root behind EN cutlery. The stem does not move. Correct spelling picked out of three close variants on 2026-08-19, in the unscored probe that separated stored letters from written letters. All six items were right. A probe moves no tier and no interval; the reading is about the output path, not about cuțit. Produced clean bare on 2026-08-19 and held rather than promoted: the spelling was on screen that morning in the discrimination probe. Clean EN→RO bare on 2026-08-25 — promoted to 3. Plural produced as cuțiți on 2026-09-07, cold at 13 days — the masculine -i on a neuter noun that takes -e. Retrieval succeeded, so this is a form finding.

Latin dentem, matches EN dentist. On the short masculine -e list. Missed 2026-08-09 as o dinte. Clean on 2026-08-10 — the only masculine -e noun produced entirely right. Produced as un dentă with plural denți on 2026-08-19 — article, t to ț alternation and -i ending all correct, and the stem built out of this row's own EN dentist hook. Demoted on retrieval. See the ERR entry. The definite dintele was correct on 2026-09-06, same-day post-key. Cold it had failed the same class minutes earlier.

Latin dormire, to sleep; matches EN dormitory. A neuter in -or takes -oare. Plural given as dormituri on 2026-08-10 — the ending was chosen wrong at step 2, not the stem at step 3; corrected to dormitoare in practice the same session, post-key. 2026-08-20 full package: bare form right, dormitoara for dormitorul and dormitori for dormitoare. The neuter's two borrowings inverted — feminine in the singular, masculine in the plural — which is the rule present and reversed, not absent.

Turkish farfuri, ultimately Greek. -ie → -ii, same class as bucătărie. Not retrievable at 72 hours on 2026-08-13, EN→RO bare — furfoare. One of only two of 32 that were genuinely gone. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Recognised correctly — the 2026-08-13 note calling it genuinely gone describes PRODUCTION only; the form→meaning link never broke.

Latin fenestra, matches EN defenestration; ea → e before the plural -e. Gender taken as neuter on 2026-08-07 despite the -ă ending. Corrected 2026-08-07, right on 2026-08-09 — the ea to e stem shift fired Correct spelling picked out of three close variants on 2026-08-19, in the unscored probe that separated stored letters from written letters. All six items were right. A probe moves no tier and no interval; the reading is about the output path, not about fereastră. Produced clean bare on 2026-08-19, and held rather than promoted: the same spelling had been on screen that morning in the discrimination probe, so the reading is same-day and post-key. Clean EN→RO bare on 2026-08-25, with its diacritic — promoted to 3. Singular right and the plural produced as ferăstre on 2026-09-07, cold at 13 days — the plural takes plain e.

Latin florem, matches EN floral. Feminine -e, the default for the gate. Clean on first contact 2026-08-09. Produced as floară on 2026-08-10 — the gender was called f correctly and the ending was then rewritten to match it. The clearest single instance of the ending being derived from the gender instead of stored. Plural flori produced correctly on 2026-08-10, oa flattening to o before -i, the day after floară. Post-key, so no tier move. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 produced as floară from English — the -e ending stored as -ă, same move as cartă in the same session. The definite floarea also failed, one day after it was produced correctly. Definite form asked cold 2026-08-24 and produced as flora, character for character the 2026-08-20 answer after the key had been published. The final -e was deleted and the oa diphthong reduced. Stored wrong rather than re-guessed (ERR-071). The definite came back as florea on 2026-09-06, cold — the ending right, the stem diphthong collapsed. floarea keeps the oa; only the plural flori loses it. ERR-089. Clean EN→RO on 2026-09-07 — counts as Clean RO→EN pass 1 of 2.

Latin frigidus, cold; matches EN frigid and refrigerator. The stem does not move. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. A second clean RO→EN pass exists and predates it: 12 of 12 on 2026-08-13, asked cold at a one-day interval in SES-019 part 1 and recorded only at that session's close-out, because the session stayed open across two days and closed after SES-018 had already run. The gate's two passes are therefore met; the promotion was deliberately not taken here because SES-018 owns the 2026-08-15 measurement and a row moves at most once a day. Promoted 1 to 2 on 2026-08-17: the two-passes gate was met and the move is arithmetic, deferred by SES-018 and SES-019 because a row moves at most once a day and routed here by SES-020's Next pointer. 2026-08-20 produced as friger — the middle syllable dropped out of a word whose meaning and shape are otherwise held. Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 18 days.

UA лампа maps directly. a → ă before the plural -i. 2026-08-20 full package: lampă and lampa right, plural given as lampe. The a→ă stem alternation before -i did not fire.

same root as a lucra to work; ends in -u not -ă, and was taken as feminine on 2026-08-07. Corrected 2026-08-07, right on 2026-08-09 — the fix held Clean bare on 2026-08-19. Retrieval clean EN→RO on 2026-09-06; plural produced as lucrul, the definite singular in a plural slot. Held at 3 — the form failed, not the word.

Latin mensa — the stem e resurfaces in the plural. Produced as mase on 2026-08-07, the stem shift did not fire. o masă two mese clean on 2026-08-09 — the a to e stem shift fired unprompted Clean package on 2026-08-19 — o masă, mese, the a to e shift unprompted for the second time.

Latin montem, matches EN mount and mountain. On the short masculine -e list — the ending says nothing, store the gender. Missed 2026-08-09 as o munte. Plural written munti on 2026-08-10, the ț without its comma, while cărți and dinți kept theirs in the same block. Article given as o on 2026-08-12 while the plural munți came out perfect — the word is retrieved and the gender is not attached to it. Ends in -e, which is gate 4 of the gender procedure, the gate that has never scored. Produced as muntă on 2026-08-19 — not a wrong article this time but a rewritten word, the -ă attached to match the feminine gender this row keeps attracting. Demoted. See the ERR entry. Clean RO→EN pass 1 of 2. Cold at 18 days, 2026-09-06. The definite muntea was wrong the same session — the feminine ending on a masculine noun. ERR-089.

Latin oculus, matches EN ocular; plural identical to singular, so un ochi and niște ochi both stand Clean bare on 2026-08-19. Retrieval clean EN→RO on 2026-09-06. The plural item leaked its own answer, so the form half is unmeasured. Held at 3.

Slavic root glend-, as in UA глядіти, to look at. d → z before the plural -i. Written oglidă on 2026-08-10 with the n present in the plural on the same line, and oglindă produced correctly minutes later — a slip, not a memory failure, and not tallied. Correct spelling picked out of three close variants on 2026-08-19, in the unscored probe that separated stored letters from written letters. All six items were right. A probe moves no tier and no interval; the reading is about the output path, not about oglindă. Produced clean bare on 2026-08-19 and held rather than promoted: the spelling was on screen that morning in the discrimination probe. Clean EN→RO bare on 2026-08-25 — promoted to 3. Singular and plural both right on 2026-09-07, cold at 13 days, with the d to z alternation fired. The article half was not asked, so the gate stays open.

Hungarian pohár. No Slavic or Romance hook — store it bare. The stem does not move. Clean on 2026-08-12 — un pahar, pahare, both parts right. The stem has no o to open, and it was correctly left alone, which is the input-vowel gate holding on the word that motivated it. Clean bare on 2026-08-19, cold — one of only two words in its block that had not been shown that morning. Plural produced as pahari on 2026-09-07, cold at 19 days — masculine -i for neuter -e, the same choice as cuțit in the same round.

no hook, memorize flat; -uri plural marks it neuter Produced as pât with plural pâturi on 2026-08-19 — the article and the -uri plural both right, and an â written into a word whose stem vowel is a plain a. Demoted: the letters of the word failed, not the rule. See the ERR entry. Clean EN→RO bare on 2026-08-25 — promoted to 3. Blank in both forms EN to RO on 2026-09-07, cold at 13 days — retrieval failed, not the form. Demoted 3 to 2.

Turkish perde. The stressed -ea class: plural -ele. The only shape in the ledger whose feminine singular does NOT end in -ă. Not retained on 2026-08-10 — un perdol on the check, then perdeă in practice: the right word with an -ă where the class ends in a bare stressed a. Correct spelling picked out of three close variants on 2026-08-19, in the unscored probe that separated stored letters from written letters. All six items were right. A probe moves no tier and no interval; the reading is about the output path, not about perdea. Produced clean bare on 2026-08-19 and held rather than promoted: the spelling was on screen that morning in the discrimination probe. On 2026-08-24 the definite form was asked with the bare noun printed in the prompt and came back pardeaua: the -aua ending correct, the stem vowel changed from e to a with the source visible (ERR-073). Demoted from tier 2 — the word was not copied, let alone recalled. Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 14 days.

masculine despite the -e ending; t → ț before the plural -i. Read as a plural on 2026-08-07. Split answer on 2026-08-09 — o perete with doi pereți. The plural knew the gender and the article did not. On the short masculine -e list. Produced as peret on 2026-08-10 — the final -e came off the word; gender m was right. Retested 2026-08-13, EN→RO bare: peret, the final -e dropped. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. The definite came back as peretea on 2026-09-06, cold — the feminine ending on a masculine noun. ERR-089.

Slavic per-, feather — the root of UA перина. The stem does not move. Retested 2026-08-13 at 72 hours, EN→RO bare: produced as perină — the Ukrainian anchor перина itself, with a Romanian ending. The hook delivered the source language. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Produced as pernea on 2026-09-07, cold at 18 days — the -ea class of perdea and canapea applied to a plain -ă noun learned in the same room batch. Demoted 2 to 1.

