0) Purpose

Goal: Convert exposure → stable retrieval → correct usage → transfer under load.

Primary output: ActiveVocabularyReliability increases (fast recall + correct usage + transfer).

Start Here: https://edukatesg.com/how-to-improve-vocabulary-vocabulary-os/

1) Definitions (LOCK)

1.1 Vocabulary Unit Types

UnitType.word
UnitType.phrase (preferred)
UnitType.collocation
UnitType.idiom
UnitType.pattern (sentence frame)

1.2 Passive vs Active

PassiveVocab = recognition (read/listen)
ActiveVocab = production (write/speak) without prompts

1.3 Reliability Levels

P0 = cannot recall / frequent misuse
P1 = sometimes recall, slow, unstable
P2 = recall mostly correct, fails under load
P3 = reliable under load + transfers across contexts

2) Core Records

2.1 `UnitRecord v1.1`

Required fields (minimal to prevent “fake learning”):

id (stable)
unit (string)
type (UnitType)
meaning_simple (1 line)
when_to_use (context)
when_not_to_use (common misuse)
bind (grammar + collocation constraints)
ex_easy (simple sentence)
ex_exam (higher load sentence)
transfer_1 (different topic sentence)
pron_spell (optional)
error_log (user-specific)

2.2 `UnitState`

last_seen_date
last_recall_date
recall_latency_sec
recall_success (bool)
usage_accuracy (0–1)
transfer_success_count (0–3)
load_stability (0–1)
status ∈ {red, yellow, green}

3) Sensor Pack (Metrics)

3.1 Capture Sensors

S.capture_units_per_day
S.capture_phrase_ratio = phrases / total units

3.2 Retrieval Sensors

S.active_recall_rate_7d (0–1)
S.recall_latency_avg_7d (seconds)
S.drop_rate_7d = (units learned last 14d but failed recall this week) / (units learned last 14d)

3.3 Usage Sensors

S.collocation_accuracy_7d
S.grammar_binding_accuracy_7d
S.register_accuracy_7d

3.4 Transfer Sensors

S.transfer_rate_7d = (units used correctly in ≥2 contexts) / (eligible units)

3.5 Load Sensors

S.timed_output_retention = (new units used correctly in timed task) / (target units)
S.load_crash_indicator = 1 if performance drops ≥30% under time pressure, else 0

4) Thresholds (Defaults)

4.1 Minimum effective input

T.capture_units_per_day >= 3
T.capture_phrase_ratio >= 0.60

4.2 Retrieval thresholds

T.active_recall_rate_7d >= 0.70
T.recall_latency_avg_7d <= 2.0 sec
T.drop_rate_7d <= 0.25

4.3 Usage thresholds

T.grammar_binding_accuracy_7d >= 0.80
T.collocation_accuracy_7d >= 0.75

4.4 Transfer thresholds

T.transfer_rate_7d >= 0.50

4.5 Load thresholds

T.timed_output_retention >= 0.40
T.load_crash_indicator == 0

5) Grading Rule (Green/Yellow/Red)

For each UnitRecord weekly:

5.1 `Green`

recall_success == true
recall_latency_sec <= 2
usage_accuracy >= 0.85
transfer_success_count >= 2

5.2 `Yellow`

recall succeeds but slow, or usage unstable
Examples:
2 < recall_latency_sec <= 5, or
usage_accuracy < 0.85, or
transfer_success_count == 1

5.3 `Red`

recall_success == false OR frequent misuse
usage_accuracy < 0.60

6) Repair Library (Interventions)

Each repair has: trigger → action → exit_condition.

6.1 `R1.SpacedRetrieval`

Trigger: S.active_recall_rate_7d < 0.70 OR unit is red
Action: daily recall drill (no looking) + 1 immediate correct sentence
Exit: unit becomes yellow for 2 checks

6.2 `R2.BindingFix`

Trigger: S.grammar_binding_accuracy_7d < 0.80 OR repeated same mistake in error log
Action: extract bind rule + do 5 micro-sentences varying subject/tense/prep
Exit: usage_accuracy ≥0.80 for 7 days

6.3 `R3.CollocationUpgrade`

Trigger: S.collocation_accuracy_7d < 0.75
Action: learn 3 collocations + substitution drill (“replace simple phrase with target collocation”)
Exit: collocation accuracy ≥0.75 for 7 days

6.4 `R4.TransferTriple`

Trigger: S.transfer_rate_7d < 0.50 OR unit only works in one topic
Action: force 3 contexts: (school), (home), (news/real-world)
Exit: transfer_success_count >= 2

6.5 `R5.LoadTraining`

Trigger: S.load_crash_indicator == 1 OR S.timed_output_retention < 0.40
Action: timed paragraph/speaking drill with “must-use 6 units”
Exit: timed retention ≥0.40 for 2 weeks

6.6 `R6.InputCalibration`

Trigger: high capture but low recall (capture_units high, recall rate low)
Action: reduce new units/day by 30–50%, increase retrieval cycles
Exit: drop_rate ≤0.25

7) Repair Router (Decision Logic)

7.1 `VocabularyRepairRouter v1.1`

Inputs: sensor pack S + unit states.

Routing order (highest leverage first):

If S.active_recall_rate_7d < 0.70 or S.drop_rate_7d > 0.25
→ run R1.SpacedRetrieval + R6.InputCalibration
Else if S.grammar_binding_accuracy_7d < 0.80
→ run R2.BindingFix
Else if S.collocation_accuracy_7d < 0.75
→ run R3.CollocationUpgrade
Else if S.transfer_rate_7d < 0.50
→ run R4.TransferTriple
Else if S.load_crash_indicator == 1 or S.timed_output_retention < 0.40
→ run R5.LoadTraining
Else
→ keep baseline protocol + expand topics (increase phrase bank breadth)

8) Baseline Protocol (Daily / Weekly)

8.1 Daily (10–15 min)

Capture(3–5 units) where phrase_ratio >= 0.60
Encode(UnitRecord) using required fields
Retrieve(yesterday_units) without looking
Output(2 sentences OR 30s speech) must include today’s units
ErrorLog(1–2 corrections) (bind/collocation)

8.2 Weekly (30 min)

Run GreenYellowRedReview
Compute sensor pack S
Run VocabularyRepairRouter
Assign next-week focus: red -> yellow -> green

9) 12-Week Upgrade Plan (Phase Schedule)

Weeks 1–3: Bank + Input calibration (build clean UnitRecords, phrase-first)
Weeks 4–6: Retrieval engine (R1 dominates)
Weeks 7–9: Load training (R5 introduced)
Weeks 10–12: Transfer + exam application (R4 + mixed prompts)

V1.1 Explanation Layer (Human-Readable)

Vocabulary improves when the recall pipeline gets trained, not when you “collect more words.” The system above forces the conversion:

Capture only a small amount daily (3–5), mostly phrases.
Encode each unit with binding rules + “when not to use,” so you stop misusing it.
Retrieve with spaced recall, so it becomes active.
Output daily, because output is where vocabulary becomes real.
Load train so it survives exams and real conversation speed.
Transfer across topics so it isn’t trapped in one story or one theme.

If you follow only one rule:
Stop counting words. Start measuring recall rate and latency. That’s the real indicator of vocabulary growth.

V1.1 — How to Improve Vocabulary (Almost-Code / Directory Spec)