Skip to main content

absence-measurement-validity-check

Before concluding "X doesn't happen" in an integrated system, verify the pipeline that would produce X is actually running. Measuring zero is trivially easy when the producer is off.

Ir a la instalaciรณn

Datos de origen

Repositorio
blas1n/claude-skills
รšltima actividad en el origen
26 de agosto de 2026 a las 03:01
Idioma detectado de SKILL.md
Varios idiomas
Estrellas
2
Forks
0

Opciones de instalaciรณn

De forma predeterminada estรก seleccionado el prompt que primero revisa el origen. Puedes cambiar a un comando directo o descargar una copia local.

Revisa los archivos de origen

Lee SKILL.md y los archivos complementarios que muestra SkillsMP antes de decidir si quieres instalarlo.

Mostrando SKILL.md

SKILL.md
Instrucciones de origen ยท Vista previa de solo lectura
name
absence-measurement-validity-check
description
Before concluding "X doesn't happen" in an integrated system, verify the pipeline that would produce X is actually running. Measuring zero is trivially easy when the producer is off.
version
1.0.0
task_types
["debugging","design","evaluation"]
triggers
[{"pattern":"claim that a tool / feature / behavior isn't being used / fires zero times"},{"pattern":"longrun / E2E experiment showing 0 count of some event"},{"pattern":"prompt engineering attempt judged 'failed' because LLM didn't do X"}]
category
trap
# Absence Measurement Validity Check ## The Pattern You run an experiment and observe `X = 0` (tool never called, artifact never created, path never hit). You conclude "X doesn't work / the prompt failed / the model has a limit". This conclusion is only valid if the pipeline that would produce X is actually