Slavic pod-, the root of UA під and підлога. Stressed -ea class, plural -ele. Clean bare on 2026-08-19, cold — one of only two words in its block that had not been shown that morning. Plural produced as podei on 2026-09-07, cold at 19 days.

Greek prosopon, face — a face-cloth. o → oa before the plural -e. All three parts correct on first test 2026-08-10. Not retrievable at 72 hours on 2026-08-13, EN→RO bare — propol. One of only two of 32 that were genuinely gone. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Recognised correctly — the 2026-08-13 note calling it genuinely gone describes PRODUCTION only; the form→meaning link never broke. Clean EN→RO bare on 2026-09-07, cold at 18 days. Promoted 2 to 3.

Latin panem, matches EN pantry and companion. Feminine -e. Taken as masculine 2026-08-09. Taken as masculine again on 2026-08-10; the word and the plural were both right. Given in a graded item on 2026-08-19 with only the article state scored: the definite form pâinea was built correctly, feminine -e to -ea. Not a retrieval measurement, so no tier or interval move. Produced clean bare on 2026-08-19 and held rather than promoted: the word had appeared in that morning's graded item. Bare word clean EN→RO on 2026-08-25 and promoted; the volunteered article was un, so the feminine gender is not held. The tier-3 gate asks it next. The definite pâinea was correct on 2026-09-06, cold — as it was on 2026-08-24, the one right answer in a set of ten. Correct EN to RO bare on 2026-09-07; the full package was not asked, so the tier-3 gate is untouched.

Latin scamnum; RU скамейка is the same family. Taken as masculine on 2026-08-07 — doi scani for două scaune. Corrected 2026-08-07, right on 2026-08-09 — the fix held Asked for the singular on 2026-08-13 and answered scaune — the plural, correctly formed. See the plural-for-singular pattern. Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. The definite scaunul was correct on 2026-09-06, same-day post-key.

Latin solem, matches EN solar. On the short masculine -e list; inanimate but masculine. 0 of 3 on 2026-08-10 — soar for soare, taken as neuter, and the singular offered as the plural. The worst-scoring noun of the session. On 2026-08-12, second drill, the word came back correct as soare and the article did not follow — o soare. Word and gender are stored separately here and only the word has been rebuilt. It is masculine in -e, which is the open gate 4 of the gender procedure. Clean bare on 2026-08-19 — soare intact, three sessions after it was the worst-scoring noun of 2026-08-10. Singular produced as soară on 2026-09-07, cold at 19 days — the -e ending written -ă, the second sighting that session after cartă. The plural sori was right.

no hook, memorize flat; -i plural, not -e Clean RO→EN recognition pass 1 of 2 on 2026-08-15, cold after two days off. Blank EN to RO cold at 18 days on 2026-09-07. Demoted 2 to 1.

not a vowel but a softening on the consonant, like UA день and кінь. After c or g it merges into the ce and ge sounds instead, so medici is two syllables. Italian Medici with three is the trap. Clean on 2026-08-09 — medici, pereți, dinți, uși, flori all produced correctly medici produced and correctly counted as two syllables on 2026-08-19.

the ending reaches back: t→ț, d→z, s→ș, and the vowels a→ă, a→e, ea→e, o→oa. UA has the same reflex in рука руці and стіл стола. Recognised correctly but produced 0 of 4 on 2026-08-07, then 5 of 5 on 2026-08-09 after the per-word cheat sheet built the same day — mese, ferestre, calculatoare, pereți, uși. It held only on rehearsed nouns: străzi came out as strazi and cărți as cârți the same day. On 2026-08-10 the alternation fired only in cărți. țare, numare, creone, librare, ucraineani, ucraineance and strazi all left the singular stem untouched — the plural is being assembled by swapping an ending onto an unchanged stem. Taught as an ordered three-step routine on 2026-08-10 (gender → ending → the ending rewrites the stem) after the ledger showed it failing on every unrehearsed noun for three sessions. Retested the same session on five nouns never seen, answers printed nowhere: 4 of 5. Two needed a change and got one, two needed none and got none, and the single miss was saete for sate — an invented change, the first OVER-application on record. Promoted to 2 on that evidence; the misses before it were all under-application. Sampled again on 2026-08-12 in the second drill, on ten fresh neuters with the -e ending supplied so that only the stem was scored: 9 of 10. Four of five o-stems opened to oa and five of five non-o stems were correctly left alone, so zero over-application. Held at 2 rather than promoted — the look fell two days into a three-day interval and came hours after the same gate was corrected, so it measures repair and not retention. The single miss was piciore for picioare, declared a real error by the learner, where the o sits inside cio and the plural needs three vowel letters in a row. 3 of 3 alternations on 2026-08-15 with the singular supplied: telefon→telefoane (o→oa), stradă→străzi (a→ă, d→z), carte→cărți (t→ț). The one error on that set was a stem vowel written â for ă, not an alternation. On 2026-08-19 telefoane was right and strazi left the a to ă shift unfired while the d to z fired. Demoted on the partial. On 2026-08-24 obraz was correctly called masculine and still gave obrazi for obraji — the z to j alternation did not fire (ERR-074). florea for floarea on 2026-09-06 — the plural stem alternation applied inside the definite singular, where the diphthong must stay.

un covers m and n, o is f, niște is every plural. German ein und eine carries gender the same way; UA and RU have no articles at all. Refused after a fi plus a profession and after a negated verb — N-am telefon. Volunteered unprompted on all 12 nouns of one block on 2026-08-10 and correct on every one, including un for the neuters. Promoted to 2 on that evidence. 3 of 3 on 2026-08-15: un telefon, o stradă, un inginer. Includes a neuter given *un* correctly while the same learner named a different neuter 'm' — consistent, since m and n share the singular article. niște studenți and o oglindă both clean on 2026-08-19. The learner volunteered ună beside o, which is not a form.

neuter is NOT UA середній рід — Romanian has only two agreement patterns, so a neuter noun is masculine in the singular and feminine in the plural. Test by counting: doi versus două. Assignment failed 5 of 7 on 2026-08-07. On 2026-08-09 the heuristic was rewritten as an ordered four-gate procedure — -ă means feminine, alive means masculine, consonant or -u and not alive means neuter, -e decides nothing. Gates 1 to 3 scored 6 of 7 and gate 4 scored 2 of 5, so the -e gate is the open wound. Gate exceptions to name when they arrive: tată is masculine despite -ă (U04 — person-sex outranks the ending), an and U08's leu are masculine despite gate 3. On 2026-08-10 gates 1 to 3 scored 12 of 15 and gate 4, the -e gate, scored 6 of 9 against 2 of 5 on 2026-08-09. oraș and ceas were taken masculine for the second consecutive session. On 2026-08-12, second drill, the article split cleanly by whether the word was retrieved: un was written six times and was correct six times, o was written six times and was correct only on the three real feminines. The article is not being chosen — un rides along with a retrieved word and o fills the space when the word is missing. See ERR-033. 1 of 3 on 2026-08-15: *scaun* called m (neuter), *perete* called n (masculine), *floare* correctly f. The session withheld all plurals by design, which is the only place the neuter is visible — half this result is instrument. 1 of 3 on 2026-08-19 — perete m right, nume called f and oraș called m. Gate 4 again, and this time it also rewrote a word's final letter elsewhere in the session. On 2026-08-24 three tier-3 nouns were all declared masculine and two of them are neuter. The class is being read off the singular article, where neuter and masculine are identical (ERR-074). Re-opens a zone measured 90% on 2026-08-12.

The three steps are not independent: the ending is what rewrites the stem, so the stem cannot be touched until the ending is chosen, and gender never touches the stem at all. Added 2026-08-10 because every plural error on record is this ordering breaking in one of three ways — creione, the ending taken and the stem left alone; țăre, the stem rewritten and an ending from the wrong shelf attached after; floară, the stem rewritten from the GENDER, which has no power over it. The rule is one sequence, not three facts, and it is the reason the alternation table alone did not land on 2026-08-09. Asked defectively on 2026-08-15 — the plural and the three-step order were put on one line, and only the plural came back. Second occurrence of the one-line multi-part defect (after SES-014), and the third in its family counting SES-016's parallel-column variant. The plurals themselves were 2 of 3. Named in order on 2026-08-19 — gender, ending, stem — asked on its own line, which is what the 2026-08-15 defect had prevented. Scored 5 of 8 on 2026-09-06, cold — the feminine -ă class resolved as always -e, so țări and luni came back as țare and lune. Demoted from 3. ERR-092.

-uri is neuter at this level — vremuri and treburi are feminine and arrive later (hedged 2026-08-10); -e is never masculine; masculine is -i almost without exception. German is the anchor — the plural is memorised with the noun and unpredictable by design; UA plurals are regular from the stem and give no hook here. 12 of 14 on drilled nouns on 2026-08-09 but 7 of 12 on undrilled ones — the plural is stored per word, so a noun whose plural was never rehearsed does not have one. On 2026-08-10 plurals scored 4 of 13 — the weakest zone of the session by a wide margin, and far below the 75% the gender assignment it depends on reached. 2 of 3 on fresh carrier nouns on 2026-08-19 — pantofi and tablouri right, pădurele for păduri. Demoted. On 2026-08-24 all three plurals in the full-package round were wrong; two follow faithfully from a wrong gender call and are not independent evidence (ERR-074). Scored 5 of 8 on 2026-09-06 cold. The four feminine -e plurals were all clean; the two failures took -e where -i belongs.

