Vocabulary Sensors Pack (Instrument Panel) — Standalone Page (Almost-Code, Paste-Ready)

If you cannot measure vocabulary reliably,
you will train the wrong thing (recognition),
and wonder why output doesn't improve.

Vocabulary_Sensor := a measurable signal that predicts
whether a word/phrase/structure will appear correctly in real output
(speaking/writing) under load.

Law1: Recognition ≠ Retrieval
Law2: RetrievalSpeed predicts usability
Law3: Load reveals the real phase (timer/stress)
Law4: Output is the only final proof
Law5: Stability across days beats single-day performance

PhaseClassifier(word) {
  if unknown → P0
  else if recognises but cannot produce → P1
  else if produces slowly / needs prompting → P2
  else if produces <2s correctly across 3 sessions + under load → P3
}

RT(word) = time to produce word meaningfully (no prompting)
Thresholds:
  RT > 5s      → P1/P2 (not usable)
  2s < RT ≤5s  → P2 (weak active)
  RT ≤ 2s      → candidate P3 (needs stability checks)

UC(word) = correct meaning + correct collocation + correct tone/register
UC must be TRUE for P3.

LS(word) = word still appears correctly during timed writing/speaking
If LS fails → not P3.

SD(word) = correct retrieval on 3 spaced sessions:
  T+0, T+24h, T+7d
If SD fails → drift exists (P2 masquerading as P3).

OAR = how often target vocab appears in real output when appropriate
Low OAR = either retrieval weak OR selection habit weak.

CER = wrong word pairing frequency
High CER → missing chunk bank (Z1 weakness)

RER = formal/informal mismatch frequency
High RER → missing usage frame

SSF = meaning breaks when swapping to a “synonym”
High SSF → shallow semantics (word is not truly owned)

HE = pauses mid-sentence caused by word retrieval
High HE → retrieval latency + load sensitivity

GVR = count(do/make/get/put) per paragraph
High GVR → vocabulary depth missing (weak verb bank)

VI = unique strong verbs + unique descriptive phrases per 120 words
Low VI → thin synonym families + chunk shortage

Pick 10 target words (from recent reading / current topic).
For each word:
  1) define it in 1 simple sentence (meaning)
  2) use it in 1 natural sentence (usage)
  3) say a second sentence with different context (transfer)
Record:
  - RT
  - UC

If many RT > 2s → P2 system
If UC fails often → context binding missing
If performance collapses under timer → load training missing

WEEKLY_VOCAB_DASHBOARD {
  Targets: 8 words + 4 chunks + 2 structures
  Metrics:
    Avg_RT: ____
    UC_PassRate: ____%
    LS_PassRate (timed): ____%
    SD_PassRate (T+0/24h/7d): ____%
    VI (per 120w): ____
    GVR (per paragraph): ____
    FreezeEvents (timed writing): ____
  DominantLeakTag: [Speed | Usage | Load | Drift | Selection]
  RepairPlanNextWeek: [R0/R1/R2/R3 chosen]
}

DominantLeakTag :=
  Speed     (RT high)
  Usage     (UC fails)
  Load      (LS fails)
  Drift     (SD fails)
  Selection (OAR low despite ability)

If Speed:
  → R0b 3-recall drill + short timed speaking
If Usage:
  → R0a 2-sentence anchor + R1a collocation bank
If Load:
  → timed micro-paragraph drills + vocab injection targets
If Drift:
  → spaced recall schedule + weekly recycling list
If Selection:
  → “trigger prompts” + forced injection + rewrite weakest paragraph

MV_Sensors {
  RT < 2s
  UC correct in sentence
  LS survives timed writing
}

MCQ tests recognition (P1), not production (P3).

Single success ≠ stability.
P3 requires spaced success + load survival.

One-time usage can be luck or prompting.
Measure SD + LS.

VOCAB_SENSORS_SPEC v1.0 {
  CoreSensors: RT, UC, LS, SD, OAR
  P3 Gate: RT≤2s + UC=true + LS=true + SD pass across 3 sessions
  LeakTags: Speed|Usage|Load|Drift|Selection
  Repairs: R0/R1/R2/R3 mapped to leak tags
}