running. Otherwise you measured a system that can't produce X *at all*, not a system that chose not to. ## Real Example (BSNexus Session 10) Goal: prove or disprove that GLM-4.7-flash can be prompted to call a verification tool (`shell_exec`) via TDD-style instructions. Measurement loop, three iterations: | Run | Prompt strategy | `shell_exec_ran` count | Conclusion I drew | |---|---|---|---| | v8 | direct "MUST verify" instruction | 0 | "direct instruction doesn't work" | | v9 | Q1/Q2/Q3 CoT scaffold | 0 | "CoT doesn't work" | | v10 | E2E reframing + file_read banned | 0 | "prompt-layer ceiling hit" | Three "failed" prompt strategies in a row. Looks decisive. What I missed: the user pushed me to run V11 patiently and query the DB directly. Results: - **Assigned agents**: Designer 8, CTO 4, Marketer 3, Backend_Engineer 1, Frontend_Engineer 1, QA_Lead 1โ€ฆ (routing was **working**) - **`file_read` calls**: 22 (passive workers were **active**, calling other tools) - **`create_screen` / `file_write`** also firing normally - **`shell_exec`**: 0 The passive-worker pipeline was fine. The `shell_exec` absence was a real GLM tool-preference signal, but I could not have known that from v8/v9/v10 โ€” I didn't check whether the pipeline that *could* use `shell_exec` was even dispatching. For all I knew from those three runs, every task was self-assigned to the planning agent and no passive worker was running at all. ## Why This Traps You - `X == 0` reads like a clean data point. It feels like certainty. - Positive signals are loud (logs, state transitions, artifacts); absence is invisible. There's nothing to question, only something missing. - Repeating the same measurement under different prompts **does