EN forces an article and interferes; UA having none is right here. ART-NDEF zone, full article system U03 Clean on 2026-08-19 — inginer with no article after a fi.

Sits visibly on bun, good — one of the same session's adjectives. Both plurals are the same word, bunici. Produced correctly EN to RO on 2026-08-12, first exposure. At 72 hours produced *binice* / *binică* — the bun- stem stored as bin-. Same substitution as V-0104's *bine*, three sightings out of three across the two rows.

No hook — stored cold. The plural moves twice, the diphthong ia becoming ie and t becoming ț. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06. Asked EN→RO in a sentence the same session it came back as bărbatul — man, not boy. Recognition works, production reaches for the neighbour. ERR-094. Produced as cupil EN to RO on 2026-09-07 — copil, child, with u for o. Came back as bărbatul on 2026-09-06. Two days, two different family members, never băiat.

No hook in any language the learner holds — stored cold, not given an invented one. The plural carries two i and the definite plural three, copiii, which is the same build as copacii. Dexonline also lists a rare copili; the standard plural is copii. Produced as cupil on 2026-08-17, the stem vowel o written u. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06. The definite copilul failed the same session, and the stem came out as Cupilul in a sentence.

EN family. UA фамілія is a false friend — it means surname, and the UA word for family is сім'я. The -ie to -ii plural is regular for every feminine in -ie, the class of bucătărie and farfurie. Produced correctly EN to RO on 2026-08-12, first exposure. Clean at 72 hours, EN→RO bare — the tier-2 question one rung early. The UA фамілія false friend did not fire. Feminine in -ie, so the definite article is -ia and not -ea — produced as familiea on 2026-08-17 at first attempt, then correct as familia in four later rounds of the same loop. Post-key, so no tier moved.

Already met as a sound pair on the U01 vowels page, fată against fata. The plural changes the stem vowel a to e. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06, and the definite fata was correct the same session.

Latin filia, the sister word to fiu. The plural puts c before e so it goes soft, fiice sounding fee-EE-che. Given as soră on 2026-08-17 when asked from English, with soră printed in a neighbouring prompt — the item was withdrawn as primed. Blank on the clean recognition read the same session. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Clean RO→EN pass 1 of 2 on 2026-08-25.

Latin filius, matching EN filial. Ends in -u, so the definite article is the bare -l. Produced correctly as fiul vostru on 2026-08-17. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Asked RO→EN cold at 20 days on 2026-09-06 and answered brother — frate. The family group is stored as a cluster, not as separate words. ERR-094. Clean EN→RO on 2026-09-07 at a 1-day interval, one day after coming back as brother. EN→RO is a rung above the gate's question, so this counts as Clean RO→EN pass 1 of 2.

EN fraternal and fraternity, Latin frater. Ends in -e and is alive, so gate 2 makes it masculine. The plural -i rewrites t to ț. At 72 hours produced *fratere* — stem correct, *-re* added. Practised heavily on 2026-08-17 inside the possessive loop, post-key throughout, so no tier moved. Produced as fratel nostru — the definite ending -l where a masculine in -e takes -le. Third sighting of that shape after muntel and câinel on 2026-08-15; see the ERR entry. Correct as fratele by the end of the same loop. The definite came back as fratea on 2026-09-06, cold — the feminine ending on a masculine noun. Failed the same way on 2026-08-24. ERR-089. Clean EN→RO bare on 2026-09-07 but at a 1-day interval against a 3-day one, so the row holds at 2 rather than promoting.

UA мама maps directly. The stem does not move before the plural -e. Clean at 72 hours, EN→RO bare — the tier-2 question one rung early.

No hook — stored cold. Describe someone with the singular, părul; the plural counts individual hairs. Produced as par on 2026-08-17, the missing ă making it a different word — par is a pole and also the first person of a părea, to seem. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Blank RO→EN on 2026-08-25, eight days after intake — first miss on this row and consistent with having no hook.

EN parent, Latin parens. Ends in -e and is alive, so gate 2 routes it masculine, the same road as frate. Mostly used in the plural. Produced correctly as părintele său on 2026-08-17 — the hardest single item of the session, being a masculine -e noun plus a family 1 possessive. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Clean RO→EN pass 1 of 2 on 2026-08-25. The definite părintele was correct on 2026-09-06, same-day post-key. In a sentence the same session the plural came out as Parenții — the stem drifting toward English parent, and the ă lost.

EN sorority, Latin soror. The plural is irregular twice over — o becomes u and a syllable is added. Recognised correctly RO to EN on 2026-08-12, first exposure. At 72 hours produced *sorore* — stem correct, ending Latinate. Sits near the plural *surori*; whether it was a reach for the plural is UNRESOLVED — the question was posed inside a batch of typo questions and got a typo verdict. Practised on 2026-08-17 inside the possessive loop, post-key, so no tier moved. Produced as sara noastre — the stem vowel o written a, a fifth shape for this row after sorore. Correct as sora in the four later rounds of the same loop. Produced as fia EN to RO on 2026-09-07, cold at 18 days — not a Romanian word, and shaped like a member of the fiu/fiică group. The family cluster reaching again. Demoted 2 to 1.

Latin socius, a companion, the root of EN social and associate. The plural soți also names a married couple. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Clean RO→EN pass 1 of 2 on 2026-08-25. Produced as soți EN to RO on 2026-09-07 — the plural in a singular slot, in the same round where soție came back on this row.

Built on soț with the feminine -ie, the same one-root-two-people shape as bunic and bunică. Produced as sață on 2026-08-17 from English — an invention, not a near miss. Recognised correctly RO to EN minutes later, but after the key had been shown, so that pass does not count. Gender and plural verified dexonline 2026-08-15 with scripts/dex.mjs. Asked RO→EN cold at 20 days on 2026-09-06 and answered husband — that is soț. The -ie that makes it feminine was not read. The definite soția was correct the same session. ERR-094. Asked EN to RO on 2026-09-07 and produced as sota — the soț root, a wrong ending and the ț absent. The husband and wife rows swapped into each other on both days.

UA тато. Masculine although it ends in -ă — for a person, sex outranks the ending, so gate 1 of the gender procedure does not run. This is the exception G-0017 has been carrying since 2026-08-09. Asked for article and plural on 2026-08-12 and answered dad — the meaning correct and recorded, the forms not attempted. Clean at 72 hours, EN→RO bare — the tier-2 question one rung early.

EN bonus and bounty, Latin bonus. Held inside bună ziua since 2026-07-30 and isolated as an adjective on 2026-08-12. The default four-form class. Produced correctly as bună the same day. At 72 hours on 2026-08-15 produced *bine* — the adverb, V-0025, which sits at tier 4 and holds the same meaning slot. The failure is competition from an entrenched neighbour, not decay: *bună* inside bună ziua has never been wrong. Clean RO→EN pass 1 of 2 on 2026-08-25.

No hook — stored cold. Four regular forms and the stem never moves. Use it of people; a thing that is old is vechi. Blank on the clean unprimed recognition read of 2026-08-17. Adjective forms verified dexonline 2026-08-15. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06.

Latin forma, well-formed. The o opens to oa before -ă and -e — the same conditioning as the noun plurals, not a new rule. Produced as frumaosă on 2026-08-12: oa written ao, and a feminine singular ending on a neuter singular target that takes the bare form. At 72 hours produced *fearoșe* — declared a real error, not a slip. The only word of 13 whose BEGINNING was also wrong, which is the first exception to the all-beginnings-correct observation of 2026-08-13.

Two forms — the masculine singular ends in -e, so gender does nothing and there is no such word as mară. Also a feminine noun, marea, the sea. Produced correctly on 2026-08-12 inside Un oraș mare. At 72 hours produced *maron* — reached for maro (V-0109). The two collapsed into one entry; both taught 2026-08-12, both begin mar-. The separator is that maro is invariable and mare is not.

EN maroon, French marron, a chestnut. Borrowed, so it never takes an ending in any slot. Produced as maroă on 2026-08-12 — an ending attached where the language has none. See ERR-034. At 72 hours produced *meron* — *e* for *a* plus an *-n* imported from EN maroon / FR marron / RU марон. Confirmed invariable on dexonline 2026-08-15: maro, maro, maro, maro. Clean EN→RO bare on 2026-09-07, cold at 18 days. Promoted 2 to 3.

EN micro. Three forms — a short c-stem, so both plurals merge into mici. The borrowed -ic suffix does not merge: simpatic gives simpatice. At 72 hours produced *mice* — stem correct, *-e* added.

No hook — stored cold. The plural runs a rule already held: d becomes z before -i, as in verde to verzi. Recognised correctly RO to EN on 2026-08-17, clean and unprimed. Adjective forms verified dexonline 2026-08-15. Clean RO→EN pass 1 of 2 on 2026-08-25.

UA симпатичний maps directly. Not EN sympathetic, which is înțelegător — LEX-FF. The borrowed -ic suffix keeps the two plurals apart, simpatici against simpatice, which is the exception the adjective agreement page named. Adjective forms verified dexonline 2026-08-15. Asked RO→EN cold at 20 days on 2026-09-06 and answered cute. UA симпатичний is about looks; Romanian simpatic is about character. A false friend that fired. ERR-094. Gone in both directions: answered cute RO to EN on 2026-09-06 and buna EN to RO on 2026-09-07. The English nice pulled bun across.

