Vocabulary Repair Router (Standalone Page) — Checklist + Decision Tree (Almost-Code, Paste-Ready)

Do not “practice more”.
Repair the dominant leak.
Then train under load.

Router := function that maps:
  Symptom → LeakTag → RepairLoop → OutputTest → NextStep

BASELINE_TEST {
  Pick 10 target words (from recent passage / current theme).
  For each word:
    1) define in 1 simple sentence
    2) use in 1 natural sentence
    3) say 1 alternate-context sentence
  Record:
    RT (recall time), UC (usage correct)
}

If RT>2s often → Speed leak
If UC fails often → Usage leak
Then do Load test to confirm.

LeakTag :=
  Speed      (RT high)
  Usage      (wrong collocation/tone/meaning slice)
  Load       (collapses under timer)
  Drift      (works today, lost next week)
  Selection  (can do it, but doesn’t appear in writing)

IF RT(word) > 2s frequently
  → LeakTag = Speed
ELSE IF UC fails (meaning/collocation/tone)
  → LeakTag = Usage
ELSE DO Timed Output Test:
  write 8-min paragraph with 6 target upgrades
  IF targets disappear or errors spike
    → LeakTag = Load
  ELSE DO 7-day ReTest:
    IF words decay / cannot recall at T+7d
      → LeakTag = Drift
    ELSE
      → LeakTag = Selection

R_SPEED {
  Daily (10 min):
    - 3-Recall Drill for 5 words: T+0, +10m, +24h
    - Speak 5 sentences fast (no pausing allowed)
  Rule:
    - Do not write long essays.
    - Fix latency first.
}

Pass if: RT ≤ 2s for 80% of targets across 3 sessions.

R_USAGE {
  For each target word:
    - 2-Sentence Anchor (neutral + emotional)
    - Collocation Pair Bank (3 strong pairs)
    - Register label: [Formal | Neutral | Casual]
    - Anti-synonym trap: 1 “close word” + explain difference
}

Pass if: UC true in 5 original sentences + 2 transfer contexts.

R_LOAD {
  2–3x/week:
    - 8–12 min timed paragraph
    - forced injection: 6 upgrades (words/chunks)
    - forced connectors: 4 types
  After each session:
    - Rewrite weakest paragraph once
}

Pass if: targets appear correctly during timed writing (LS true).

R_DRIFT {
  Spaced schedule (non-negotiable):
    - Day 1: learn + use
    - Day 2: recall + use
    - Day 7: recall + timed use
  Weekly:
    - Recycling list: 20 old targets rotated back
}

Pass if: SD pass rate ≥ 80% at Day 7.

R_SELECTION {
  For every writing task:
    - pick 8 target words + 4 chunks + 2 structures
    - must inject at least 6 upgrades
    - highlight upgrades after writing
  Build triggers:
    - “emotion triggers” (angry/scared/proud)
    - “action triggers” (run/argue/escape)
    - “description triggers” (weather/setting)
}

Pass if: OAR increases (targets appear when appropriate) for 3 sessions.

10MIN_PROTOCOL {
  5 min: Speed drill (3-recall) for 5 words
  3 min: 2-sentence anchor for 1 word
  2 min: speak-then-write 1 sentence
}

20MIN_PROTOCOL {
  8 min: collocation bank (5 items)
  10 min: timed paragraph (inject 6 upgrades)
  2 min: rewrite worst sentence
}

40MIN_PROTOCOL {
  10 min: input (story excerpt / article)
  12 min: timed paragraph with targets
  12 min: feedback + rewrite weakest paragraph
  6 min: spaced review + recycle list
}

- Run baseline test
- Identify dominant leak
- Run the matching repair loop 4 days
- 2 timed paragraphs (if Load/Selection)

- 8 words + 4 chunks/week
- force injection in 2 timed pieces
- begin weekly recycle list

- do 3 timed paragraphs
- vary prompts/topics
- tighten UC correctness (collocation/tone)

- 1 full continuous writing piece
- 1 situational writing task (if Secondary)
- post-mortem: dominant leak shifts? update router

VOCAB_REPAIR_LOG {
  Date:
  Targets (8W/4C/2S):
  DominantLeakTag:
  RepairLoopUsed:
  TimedOutputDone? (Y/N):
  Metrics:
    Avg_RT:
    UC_PassRate:
    LS_PassRate:
    SD_PassRate:
    OAR:
  Notes (dominant mistakes):
  Next session plan:
}

VOCAB_REPAIR_ROUTER_SPEC v1.0 {
  Baseline: 10-word test with RT+UC
  LeakTags: Speed|Usage|Load|Drift|Selection
  DecisionTree: RT → UC → TimedOutput → Day7ReTest
  Repairs: R_SPEED/R_USAGE/R_LOAD/R_DRIFT/R_SELECTION
  KPI: P3 Gate = RT≤2s + UC=true + LS=true + SD pass
}