not** increase confidence in the conclusion. All three runs can be contaminated by the same upstream gap. - A later negative result that's real confirms the wrong early reasoning, so the lesson never surfaces. ## The Validity Check โ€” Before Claiming Absence Before writing down "X didn't happen โ†’ Y caused it", verify three layers: 1. **Producer liveness**: is the process that would emit X actually dispatched/active in this run? Check logs for "its kind of event happened at all", not just the specific signal. 2. **Sibling signals**: does the same code path emit *anything*? If the producer fires same-category events successfully (other tool calls, other file writes), you're comparing "made a choice not to" vs "never got the chance". Different conclusion, different fix. 3. **Artifact vs state**: state transitions (`status=done`, `phase=completed`) are cheap to fake. File-level / command-level artifacts are what matter. Query the DB / filesystem directly; don't trust API responses that might be showing creator vs assignee, cached vs live, etc. Only after those three layers check out is `X = 0` evidence of intentional absence rather than blocked pipeline. ## Heuristic > If your measurement is "how often did the LLM / agent / process do X?", and X is zero across N runs, your first follow-up question must be: **did the thing that produces X even execute?** Query the producer's *sibling* signals. If the siblings are also zero, the pipeline is dead; the experiment was invalid and the prompt / model / feature has not actually been tested. ## When This Skill Applies - Multi-agent systems where a tool might be offered but never called - Any E2E test claiming "feature X didn't fire" - Prompt engineering