The one irregular of the six: the plural stem is tiner-, not tânăr-, and both vowels move. Store the plural as its own fact. Blank from English on 2026-08-17; the later correct reading was primed and does not count. Adjective forms verified dexonline 2026-08-15. Clean RO→EN pass 1 of 2 on 2026-08-25.

EN verdant. Two forms. The plural -i rewrites d to z, as in oglindă to oglinzi. Left unchanged where a plural was wanted on 2026-08-12. Clean at 72 hours, EN→RO bare — the tier-2 question one rung early.

Latin altus, matching EN altitude. The plain four-form class, like bun. Blank when asked from English on 2026-08-17, correct RO to EN the same session but after the key. Adjective forms verified dexonline 2026-08-15. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06. Asked EN→RO in a sentence the same session it came back blank — the direction is not symmetric on this row.

Four form-classes, counted by how many distinct forms an adjective has: 4 (bun), 3 (mic, nou), 2 (mare, verde), 1 (maro). The endings are fixed and carry no choice — -ă feminine singular, -i masculine plural, -e feminine plural — and there is no -uri anywhere on an adjective, so the -e versus -i coin flip that breaks the noun plural does not exist here. Neuter has no form of its own: masculine cell in the singular, feminine cell in the plural. Taught 2026-08-12. Measured the same day on five retest items with the noun supplied so only the adjective was scored: vechi left bare and negre built from negru both correct, plină for pline, alb for albe, receă for rece. Three feminine-plural items produced three different answers, so the rule is not being run and an -ă fills the slot when it does not fire. See ERR-034. 1 of 3 on 2026-08-19 with both the noun and the adjective supplied — buni right, albele for albe and marele for mari. The two misses are definite endings, not agreement endings. Demoted. Asked 2026-08-25 as casă plus mare in the plural and came back case mare — the noun pluralised, the adjective did not follow. See ERR-081.

UA, RU and EN all put the adjective in front, so this is the position error to expect. Taught 2026-08-12 and correct in all eleven phrases produced that day, graded check and retest together. The position landed immediately and the endings did not — worth separating from G-0023 because the two fail independently. Correct again on 2026-08-19 — the learner left the slot after the noun even while declining to guess the adjective itself.

First contact 2026-08-17; the genitive use is deferred to U12. It appears only in the gap between an indefinite noun and the possessive, and it carries the definite endings -l -a -i -le standing alone. The article and the possessive after it always sit in the same cell. First attempt collapsed to Ukrainian word order; by the last round both halves moved together in 6 of 6. 2026-08-20 the linker a was selected correctly for a feminine singular noun; the possessive after it did not agree. CONTAMINATED: table scanned minutes before, held. First clean reading on 2026-08-24, the possessive table unopened: 2 of 3 parts. The linker and the possessive agree independently and only one fired per phrase — a noi has the right linker and the wrong possessive, ai mele the reverse (ERR-075).

Taught 2026-08-17 alongside the possessive, whose paradigm carries both. Every vowel-order error on record is a rising pair written a-first, and the falling pairs have never been mis-ordered. oa exists only under stress, which is the same fact as the o to oa alternation already running in nouns and adjectives. Naming the rule immediately produced over-correction — oa pushed into noastru and noaștrae, cells that have a plain o. 2026-08-20 both diphthong cells correct — noastră with oa and nostru with plain o, which is exactly the over-correction pair named when this row was taught. CONTAMINATED: table scanned minutes before, held. On 2026-08-24 floare gave flora: the oa diphthong reduced to o while the definite ending was also wrong (ERR-071). The rising diphthong is not surviving suffixation.

Taught 2026-08-17. Two families — meu tău său take -u -a -i -ale, nostru vostru take -u -ă -i -e with o opening to oa in both feminines only. lor never changes. The single cell where the families disagree is the feminine plural, mele against noastre, and that is where the errors landed. All six paradigms verified against dexonline paradigm tables 2026-08-15. 2026-08-20 noastră and nostru both correct in their own cells — the o/oa over-correction this row was demoted for did not recur. mea failed after the linker. CONTAMINATED: table scanned minutes before, so held rather than promoted. On 2026-08-24 the subject pronoun noi was supplied in a possessive slot needing noastra, while mele was produced correctly in another item of the same set (ERR-075).

Taught 2026-08-17. The noun takes its article first and the possessive follows it, agreeing with what is owned and never with the owner. The learner dropped the article step in 4 of 4 first attempts, held it in 6 of 6 after one isolating round, and lost it again when the possessive got harder — the step falls out of the sequence rather than being unknown. Ukrainian agrees the same way but puts the possessive first; the position rule is the one already taught for adjectives. 2026-08-20 prieteni meu for prietenul meu — the plural produced where the definite singular belonged, so the possessive attached to an unarticulated noun. CONTAMINATED: the learner volunteered that they scanned the possessive table minutes before answering.

UA банк is masculine and this is feminine, so the definite article is -a and not -ul. Also means bench. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours.

Latin basilica, the same word as EN basilica with the l worn down to r. Read as market on 2026-08-17 — which is a real translation of piață, answered correctly four lines earlier in the same set. A neighbour from the same intake arrived instead of the target, not a missing word. Clean RO→EN pass 1 of 2 on 2026-08-25. Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 13 days.

UA центр. Romanian adds -u, and that u is exactly what makes the definite article just -l rather than -ul. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06. Also stayed bare after în in a sentence the same session, which is correct.

No hook anywhere — not Latin, not Slavic, origin disputed. Stored cold. The only masculine in the U05 city set, so the only definite plural in -ii. Blank on 2026-08-17 at 48 hours — the second of the two words this set declared hook-less, and both of those died. Clean RO→EN pass 1 of 2 on 2026-08-25. Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 13 days.

EN pharmacy, UA фармацевт. The everyday UA word is аптека, so the hook is the specialist and not the shop. Read correctly on 2026-08-17 but flagged hesitant by the learner, so it does not count as a clean pass. The definite farmacia was correct on 2026-09-06, cold — this exact item failed on 2026-08-24. The feminine -ie branch has held 13 days.

French gare. No hook in any language the learner holds — вокзал is unrelated — and the page declared it stored cold rather than inventing one. Not answered at all on 2026-08-17, 48 hours after the last reading — one of the two words this set declared hook-less, and both of those died. Clean EN→RO bare on 2026-09-07, cold at 18 days. Promoted 2 to 3 — the next asking is the full package.

UA готель with no soft sign in Romanian. Plural produced as hotele on 2026-08-15 — the plural is -uri, and the definite plural built off it was correct, so the article rule ran on a wrong input. Plural asked alone on 2026-08-17 and produced as hoteli — the masculine -i ending on a neuter noun. This is worse than the 2026-08-15 hotele, which had the right shelf and the wrong ending. RO to EN was not asked because the word is identical in English, so there is no retrieval evidence either way. Hotelul produced correctly inside a sentence on 2026-09-06, subject position. The article round that day was post-key, so this is not a cold reading.

UA магазин, same meaning. False friend in English — a magazine you read is revistă. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours.

UA музей. The ending is the whole difference — UA -ей against Romanian -eu. Produced as muzea on 2026-08-15, LEX-END at first contact, 48 hours after it was taught. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours.

UA парк — the same word and the ending matches too, which is rare in this set. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours.

Latin platea, the same root as EN place and IT piazza. No Ukrainian form. The plural shifts ia to ie. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours. Given in a graded item on 2026-08-19 with only the article state scored: correctly left bare after the preposition la. Not a retrieval measurement. Produced as piața on 2026-09-07 when the bare form was asked — either the definite form in a bare slot or the final ă dropped. Demoted 2 to 1.

False friend — UA під and RU под mean under. Also means attic. Produced as pot on 2026-08-15: the RU devoiced pronunciation of под, so the anchor handed over the wrong language. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours. Clean EN→RO bare on 2026-09-07, cold at 18 days. Promoted 2 to 3.

UA ресторан. Romanian keeps the French -t that Ukrainian dropped, and it is pronounced. Plural asked alone on 2026-08-17 and produced as restauranturi — the right neuter shelf, the wrong one of its two endings. This is the -e against -uri coin flip that no rule reaches. RO to EN was not asked; the word is identical in English.

UA шпиталь, the same Latin root. Romanian drops the soft sign and the EN ho-. Clean RO to EN pass 1 of 2 on 2026-08-17, cold at 48 hours.

Taught 2026-08-17 with the group I present. eu ajut (-a class, bare). Latin adiutare — EN adjutant is the same word. The tu form softens: ajuți. Conjugation class verified dexonline 2026-08-17. Conjugated cold on 2026-08-19, 48 hours after the teaching: tu form ajuți correct, including the t to ț softening.

Taught 2026-08-17 with the group I present. eu ascult (-a class, bare). Latin auscultare, EN auscultate. Takes its object directly — ascult muzica. The tu form softens: asculți. Conjugation class verified dexonline 2026-08-17. Answered to search on 2026-08-25, which is a căuta. See ERR-078 — the two collided, and the meaning that was retrieved landed on the wrong verb. Blank EN to RO on 2026-09-07, cold at 13 days.

Taught 2026-08-17 with the group I present. eu aștept (-a class, bare). Takes its object directly, with no preposition — aștept autobuzul. EN says wait for and UA чекати на; Romanian says neither. Stem opens before -ă: așteaptă. Conjugation class verified dexonline 2026-08-17. Produced EN to RO on 2026-08-19 as a așepta, with the t of the cluster missing. The learner declared it genuine, not a slip. The lexeme was retrieved correctly; the letter sequence was not. Asked RO to EN cold at 19 days on 2026-09-07 and answered to answer — a confident wrong lexeme, not a blank. The Latin exspectare hook was never given.