iterations producing the same null count - Backend enforcement rules that supposedly triggered no rejections - Dashboards showing a suspicious zero when activity is expected ## When It Does NOT Apply - Unit tests with synthetic input (the pipeline is explicit) - Experiments where the producer is obviously controlled (the call site is in your test code) ## Related Skills - `systematic-debugging` โ€” broader root-cause investigation; this skill is the "check absence first" corner of it. - `verification-before-completion` โ€” the inverse problem (premature success claims); this one covers premature failure claims. - `test-against-source-contracts` โ€” when API field semantics confuse you (e.g., `agent_name` returning creator vs assignee), this ties in. --- ## ์‚ฌ๋ก€ โ€” **์ธก์ • ๋ช…๋ น ์ž์ฒด๊ฐ€ ์‹คํŒจํ–ˆ์„ ๋•Œ** (2026-08-17) ์ด ์Šคํ‚ฌ์€ "0์„ ๋งŒ๋“ค์–ด๋‚ด๋Š” *์ƒ์‚ฐ์ž*๊ฐ€ ๋„๋Š”๊ฐ€"๋ฅผ ๋ฌป๋Š”๋‹ค. ๊ทธ ํ•œ ๊ฒน ์•„๋ž˜์— ๋” ํ”ํ•œ ๊ฒƒ์ด ์žˆ๋‹ค: **์ธก์ • ๋ช…๋ น ์ž์ฒด๊ฐ€ ๋Œ๊ธฐ๋Š” ํ–ˆ๋Š”๊ฐ€.** BSVibe ์‹ค์ธก. ๋ ˆํฌ ๋ฃจํŠธ์˜ ์ž„์‹œ ํŒŒ์ผ 5๊ฐœ๋ฅผ ์ง€์šฐ๊ธฐ ์ „์— ์ฐธ์กฐ๋ฅผ ํ™•์ธํ•˜๋ ค๊ณ : ```bash grep -rn "_patch_judge\|_patch_parse" --include=*.py backend/ tests/ # โ†’ (zsh) no matches found: --include=*.py โ† ๋ช…๋ น์ด ์ฃฝ์—ˆ๋‹ค ``` zsh ๊ฐ€ `--include=*.py` ๋ฅผ ๊ธ€๋กญ์œผ๋กœ ๋จน์–ด **๋ช…๋ น์ด ์‹คํ–‰๋˜์ง€ ์•Š์•˜๋‹ค.** ์ถœ๋ ฅ์€ ๋น„์–ด ์žˆ์—ˆ๊ณ , ๋‚˜๋Š” ๊ทธ ๊ณต๋ฐฑ์„ **"์•„๋ฌด๋„ ์ฐธ์กฐํ•˜์ง€ ์•Š๋Š”๋‹ค"** ๋กœ ์ฝ๊ณ  ํŒŒ์ผ์„ ์ง€์› ๋‹ค. > **๋นˆ ์ถœ๋ ฅ์€ "์—†์Œ"์ด ์•„๋‹ˆ๋‹ค. ๊ทธ ๋ช…๋ น์ด ์‹คํ–‰๋๋‹ค๋Š” ์ฆ๊ฑฐ๊ฐ€ ๋จผ์ € ํ•„์š”ํ•˜๋‹ค.