Taught 2026-08-17 with the group I present. eu cumpăr (-a class, bare). No hook — stored cold. Stem vowel moves: cumperi at tu, because ă becomes e before the -i ending. It does not open at el, because that ă is unstressed. Conjugation class verified dexonline 2026-08-17. On 2026-08-24 the tu form came back cumpari again, character for character ERR-070, with a cumpara printed in the prompt. Second failure of the stem-vowel rule on the verb it was taught with (ERR-073). Clean EN→RO on 2026-09-07, cold at 14 days — counts as Clean RO→EN pass 1 of 2.

Taught 2026-08-17 with the group I present. eu caut (-a class, bare). No hook — stored cold. Takes its object directly, with no for. Stem vowel moves: caut under stress, căutăm and căutați when the stress leaves it. Conjugation class verified dexonline 2026-08-17. Blank RO→EN on 2026-08-25 while its meaning surfaced one line later on a asculta. Not decay — a collision between two verbs that entered on the same day. Blank EN to RO on 2026-09-07, cold at 13 days.

Taught 2026-08-17 with the group I present. eu dansez (-a class, -ez- infix). EN dance, FR danser. UA танцювати is a different root. Takes -ez-, predicted by the international-word test. Conjugation class verified dexonline 2026-08-17. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06. Production is behind recognition: dansăm lost its ă and the third person came out as dansă, the -ez- infix omitted. ERR-093.

Taught 2026-08-17 with the group I present. eu desenez (-a class, -ez- infix). FR dessiner, EN design. Takes -ez-, predicted by the international-word test. Conjugation class verified dexonline 2026-08-17. Produced EN to RO on 2026-08-19 as a denesa — s and n in each other's places. Declared genuine. The lexeme was retrieved correctly; the letter sequence was not.

Taught 2026-08-17 with the group I present. eu intru (-a class, bare). EN enter, the same Latin root. The only verb in this intake whose eu form adds -u, because the stem ends in a consonant plus r. Conjugation class verified dexonline 2026-08-17. Conjugated cold on 2026-08-19 and produced as intrez — the -ez- infix applied to a bare verb, and the -u of the eu form missed. This row asks two stored facts at once and both failed; it is the hardest cell of the intake.

Taught 2026-08-17 with the group I present. eu plec (-a class, bare). No hook in any held language — stored cold. Stem opens before -ă: pleacă. Conjugation class verified dexonline 2026-08-17. Asked RO→EN cold at 20 days on 2026-09-06 and answered to search — a căuta. The same wrong English this learner gave a asculta on 2026-08-25. A stored default answer, not a fresh guess. ERR-090. Answered to search RO to EN on 2026-09-06 and blank EN to RO on 2026-09-07. Two directions, two days, nothing correct.

Taught 2026-08-17 with the group I present. eu semnez (-a class, -ez- infix). Latin signum, EN sign. The Writing section asks for signed forms. Takes -ez-, predicted by the international-word test. Conjugation class verified dexonline 2026-08-17. Conjugated cold on 2026-08-19: noi form semnăm correct, so the infix-drop rule that broke on vorbem in SES-007 held here on first use. Asked RO to EN cold at 19 days on 2026-09-07 and answered to mean — that is a înseamna, one prefix away, and ce înseamnă is already a ledger row. The disambiguator is the în- prefix.

Taught 2026-08-17 with the group I present. eu telefonez (-a class, -ez- infix). UA телефонувати — the same word and the same job. Takes -ez-, predicted by the international-word test. Conjugation class verified dexonline 2026-08-17. Conjugated cold on 2026-08-19 and produced as telefoneză. The infix and the person ending are both right; the missing piece is the opening e to ea before -ă, which is the rule already held for frumos to frumoasă. The rule has not transferred from adjectives to verbs.

Taught 2026-08-17 with the group I present. eu uit (-a class, bare). With a reflexive pronoun it becomes a se uita, to watch — a different verb in the same body. Reflexives are U06. Conjugation class verified dexonline 2026-08-17. Produced EN to RO on 2026-08-19 as a ulia. Declared genuine. The lexeme was retrieved correctly; the letter sequence was not.

Taught 2026-08-17 with the group I present. eu vizitez (-a class, -ez- infix). EN visit; UA has the noun візит but not this verb. Takes -ez-, and the international-word test predicts it. Conjugation class verified dexonline 2026-08-17. Asked RO→EN cold at 20 days on 2026-09-06 and answered to enter — a intra. The third person vizitează also failed the same session. ERR-094. Clean EN→RO on 2026-09-07 at a 1-day interval, one day after coming back as to enter. Counts as Clean RO→EN pass 1 of 2.

Taught 2026-08-17 with the group I present. eu întreb (-a class, bare). Latin interrogare, the root of EN interrogate. No Ukrainian form. The stem opens before -ă: întreabă. Conjugation class verified dexonline 2026-08-17. Clean RO→EN pass 1 of 2. Cold at 20 days, 2026-09-06. The third-person întreabă was asked the same session but withdrawn — the learner misread ea as eu and said so.

Taught 2026-08-25. The same stress mechanism as G-0048 with a wider gap: the singular ia- shares nothing with the infinitive. tu iei is the a to e fronting in front of the -i ending, the same shift that turns spăl into speli, running on a diphthong instead of a single vowel. Full present verified dexonline 2026-08-25. Not tested in production on 2026-09-06 — the infinitive was asked and came back clean; no inflected form was. luați produced correctly on 2026-09-15, cold alone and again in a sentence — the lu- stem at voi held at 9 days.

high-alert zone ART-DEF — no UA or RU analog at all, so German der-die-das is the only anchor for the concept. Taught 2026-08-15 (SES-019), promoted from the tier-0 seed by the session that delivered it. The singular ending is picked by the noun's last letter: consonant takes -ul, -u takes -l, masculine -e takes -le, feminine -ă takes -a, feminine -e takes -ea, -ie takes -ia, stressed -ea takes -ua. Masculine -e is the weak cell — produced as muntel and then câinel on 2026-08-15, twice in one session and the second time after the correct form had been published, so it is a procedure that is not running rather than a fact that is missing. 2 of 3 on 2026-08-19 — telefonul and floarea both right, lucrua for lucrul. The masculine -e cell was withheld because this session had already published that ending in an earlier marking. 2026-08-20 the feminine -e branch measured 0.5 of 6 on a leak-checked set, 24 hours after the same branch read 3 of 3 cold and was logged as the strongest result of SES-023. floarea itself was correct on 2026-08-19 and came back as flora. Wrong answers were flora, carta, cheiele, baiela, bucătăriea, frafuriele — three of them plural-shaped, one (bucătăriea) the near-miss where -a arrives and -e fails to come off. This is the first next-day measurement of this cell and it did not survive. Asked as ten cold production items on 2026-08-24 and every one treated the final -e backwards: deleted where it must stay, kept where it must go (ERR-071). Reframed mid-session from which ending to does the word lose a letter — four of the six endings are pure addition, only -a and -ie replace anything — and the retest ran 7 of 7 cold on nouns never asked before. Same sitting, so no Measured token; the cold retest is booked. Cold retest on 2026-09-06 at 13 days scored 7.5 of 12, against 0 of 10 on 2026-08-24. It splits by gender with nothing left over: 7 of 7 feminine correct, 0 of 4 masculine. Every masculine noun took the feminine -a. ERR-089.

Only two endings — masculine adds -i and gives a double ii, everything else adds -le whatever the plural ends in. Split from G-0002 because the two halves behaved differently on the day they were taught: the plural read 2 of 4 on the graded check with the masculine and neuter branches swapped, then 4 of 4 on a same-sitting retest with new nouns, while the singular -e cell failed twice. Same-sitting, so the retest carries no retention claim. frații and medicii both correct on 2026-08-19, masculine branch only; the -le branch was withheld because this session had already published that ending in an earlier marking. Părinții produced with the correct -ii on 2026-09-06 inside a sentence, cold at 18 days.

Taught 2026-08-17. Three ordered questions. After a preposition the article comes off unless something is added to the noun or the preposition is cu meaning by. A whole class, a title before a name and a noun used to say when all take it where Ukrainian marks nothing. Everything else behaves like English. Where Ukrainian uses a case ending Romanian uses preposition plus article. Stated from descriptive grammar and unverified in this workspace's sense. Checked cold on 2026-08-19, 48 hours after teaching: 3 of 3, including the cu exception and a feminine -e to -ea definite built from scratch. The strongest result of the session. Hotelul este în centru produced correctly on 2026-09-06 — the subject articled and the noun after în bare, both rules in one sentence. Post-key on the article, so not a cold reading.

Taught 2026-08-17. One stored bit per verb, and the eu form in this ledger is that bit. -ează is not a seventh ending: it is -ez- plus the ordinary -ă opened by the rule already held. The guess that works is whether the verb is the same word the learner knows in English or French, right on 14 of 14 of this intake, with a lucra and a repara as the two named breakers. Checked cold on 2026-08-19: the infix-drop at noi held (semnăm), the infix itself was applied correctly to an -ez- verb (telefonează stem), and it was wrongly applied to a bare verb (intrez). The bit is the failure point, not the mechanics. Two clean cold productions on 2026-08-24: lucram with -ez- correctly dropped at noi, and lucrezi correct at tu. The rule that the infix disappears at noi and voi is known; which verbs have one is not (ERR-072). Demoted 2026-08-25: a lucra tu produced as lucri, the infix dropped and the bare group I pattern run in its place. Second stored-bit failure on this row. Inverted on 2026-09-06: dansă produced for dansează, the infix omitted in the third person where it is required. The row's own text already says niciodată la noi și voi. ERR-093. On 2026-09-15 traversează came back as traversă three times, once in a plain statement and once after the key printed it — the -ez- third person missing for the second session running (ERR-093, ERR-105).