** ์ œ๋Œ€๋กœ ๋‹ค์‹œ ๋Œ๋ฆฌ๋‹ˆ 13๊ฐœ ํŒŒ์ผ์ด ๋งค์นญ๋๋‹ค(์ „๋ถ€ `monkeypatch_resolver` ๊ฐ™์€ ๋ฌด๊ด€ํ•œ ๋ถ€๋ถ„ ๋ฌธ์ž์—ด์ด๋ผ ์‚ญ์ œ๋Š” ๊ฒฐ๊ณผ์ ์œผ๋กœ ์•ˆ์ „ํ–ˆ์ง€๋งŒ, **๊ทธ๊ฑด ์šด์ด์—ˆ๋‹ค**). ์‚ญ์ œยท๋จธ์ง€ยท๋ฐฐํฌ์ฒ˜๋Ÿผ **๋น„๊ฐ€์—ญ** ๋™์ž‘ ์•ž์—์„œ๋Š” ์ด ํ™•์ธ์ด ํ•„์ˆ˜๋‹ค. **์ฒ˜๋ฐฉ** - ๋ถ€์žฌ๋ฅผ ๊ทผ๊ฑฐ๋กœ ํ–‰๋™ํ•˜๊ธฐ ์ „์— **์–‘์„ฑ ๋Œ€์กฐ**๋ฅผ ํ•˜๋‚˜ ๋ผ์›Œ๋ผ โ€” ๋ฐ˜๋“œ์‹œ ๋งค์นญ๋  ๋ฌธ์ž์—ด๋กœ ๊ฐ™์€ ๋ช…๋ น์„ ๋Œ๋ ค **0์ด ์•„๋‹Œ ๊ฒฐ๊ณผ**๊ฐ€ ๋‚˜์˜ค๋Š”์ง€ ๋ณธ๋‹ค. ์•ˆ ๋‚˜์˜ค๋ฉด ๋ช…๋ น์ด ์ฃฝ์€ ๊ฒƒ์ด๋‹ค. - ์ข…๋ฃŒ ์ฝ”๋“œ๋ฅผ ๋ด๋ผ. `grep` ์˜ "๋งค์นญ ์—†์Œ"์€ 1, **์˜ค๋ฅ˜๋Š” 2**๋‹ค. ์…ธ์ด ์ฃฝ์˜€์œผ๋ฉด ๊ทธ๋งˆ์ € ์•ˆ ๋‚˜์˜จ๋‹ค. - ์…ธ ํ™•์žฅ์ด ๊ฐœ์ž…ํ•  ์ˆ˜ ์žˆ๋Š” ์ธ์ž(`*`, `?`, `[`)๋Š” **๋”ฐ์˜ดํ‘œ๋กœ ๊ฐ์‹ธ๋ผ**. ```bash # ์–‘์„ฑ ๋Œ€์กฐ๋ฅผ ๊ฐ™์ด ๋Œ๋ฆฐ๋‹ค grep -rn "_patch_judge" backend tests; echo "exit=$?" grep -rn "def test_" tests | head -1 # โ† ์ด๊ฒŒ ๋น„๋ฉด ๋ช…๋ น์ด ์ฃฝ์€ ๊ฒƒ์ด๋‹ค ``` **Detection**: ๋นˆ ์ถœ๋ ฅ์„ ๊ทผ๊ฑฐ๋กœ **์‚ญ์ œ/์ •๋ฆฌ/๋กค๋ฐฑ**์„ ํ•˜๋ ค ํ•œ๋‹ค ยท ์…ธ ์˜ค๋ฅ˜ ๋ฉ”์‹œ์ง€๊ฐ€ ์ถœ๋ ฅ์— ์„ž์—ฌ ์žˆ๋Š”๋ฐ ๊ฒฐ๊ณผ๋ถ€๋งŒ ์ฝ์—ˆ๋‹ค ยท `tail`/`head` ๋กœ ์ž˜๋ผ ๋ณด๋А๋ผ ์•ž์˜ ์—๋Ÿฌ๋ฅผ ๋ชป ๋ดค๋‹ค(๊ฐ™์€ ์„ธ์…˜์—์„œ `ruff` ์˜ F821 ์„ `tail` ๋กœ ๋†“์ณ 85๊ฐœ ํ…Œ์ŠคํŠธ๋ฅผ ๊นจ๋œจ๋ ธ๋‹ค). --- ## ์‚ฌ๋ก€ โ€” **์ง€ํ‘œ๋กœ ๊ณ ๋ฅธ ํ…Œ์ด๋ธ” ์ž์ฒด์— producer ๊ฐ€ ์—†์—ˆ๋‹ค** (BSVibe, 2026-08-20) ๊ฐ€์žฅ ๋น„์‹ผ ํ˜•ํƒœ๋‹ค: ๋ถ€์žฌ๋ฅผ ์ž˜๋ชป ์ฝ์€ ๊ฒŒ ์•„๋‹ˆ๋ผ, **๋ถ€์žฌ๋ฅผ ์žฌ๋ ค๊ณ  ๊ณ ๋ฅธ ๊ณ„๊ธฐํŒ์ด ์• ์ดˆ์— ์•ˆ ๊ฝ‚ํ˜€ ์žˆ์—ˆ๋‹ค.** `IngestCompiler` ๊ฐ€ `retriever=` ๋ฅผ ๋ชป ๋ฐ›์•„ ๋ˆˆ๋ฉ€์–ด ์žˆ๋˜ ๊ฒƒ์„ ๊ณ ์น˜๊ณ , PR ๋ณธ๋ฌธ์— ๊ฒ€์ฆ ๋ฐฉ๋ฒ•์„ ์ด๋ ‡๊ฒŒ ์ ์—ˆ๋‹ค โ€” *"๋ฐฐํฌ ํ›„ `ingest_batches` ์˜ `notes_updated` ๊ฐ€ 0์—์„œ ์›€์ง์ด๋Š”์ง€ ๋ณด๋ผ."* ์‹ค์ œ๋กœ ์žฌ๋ณด๋‹ˆ: | | | |---|---| | `ingest_batches` ํ–‰ ์ˆ˜ | **0** | | `IngestBatchRecorder` ํ”„๋กœ๋•์…˜ ๊ตฌํ˜„์ฒด | **0** (Protocol ์„ ์–ธ๋งŒ) | | `batch_recorder=` ๋ฅผ ๋„˜๊ธฐ๋Š” ์ƒ์„ฑ ์ง€์  | **0** | **๋‚ด๊ฐ€ ๋ฐฉ๊ธˆ ๊ณ ์นœ ๊ฒฐํ•จ๊ณผ ๋˜‘๊ฐ™์€ ๊ฒฐํ•จ์ด ๊ฐ™์€ ์ƒ์„ฑ์ž์˜ ๋ฐ”๋กœ ์˜† ์ธ์ž์— ์žˆ์—ˆ๋‹ค.** ๊ทธ ์ง€ํ‘œ๋Š” ์›๋ฆฌ์ƒ ์˜์›ํžˆ 0์ด๋ฏ€๋กœ, ๋‚ด ์ˆ˜์ •์ด ๋™์ž‘ํ•˜๋“  ์•ˆ ํ•˜๋“  "๊ฒ€์ฆ ์‹คํŒจ"๋กœ ์ฝํ˜”์„ ๊ฒƒ์ด๋‹ค. > **์ง€ํ‘œ๋ฅผ ๊ณ ๋ฅด๋Š” ๊ฒƒ๋„ ์ธก์ •์ด๋‹ค. ๊ณ„๊ธฐํŒ์— producer ๊ฐ€ ์žˆ๋Š”์ง€๋ถ€ํ„ฐ ์„ธ๋ผ.** **์ฒ˜๋ฐฉ** โ€” ๊ฒ€์ฆ ์ง€ํ‘œ(ํ…Œ์ด๋ธ”/์นด์šดํ„ฐ/๋กœ๊ทธ/์ด๋ฒคํŠธ)๋ฅผ ์ •ํ•  ๋•Œ **์“ฐ๊ธฐ ์ „์—** ํ™•์ธ: ```bash # 1) ๊ทธ ์ง€ํ‘œ์— ์ง€๊ธˆ ๊ฐ’์ด ๋“ค์–ด ์žˆ๋‚˜ (์—ญ๋Œ€ ํ•œ ๋ฒˆ์ด๋ผ๋„) psql -tAc "select count(*) from <metric_table>" # 2) ์—†๋‹ค๋ฉด โ€” ํ”„๋กœ๋•์…˜ producer ๊ฐ€ ์กด์žฌํ•˜๋‚˜ grep -rn "<RecorderClass>(" backend/ | grep -v tests # 0๊ฐœ๋ฉด ์ง€ํ‘œ๊ฐ€ ์•„๋‹ˆ๋ผ ์œ ๋ น์ด๋‹ค ``` 0์ด๋ฉด **์ง€ํ‘œ๋ฅผ ๋ฐ”๊พธ๊ฑฐ๋‚˜, ์ง€ํ‘œ๋ฅผ ๋จผ์ € ๋ฐฐ์„ ํ•˜๋ผ.** ๊ทธ๋ฆฌ๊ณ  "๋ฐฐ์„ ํ• ์ง€ ์ง€์šธ์ง€"๋Š” ๋Œ€๊ฐœ ์‚ฌ๋žŒ์˜ ๊ฒฐ์ •์ด๋‹ค โ€” producer ์—†๋Š” ํ…Œ์ด๋ธ”์€ *"์—ฐ๊ฒฐ์ด ๋น ์ง„ ๊ฒƒ"* ์ด ์•„๋‹ˆ๋ผ *"๋งŒ๋“  ์  ์—†๋Š” ๋‘ ๋ฒˆ์งธ ํ‘œํ˜„"* ์ผ ์ˆ˜ ์žˆ๊ณ , ๊ทธ๋Ÿฌ๋ฉด ๋‹ต์€ ์‚ญ์ œ๋‹ค. ### ๊ณ๊ฐ€์ง€ ํ•จ์ • โ€” ๋ฐฐํฌ๋Š” ๋กœ๊ทธ ํžˆ์Šคํ† ๋ฆฌ๋ฅผ ์ž๋ฅธ๋‹ค ๊ฐ™์€ ๊ฒ€์ฆ์—์„œ `docker logs <worker> --since 12h | grep ingest_compile` ๊ฐ€ **0** ์„ ๋ƒˆ๋‹ค. ๊ฒฐํ•จ์œผ๋กœ ์ฝ์„ ๋ป”ํ–ˆ๋Š”๋ฐ, ์›์ธ์€ **๋ฐฐํฌ๊ฐ€ ์ปจํ…Œ์ด๋„ˆ๋ฅผ ์žฌ์ƒ์„ฑ**ํ•ด์„œ ๊ทธ ์ด์ „ ๋กœ๊ทธ๊ฐ€ ์•„์˜ˆ ์—†๋Š” ๊ฒƒ์ด์—ˆ๋‹ค. ์ปจํ…Œ์ด๋„ˆ ๋กœ๊ทธ๋กœ "์ผ์–ด๋‚œ ์  ์—†๋‹ค"๋ฅผ ์ฃผ์žฅํ•˜๋ ค๋ฉด **์ปจํ…Œ์ด๋„ˆ ์‹œ์ž‘ ์‹œ๊ฐ๋ถ€ํ„ฐ์˜ ์ฐฝ**๋งŒ ์œ ํšจํ•˜๋‹ค. ```bash docker inspect -f '{{.State.StartedAt}}' <container> # ์ด ์‹œ๊ฐ ์ดํ›„๋งŒ ๊ทผ๊ฑฐ๊ฐ€ ๋œ๋‹ค ``` ### ๋˜ ํ•˜๋‚˜ โ€” ๋กœ๊ทธ๊ฐ€ ์—†๋‹ค๊ณ  ๊ฒฝ๋กœ๊ฐ€ ์•ˆ ๋„๋Š” ๊ฒŒ ์•„๋‹ˆ๋‹ค `ingest_compile_batch_complete` ๊ฐ€ ์•ˆ ๋ณด์—ฌ์„œ "settle ์ด ์ปดํŒŒ์ผ๋Ÿฌ๋ฅผ ์•ˆ ํƒ„๋‹ค"๋กœ ๊ฒฐ๋ก ๋‚ผ ๋ป”ํ–ˆ๋‹ค. ์ฝ”๋“œ๋ฅผ ์ฝ์œผ๋‹ˆ settle ์€ `compile_batch` ๊ฐ€ ์•„๋‹ˆ๋ผ `extract_entity_names` ๋ฅผ ํƒ€๋Š”๋ฐ, **๊ทธ ๋ฉ”์„œ๋“œ๋„ ๊ฐ™์€ `_find_related` ๋ฅผ ๋ถ€๋ฅธ๋‹ค** โ€” ๋กœ๊ทธ๋งŒ ์•ˆ ๋‚จ๊ธธ ๋ฟ ๊ฒฝ๋กœ๋Š” ๋ˆ๋‹ค. **๊ด€์ธก ์žฅ์น˜์˜ ๋ถ€์žฌ๋ฅผ ๋™์ž‘์˜ ๋ถ€์žฌ๋กœ ์ฝ์ง€ ๋งˆ๋ผ.