Taught 2026-08-17. One set of six endings covering five distinct forms, because el and ei share -ă. The stem then rewrites with the same operations that build a plural: t to ț before -i, e to ea and o to oa before -ă. Checked cold on 2026-08-19, 48 hours after teaching: 2 of 4 forms clean, and the two misses are the stem rewrite and the stored bit, not the endings. No ending was ever wrong. Two clean cold productions on 2026-08-24 — noi intrebam and noi lucram — and the tu -i ending was present on both group I items in a later round even where the stem vowel failed. Promoted on production. The third person failed twice on 2026-09-06 — pleacă and întreabă both missing, and dansă for dansează. Demoted from 2. The noi ending lost its ă twice as well. ERR-093, ERR-095.

Taught 2026-08-17 after the learner asked whether these were just memorised. ă to a before -ă is the third row of the opening rule already taught as e to ea and o to oa, and it fires only under stress, which is why cumpără does not open. What is left stored per verb is where the stem stress sits, and the eu form shows it. Forms verified dexonline 2026-08-17; the unification of the three rules is stated, not attested. 2026-08-20 cumpari for cumperi, cold and uncontaminated — the possessive page carries nothing about group I. The ă→e shift before an -i ending did not fire three days after teaching. Second failure, 2026-08-24, both clauses. cumpari for cumperi is character for character ERR-070 on the same verb, and spali for speli is the same shift on a second verb. In both the correct stem was printed in the prompt, so retrieval is not the explanation (ERR-073).

resembles Latin annus, EN annual; the noun in Am 30 de ani. un an doi ani clean on 2026-08-09 — masculine despite being a consonant-final inanimate, the gate exception produced correctly Clean RO→EN on 2026-09-06 at 25 days, and anul produced correctly in the article round.

Already held inside bună dimineața. dexonline 2026-08-25. The time expression is dimineața, with the definite ending doing adverb duty — learn it attached. Written demineața on 2026-08-25.

Latin dominica, the Lord's day — IT domenica, ES domingo. dexonline 2026-08-25.

După plus amiază, midday — built on the clock rather than on the meal. Romanian also has după-prânz and după-masă, and după-masă is exactly the shape of UA після обіду. dexonline 2026-08-25.

Jupiter, Jove — FR jeudi. dexonline 2026-08-25. Blank in the repair retest inside Joi dorm. Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 13 days.

Shortens to juma in speech. Half an hour is o jumătate de oră. dexonline 2026-08-25. Stands bare in a clock reading — trei și jumătate, no article. Blank in the repair retest 2026-08-25 after its answer had been published, ERR-085. Asked RO to EN cold at 13 days on 2026-09-07 and answered afternoon — după-amiază, a same-batch neighbour from the clock lesson.

Latin Lunae dies, the moon's day — the same thought as EN Monday, and the same word as lună. Collides head-on with the plural of lună: două luni is two months, luni alone is Monday. dexonline 2026-08-25.

same double meaning as UA місяць — obvious once noticed. Plural produced as lune on 2026-08-09, reasoned from casă case. Feminine takes -e or -i and which one is stored, not derived. Plural lune again on 2026-08-10 — second identical miss, still reasoned from casă case. Clean bare on 2026-08-19, with its final ă. Retrieval clean EN→RO on 2026-09-06; plural produced as lune, the -e rule where -i belongs. luna correct in the article round. Held at 3. ERR-092.

Mars — FR mardi, and the month martie is the same god. dexonline paradigm 2026-08-25 gives feminine, plural marți, definite marțea.

Mercury — FR mercredi. dexonline 2026-08-25. Asked EN→RO at first exposure the same day and answered vineri — see ERR-085's neighbours.

Neuter, so it counts masculine in the singular and feminine in the plural. dexonline 2026-08-25. Not needed in a clock reading: și douăzeci, not și douăzeci minute.

The greeting flips the order — noapte bună, see V-0020. dexonline 2026-08-25. Time expression noaptea. 23:00 is noaptea, not seara, and that boundary was missed on 2026-08-25. The definite noaptea was correct on 2026-09-06, cold. Clean EN→RO on 2026-09-07 — counts as Clean RO→EN pass 1 of 2.

EN hour, FR heure. Not ceas, the clock on the wall, and not timp, time in general. dexonline 2026-08-25. Lives inside the frame ora plus numeral, and on 2026-08-25 was twice produced after the numeral instead of before it. Produced as ora on 2026-09-07, cold at 13 days — the final ă absent, or the definite form in a bare slot. Second sighting of that shape in one session after piața.

Both the hour and the meal, the same double duty as UA обід. La prânz takes a preposition where the other four day parts take none. dexonline 2026-08-25.

Already held inside bună seara. dexonline 2026-08-25. Time expression seara, ending attached. The one word of today's five that came back clean at first exposure. Clean EN→RO on 2026-09-07, cold at 13 days — counts as Clean RO→EN pass 1 of 2.

Lives almost entirely in un sfert de oră, a quarter of an hour, and keeps its indefinite article inside the clock phrase where jumătate does not. dexonline 2026-08-25. Produced as saft at first exposure, ERR-085. Blank RO to EN cold at 13 days on 2026-09-07 — nothing retrieved. Failed in the same round as jumătate; the two clock-fraction words went together.

The one Slavic hook in the week — UA субота. dexonline 2026-08-25, plural sâmbete, which is irregular and worth its own asking. Answered dimenion at first exposure 2026-08-25, ERR-085.

Latin septimana, the seven — șapte is still inside it. dexonline 2026-08-25. That hook misfired at first exposure and produced septemată, ERR-085: name the hook, do not let it stand in for the word.

Venus — FR vendredi. dexonline 2026-08-25. Produced 2026-08-25 in place of miercuri, so the day set is present but unordered.

Latin dies, EN diary and journal. Answered ziua on 2026-08-07 — the definite form lifted out of bună ziua. Bare noun is zi. Plural produced as zi on 2026-08-09 — zile adds a syllable and is irregular. Plural zili on 2026-08-10 — a third different wrong ending for this word. Clean bare on 2026-08-19. Plural zile produced correctly on 2026-09-06 — the small -le class, cold. Retrieval not asked, so no promotion.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu beau (-ea class, group II). Latin bibere, EN beverage imbibe. The one irregular on its page: el bea, not bee. Conjugation class and full present verified dexonline 2026-08-20. Read only on 2026-08-20, not tested. Asked EN to RO cold on 2026-08-24 and produced as a bae: the two vowels of bea swapped and one lowered. A numbered typo question is open on this item.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu citesc (-i class, group IV, -esc- infix). Slavic cognate UA читати — named out loud because the Slavic layer does not surface unprompted. Conjugation class and full present verified dexonline 2026-08-20. Asked for the el form on 2026-08-20 and produced citești, the tu form — the infix was right, the person ending was not. Blank EN to RO cold on 2026-08-24, four days after the reading. The UA citati hook did not fire and a hook-less verb on the same page did (ERR-076). The noi form the same day was citizem, a phantom z infix. Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 14 days.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu dorm (-i class, group IV, bare — no infix). Latin dormire, EN dormitory; the ledger already holds dormitor. Conjugation class and full present verified dexonline 2026-08-20. Form produced correctly on 2026-08-20 (voi dormiți) with the infinitive supplied — the word itself was not tested. Produced EN to RO cold on 2026-08-24 at a four-day interval, first drilling of any kind. Promoted on a production pass, which is a harder gate than tier 1 asks for. Produced as a dormea EN to RO cold at 14 days on 2026-09-07 — the stem right and the -i class vowel written -ea. Demoted 2 to 1.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu fac (-e class, group III). Latin facere, EN factory manufacture fact. Conjugation class and full present verified dexonline 2026-08-20. Read only on 2026-08-20, not tested. Produced EN to RO cold on 2026-08-24 at a four-day interval.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu merg (-e class, group III). Declared hook-less rather than given an invented one: Latin mergere gave EN submerge, which is the wrong meaning. Conjugation class and full present verified dexonline 2026-08-20. Form produced correctly on 2026-08-20 (tu mergi) with the infinitive supplied — the word itself was not tested. Produced EN to RO cold on 2026-08-24. Notable against a citi: this verb was deliberately given no hook and came back, while the one Slavic cognate on the page did not (ERR-076). Blank EN to RO on 2026-09-07, cold at 14 days. Demoted 2 to 1.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mănânc (-a class, group I, irregular stem). Latin manducare, FR manger. Store mănânc against noi mâncăm as one pair — the stem is not derivable. Conjugation class and full present verified dexonline 2026-08-20. Asked EN to RO on 2026-08-20, twenty minutes after reading, and came back blank. Asked EN to RO cold on 2026-08-24 and produced as a mananca: the root is right and the irregular present stem was carried into the infinitive. Half credit, held at tier 1.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu pot (-ea class, group II). Latin posse, EN possible potent. Irregular: pot poți poate against putem puteți — the o stem in the singular and the third plural, the u stem at noi and voi. Not pod, a bridge. Conjugation class and full present verified dexonline 2026-08-20. Asked for the ei form on 2026-08-20 and produced put — the bare shape was right and the stored stem was the noi one. Produced EN to RO cold on 2026-08-24 at a four-day interval. The pot/putem stem split is still untested.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu scriu (-e class, group III). Latin scribere, EN describe script. The eu form adds -u because the stem ends in a consonant cluster, the same reason a intra gives intru. Conjugation class and full present verified dexonline 2026-08-20. Read only on 2026-08-20, not tested. Produced EN to RO cold on 2026-08-24 at a four-day interval. Produced as a scribi on 2026-09-07, cold at 14 days — the stem scri- is right and the Latin scribere supplied the ending. Demoted 2 to 1.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mă culc (-a class, group I; reflexive). Latin collocare, the root of FR se coucher and EN couch. Conjugation class and full present verified dexonline 2026-08-20. Form produced correctly on 2026-08-20 (tu te culci) with the infinitive supplied — the word itself was not tested. Asked EN to RO cold on 2026-08-24 and produced as a se usia: the reflexive marker correct, the stem an invention (ERR-076).