** ### ์‹คํ—˜ ํŒ(arm)์ด **์ „๋ถ€ ๋˜‘๊ฐ™์ด** 0 ์„ ๋‚ด๋ฉด ์ฒ˜์น˜๊ฐ€ ์•„๋‹ˆ๋ผ ํ•˜๋„ค์Šค๋ฅผ ์˜์‹ฌํ•˜๋ผ BSVibe 2026-08-26. ํ”„๋กฌํ”„ํŠธ ๋ฒ„์ „์„ ๋ณ€์ˆ˜๋กœ A/B ๋ฅผ ๋Œ๋ ธ๋‹ค โ€” arm 4๊ฐœ ร— ์ž…๋ ฅ 3๊ฑด = 12์…€. **์ „๋ถ€ ๋ช…๋ น 0๊ฐœ**๊ฐ€ ๋‚˜์™”๋‹ค. "์˜› ํ”„๋กฌํ”„ํŠธ๋Š” ์ด ๊ฒ€์‚ฌ๋ฅผ ๋‚ผ ๋Šฅ๋ ฅ์ด ์—†๋‹ค"๋กœ ์ฝํž ์ˆ˜ ์žˆ๋Š” ๋ชจ์–‘์ด์—ˆ๊ณ , ์‹ค์ œ๋กœ ๊ทธ๋ ‡๊ฒŒ ์ ์„ ๋ป”ํ–ˆ๋‹ค. ์›์ธ์€ ์ฒ˜์น˜๊ฐ€ ์•„๋‹ˆ๋ผ **๋ฐฐ์„ **์ด์—ˆ๋‹ค: ํ”„๋กœ๋ธŒ๊ฐ€ resolver ์— redis ๋ฅผ ์•ˆ ๋„˜๊ฒจ์„œ 12์…€ ์ „๋ถ€ `ExecutorAdapterUnavailable` ๋กœ ์ฃฝ์—ˆ๋‹ค. LLM ์ด ํ•œ ๋ฒˆ๋„ ํ˜ธ์ถœ๋˜์ง€ ์•Š์•˜๋‹ค. **๊ฐ๋ณ„ ๊ทœ์น™ โ€” ์ฒ˜์น˜ ํšจ๊ณผ๋Š” ๊ท ์ผํ•  ์ˆ˜ ์—†๋‹ค.** | ๊ด€์ธก | ์ฝ๋Š” ๋ฒ• | |---|---| | arm ๋งˆ๋‹ค 0 ์˜ **์ •๋„๊ฐ€ ๋‹ค๋ฅด๋‹ค** | ์ง„์งœ ์ฒ˜์น˜ ํšจ๊ณผ์ผ ์ˆ˜ ์žˆ๋‹ค | | **๋ชจ๋“  arm ์ด ์ •ํ™•ํžˆ ๊ฐ™์€ 0** | ํ•˜๋„ค์Šค/๋ฐฐ์„  ๊ณ ์žฅ. ์ฒ˜์น˜๋Š” ์•„์ง ์•ˆ ์žฌ๋ดค๋‹ค | | ํ†ต์ œ(control) arm ๊นŒ์ง€ 0 | ํ™•์ •์ ์œผ๋กœ ํ•˜๋„ค์Šค ๊ณ ์žฅ โ€” ํ†ต์ œ๋Š” ๋˜๋˜ ๊ฒƒ์ด์–ด์•ผ ํ•œ๋‹ค | **How to apply** - ์‹คํ—˜ ๊ฒฐ๊ณผ์— **์˜ค๋ฅ˜ ํ•„๋“œ๋ฅผ ๋‚จ๊ธฐ๊ณ  ์ง‘๊ณ„์— ํ•จ๊ป˜ ์ถœ๋ ฅํ•˜๋ผ**. `error=None` ์ธ ์…€ ์ˆ˜๋ฅผ ๋จผ์ € ๋ณด๊ณ  ๋‚˜์„œ ํšจ๊ณผ๋ฅผ ๋ด๋ผ. ์˜ˆ์™ธ๋ฅผ ์‚ผ์ผœ 0 ์œผ๋กœ ๋งŒ๋“ค๋ฉด ๊ณ ์žฅ์ด ๋ฐœ๊ฒฌ์œผ๋กœ ์œ„์žฅํ•œ๋‹ค. - **์•Œ๋ ค์ง„-์–‘์„ฑ arm ์„ ๋ฐ˜๋“œ์‹œ ํ•˜๋‚˜ ๋„ฃ์–ด๋ผ.** ์ง€๊ธˆ ์ž˜ ๋˜๋Š” ์กฐ๊ฑด์ด 0 ์„ ๋‚ด๋ฉด ๊ทธ ํŒ์€ ํ†ต์งธ๋กœ ๋ฒ„๋ฆฐ๋‹ค. - n=1 ์Šค๋ชจํฌ๋ฅผ ๋จผ์ € ๋Œ๋ ค **๋ฐฐ์„ ์„ ํ™•์ธ**ํ•˜๊ณ  ๋‚˜์„œ ๋ณธ ์‹คํ–‰์— ๋น„์šฉ์„ ์จ๋ผ. - ๊ด€๋ จ: [[ab-on-a-few-items-cannot-recover-a-population-rate]] โ€” ํ•˜๋„ค์Šค๊ฐ€ ๋ฉ€์ฉกํ•ด๋„ **๋„๊ตฌ ์ž์ฒด๊ฐ€ ์งˆ๋ฌธ์— ์•ˆ ๋งž์„** ์ˆ˜ ์žˆ๋‹ค(๋ชจ์ง‘๋‹จ ๋น„์œจ์„ ๋ช‡ ๊ฐœ ํ•ญ๋ชฉ์œผ๋กœ ๋ฌผ์„ ๋•Œ).
Ver en GitHub