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mă grăbesc (-i class, group IV, -esc- infix; reflexive). Declared hook-less rather than given an invented one. Conjugation class and full present verified dexonline 2026-08-20. Asked EN to RO on 2026-08-20 inside Nu mă grăbesc and came back blank. Asked EN to RO cold on 2026-08-24 and produced as a gara: the g-r skeleton survived, the reflexive marker was dropped, the rest invented (ERR-076). Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 14 days.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mă odihnesc (-i class, group IV, -esc- infix; reflexive). Russian is the anchor and Ukrainian is not: отдыхать shares the root, відпочивати does not. Reflexive in Romanian and in neither Slavic verb. Conjugation class and full present verified dexonline 2026-08-20. Read only on 2026-08-20, not tested. Blank EN to RO cold on 2026-08-24 with no reflexive marker attempted (ERR-076). Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 14 days.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mă spăl (-a class, group I; reflexive). Declared hook-less. The tu form te speli carries the ă to e shift before -i. Reflexive in Ukrainian too, like митися. Conjugation class and full present verified dexonline 2026-08-20. Read only on 2026-08-20, not tested. Blank EN to RO cold on 2026-08-24, the reflexive marker a se produced without the stem (ERR-076). Separately the tu form te speli failed the a-to-e shift as spali, with the stem visible in the prompt (ERR-073).

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mă trezesc (-i class, group IV, -esc- infix; reflexive). Reflexive in Ukrainian too — прокидатися carries its own -ся. Conjugation class and full present verified dexonline 2026-08-20. Asked for the noi form on 2026-08-20 and produced ne trezem — the pronoun was right and the infix was correctly dropped, the group IV -im came out as -em. Produced EN to RO cold on 2026-08-24, the only reflexive of seven that came back (ERR-076). The voi form the same day took the noi ending. Only the reflexive marker a se came back EN to RO on 2026-09-07, cold at 14 days; the verb itself was absent. The learner knew the row was reflexive and not which verb it was. Demoted 2 to 1.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mă uit (-a class, group I; reflexive). The same body as V-0147 a uita, to forget — the pronoun is the whole difference, and they are two verbs. Reflexive in Ukrainian too, like дивитися. Conjugation class verified dexonline 2026-08-20. Read only on 2026-08-20, not tested. Blank EN to RO cold on 2026-08-24 (ERR-076).

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu mă îmbrac (-a class, group I; reflexive). Latin bracae, trousers — the root behind EN breeches. Reflexive in Ukrainian too, like одягатися. Conjugation class and full present verified dexonline 2026-08-20. Read only on 2026-08-20, not tested. Blank EN to RO cold on 2026-08-24, the reflexive marker a se produced without the stem (ERR-076).

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. eu văd (-ea class, group II). Latin videre, EN video evident. The tu form vezi carries the d to z alternation. Conjugation class and full present verified dexonline 2026-08-20. Asked for the noi form on 2026-08-20 and produced vedăm — group I's -ăm ending on a group II verb. Produced EN to RO cold on 2026-08-24. The noi form the same day inserted a phantom infix, vedezem for vedem, which is ERR-072 and not a fact about this row.

Literally how much is the clock. Written with a question mark, dropped here so the row parses. Ceas is the clock, never UA час, which is time — that is timp. The other frame is Ce oră este, and on 2026-08-25 the two were blended into Câti e oră.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. Fixed phrase, front of the sentence. Declared hook-less. obicei is a custom or habit, a neuter noun. Attested dexonline 2026-08-20. Asked EN to RO on 2026-08-20, twenty minutes after reading, and came back blank. Blank EN to RO cold on 2026-08-24 (ERR-076). Clean RO→EN pass 1 of 2 on 2026-09-07, cold at 14 days.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. Adverb. Built on vreme, a Slavic loan — RU время is the closer match, UA says час. Literally in good time. Attested dexonline 2026-08-20. Asked EN to RO on 2026-08-20 and produced davrema — the word was retrieved, both outer e came out as a. ORTH-VOW. Blank EN to RO cold on 2026-08-24. Asked twice now and never right: davrema on 2026-08-20, dovrema here. Different wrong vowels each attempt, so the form is being re-guessed rather than stored wrong (ERR-071 draws the contrast with flora and carta).

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. Adverb, and an adjective, which is how dexonline lists it. Latin tardivus, EN tardy retard. Attested dexonline 2026-08-20. Read only on 2026-08-20, not tested. Blank EN to RO cold on 2026-08-24, first test of this row. Blank RO to EN cold at 14 days on 2026-09-07.

Taught 2026-08-25. Stress picks the stem: the present stresses the stem in four forms and the ending at noi and voi, so a stem vowel that reacts to stress shows two shapes and the seam always falls in the same place. The infinitive shows the unstressed one, which is why a veni looks nothing like eu vin. Full present verified dexonline 2026-08-25. Both stems correct cold at 12 days on 2026-09-06 — venim and vin, asked in separate items.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. The mirror of group I, where el and ei share a form instead. Named as the page's trap because the collapse moves. Verified dexonline 2026-08-20. Stated correctly and unprompted on 2026-08-20.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. One stored bit per verb, carried by the eu form, and the same bit group I asks about -ez-. The infix drops at noi and voi, the same cells as -ez-. Verified dexonline 2026-08-20. On 2026-08-20 the infix itself was right in both sightings — kept in citești and correctly dropped in trezem — while the person ending and the noi vowel were not. On 2026-08-24 a citi gave citizem for citim: the infix neither dropped at noi nor was spelled -esc-. The two infixes are merging into one z-shaped chunk (ERR-072). Inverted on 2026-09-06 in both directions in one round: iub for iubesc, infix missing where required, and vorbesci for vorbiți, infix added where it must vanish. ERR-093.

Taught 2026-08-25 as one item rather than twelve rows, on the claim that the set maps one to one onto Russian and is therefore a mapping rather than twelve retrievals. That claim is testable and untested: the gender half was produced correctly on 2026-08-25 and the preposition half was answered un for în.

Taught 2026-08-25. UA needs у понеділок and EN needs on; Romanian uses nothing. Produced as Îm vineri on the graded check, then blank in the repair retest — the correction removed the wrong procedure without installing one, which is the same-sitting repair limit already on record.

Taught 2026-08-25. ora două not ora doi, ora douăsprezece not ora doisprezece. One o'clock is the single exception and is a fixed convention rather than agreement — DOOM 3's note at dexonline's unu entry, read 2026-08-25. Not yet asked.

Taught 2026-08-25. Past the half hour you name the hour you are heading for, not the one you left. This half held on 2026-08-25 — the fără jump was produced correctly twice, including cold. What failed underneath it was the numerals, ERR-082, and the word order of the frame itself.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. Differs from groups II and III in two cells only: noi and voi take -im and -iți where II and III take -em and -eți. All forms verified dexonline 2026-08-20. Asked for the noi form of a se trezi on 2026-08-20 and produced trezem. VRB-CONJ. On 2026-08-24 dormim was correct cold and the group IV -i vowel was also right in trezim, where the person was wrong rather than the group (ERR-072). ERR-064's trezem did not recur.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. Groups II (-ea) and III (-e) take identical present endings; they differ only in where the infinitive is stressed. All forms verified dexonline 2026-08-20. Asked for the noi form of a vedea on 2026-08-20 and produced vedăm — group I's ending. VRB-CONJ. On 2026-08-24 mergem and scrieti were correct cold and vedezem inserted a phantom z infix into an otherwise correct -em ending (ERR-072). The ending set held at a four-day interval; ERR-064's group-vowel error did not recur. înțelegeți produced correctly on 2026-09-06, cold.

Taught by self-study from the 2026-08-19 present-tense page, read 2026-08-20. Six slots, five words — se serves both third persons. A separate word in front of an otherwise ordinary verb, with nu outside it. Matches UA -ся on the concept; Romanian fronts it and agrees it for person. Both sightings correct on 2026-08-20 (ne trezim, te culci). On 2026-08-24 the marker held on four of five attempts cold, three times as a bare a se with the stem missing entirely, and va was matched correctly to voi. The part with no Slavic support is the part that survived (ERR-076). Promoted on production.

se ≈ UA -ся: mă numesc ≈ називаюся — true help. Taught as a chunk SES-004; question and answer still fused — Ma tu numeți for Cum te numești. Cum te numești produced correctly on 2026-08-03 in free use but not under test — still fused Mă numesc produced unfused and correct on 2026-08-19, cold at 10 days — the question-answer fusion is gone. The dropped ă belongs to the diacritic entry, not to this row.

UA автобус, the same international word, and it behaves like every borrowed neuter here. Verified dexonline 2026-08-25. Taught inside cu autobuzul, where the definite article stays on after cu. Produced 2026-09-06 cold at 12 days as autubuzul: the article and the frame were right, the second vowel went o to u. The learner declared it genuine, not a slip, and the prompt printed autobuz on the same line. Produced 2026-09-15 cold at 21 days as busul — English bus under a correct Romanian article, the Ukrainian match unused (ERR-107).

UA білет, the same borrowing from French billet. English billet is the false friend. Verified dexonline 2026-08-25. Clean EN to RO bare pass 1 of 2 on 2026-09-06, cold at 12 days with no review in between.

Taught 2026-09-07. The gender splits two words: neuter colț, colțuri is a corner; masculine colț, colți is a fang. dexonline 2026-09-07. Blank EN to RO bare cold at 8 days on 2026-09-15; produced as la colț post-key the same day.

Taught 2026-09-07. dreaptă plus -a, the same shape as stânga. Latin directus, the same word as drept, straight — the anchor for keeping the pair apart. dexonline 2026-09-15. Produced 2026-09-15 cold at 8 days for left (ERR-106).

Taught 2026-09-07. EN intersection; the -ție ending makes its plural in -ții, like stație. dexonline 2026-09-07. Not tested.

Taught 2026-09-07. UA машина matches in form and everyday meaning, and keeps the sense machine too. dexonline 2026-09-07. Not tested.

Taught 2026-09-07. UA метро never inflects; Romanian does — metroul, metrouri. dexonline 2026-09-07. Not tested.

Taught 2026-09-07. UA семафор is the same word as a railway signal. dexonline 2026-09-07. Clean EN to RO bare cold at 8 days on 2026-09-15 — clean RO→EN pass 1 of 2, met by a production answer a rung above it, as V-0005 was.

UA станція, matching closely. Kept apart from gară, which the learner already has: a railway station is a gară, a stație is where a bus or a metro stops. Verified dexonline 2026-08-25. Blank on 2026-09-06 — no recall at all, cold at 12 days. The only one of the eight U07 words that produced nothing. The definite stația was correct on 2026-09-06, cold. Named on 2026-09-06 as one of the two words that did not survive the U07 gap; the article on it is fine. Produced 2026-09-15 as stațiae for the station — spelling pending the learner's declaration; the word itself was retrieved cold.

Taught 2026-09-07. Already articled: stângă plus -a, frozen, so la stânga keeps the article after a preposition (G-0055). dexonline 2026-09-07. Produced 2026-09-15 cold at 8 days for right, while dreapta came back for left in the same set (ERR-106).

Taught 2026-09-07. UA трамвай maps directly, ending and all. dexonline 2026-09-07. Produced 2026-09-15 cold at 8 days as tramul — English tram under a correct Romanian article, the Ukrainian match unused (ERR-107).

Nothing from Ukrainian — потяг is unrelated; English and French train carry it instead. The plural takes -uri, the neuter's other ending. Verified dexonline 2026-08-25. Asked EN to RO bare on 2026-09-06, cold at 12 days with no review in between, and produced tran. The consonant frame is intact and the vowel is wrong; declared genuine. Scored half.

eu ajung (-e class, group III). No hook — stored cold. Its whole present runs a merge's pattern cell for cell, both being group III, so the stem is the only new thing. It also means to be enough, which arrives later. Full present verified dexonline 2026-08-25. Produced correctly cold at 12 days on 2026-09-06 as ajungi, inside a sentence carrying three separate rules.

Taught 2026-09-07. eu cobor (group IV in -î). No hook in any language held — stored cold. Imperative coboară / coborâți, dexonline 2026-09-07. Asked EN to RO bare cold at 8 days on 2026-09-15 and produced doca (ERR-107). Used post-key the same day as coborați, â missing (ERR-110).

eu iau (-a class, irregular singular stem). No hook in any language the learner holds, and none invented — stored cold. Group I on paper and irregular in practice: the singular is ia- and shares nothing with the infinitive. Also the everyday verb for taking transport and taking a meal. Full present verified dexonline 2026-08-25. Clean EN to RO bare pass 1 of 2 on 2026-09-06, cold at 12 days; the infinitive was asked and the inflected forms were not.

Taught 2026-09-07. eu traversez (-a class, -ez- infix). EN traverse. Imperative traversează / traversați, dexonline 2026-09-07. The lexeme came back every time on 2026-09-15; the third person came back as traversă three times, once after the key printed it (ERR-105).

Taught 2026-09-07. eu urc (-a class, no infix). The transport partner of a coborî. Imperative urcă / urcați, dexonline 2026-09-07. Not tested.

eu vin (-i class, no infix). Group IV, no infix. Ukrainian offers nothing — приходити is built from a different root; Latin venire reaches English in convene, venue and advent. Two stems, vin- in the singular and ven- at noi and voi. Full present verified dexonline 2026-08-25. Both stems produced correctly cold at 12 days on 2026-09-06 — venim in one item and vin in another.

Taught 2026-09-07. drept, straight, plus înainte, ahead; UA вперед covers the second half. Produced 2026-09-15 cold as mergeți drept — drept alone is attested as in a straight line, directly (DEX '09, checked 2026-09-15), so the Romanian was right and the chunk came back half.

An adverb, so it never changes shape and never takes an article — and it refuses every preposition, because a- is already Latin ad, to. Built on casă, V-0069. de acasă is the one prefix it accepts. Verified dexonline 2026-08-25. Produced correctly cold at 12 days on 2026-09-06. Not to be confused with la casă, which lands on the cash-desk sense of casă. Produced correctly in a sentence on 2026-09-06 but spelled acasa, the final ă dropped. The row's note already says it never changes shape. ERR-095.

Taught 2026-09-07. Latin ad prope, EN approach and approximate. Takes de before the place: aproape de. Produced correctly in a sentence on 2026-09-15, cold at 8 days, with nothing in the sitting printing it — clean RO→EN pass 1 of 2, met by production in a frame.

Taught 2026-09-07. EN depart, the same root. Takes de: departe de. dexonline 2026-09-07. Blank EN to RO bare cold at 8 days on 2026-09-15; produced post-key the same day.

Taught 2026-09-07. The mirror of înainte; UA назад. dexonline 2026-09-07. Not tested.

Taught 2026-08-25. The preposition is already inside the word, a- being Latin ad, so putting la in front says it twice. la casă is not a variant of it: bare casă after a preposition lands on the cash-desk sense of the noun, verified dexonline 2026-09-06, and la casa mea with the article is the building. la mine is what says at my place. Correct cold at 12 days on 2026-09-06.

Taught 2026-08-25 as the one exception to G-0033, which strips the article after a preposition. pe jos, on foot, takes neither cu nor an article. Iau autobuzul is the second way to say the same thing, and the article is there for the ordinary reason rather than as an exception. Correct cold at 12 days on 2026-09-06 — the -ul landed, on a word that was itself misspelled.

Taught 2026-08-25. Not a fourth set to memorise — de welds onto whichever of la, în or pe the place already took, and only de + în fuses, into din. din had been in use since July inside sunt din Ucraina without ever being seen as two words. Both branches correct cold at 12 days on 2026-09-06: de la spital and din magazin, in separate items.

Taught 2026-09-07 as a bet, not a law — vezi! breaks it. On 2026-09-15 mergi was correct, and traversează came back as traversă twice: the el/ea slot chosen right, the -ez- missing (G-0031, ERR-105).

Taught 2026-09-07; every form read off dexonline paradigms that day. Correct cold at 8 days on 2026-09-15, alone (așteptați, luați) and in sentences (luați, mergeți).

Taught 2026-09-07. Asked the familiar of a se grăbi on 2026-09-15 and given the polite grăbiți-vă: pronoun placement right, register wrong (ERR-108). The learner declared the missing ă and hyphen a slip.

Taught 2026-09-07 as the trap of U07 part 2: stânga and dreapta are articled forms, frozen — the exception family of G-0045. Both kept the article correctly on 2026-09-15, with left and right swapped (ERR-106).

Taught 2026-08-25. la for an errand, an institution or a point on the map, and the default when the choice is unclear; în for a bounded space you are physically inside; pe for a surface or a line you travel along, which is why a street is pe. The same noun takes both la and în with different meanings — la școală is studying, în școală is inside the building. The choice was correct 6 of 6 cold at 12 days on 2026-09-06, pe included: the pe item failed on its spelling, pa, and not on the choice.

Taught 2026-09-07. The irregular affirmatives vanish here — vino becomes nu veni. nu veni correct cold alone on 2026-09-15. nu face given where the polite nu faceți was asked (ERR-108).

Taught 2026-09-07; dexonline paradigms 2026-09-07. Their polite forms are the voi form except fiți. fă produced correctly cold in a sentence on 2026-09-15.

Taught 2026-08-25. Romanian has no separate where-to form: the place-phrase keeps one shape and only the verb in front of it changes. UA and RU carry the distinction on the noun instead — в магазині against в магазин, на вулиці against на вулицю — so the work here is subtraction, not a second paradigm to learn. Two cold items 12 days later on 2026-09-06, merg la teatru and suntem la poștă, both correct.