im-experiment
/im-experiment
Planuje lub ocenia eksperyment treściowy na Instagramie: hipotezę, warianty, metrykę, okno obserwacji i regułę decyzji.
- Autor
- mortired ↗
- Repozytorium
- instagram-agent-skill ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 01.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 1
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 01.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Planuje lub ocenia eksperyment treściowy na Instagramie: hipotezę, warianty, metrykę, okno obserwacji i regułę decyzji.
Wyzwalacze z opisu
Oryginalny opis (EN)
Plan or evaluate a focused Instagram content experiment: hypothesis, variants, comparison design, metric, observation window, and decision rule. Use for "test these hooks", "compare two covers", "does this CTA work", "design an experiment", or results from a planned test. Use im-audit for broad retrospective analysis of an account's past posts.
Jak działa
- Input
- Write the test card before results
- Record observations
- Evaluate
- Output and handoff
Podgląd SKILL.md
--- name: im-experiment description: >- Plan or evaluate a focused Instagram content experiment: hypothesis, variants, comparison design, metric, observation window, and decision rule. Use for "test these hooks", "compare two covers", "does this CTA work", "design an experiment", or results from a planned test. Use im-audit for broad retrospective analysis of an account's past posts. --- # im-experiment Turn a content assumption into a test the user can run and interpret. Write in the user's language. A test plan is useful even when no results exist yet. ## Input Use the user's question, available post insights, and relevant `strategy.md`, `experiments.md` or audit findings. Shared files may be attached or available under `~/.chatgpt/instagram/`. Ask for missing essentials together: - The decision the user wants to make and the alternatives they can produce. - Comparable past posts or a baseline, if available. - Which metrics are actually available, publishing capacity and review dates. If asked to evaluate a test, obtain its original plan and actual observations. If there was no plan, describe the analysis as exploratory and separate any new hypothesis from the result that suggested it. ## Write the test card before results 1. **Hypothesis.** State the changed element, expected outcome, audience and reason: "Changing X may improve Y because Z." Treat the reason as a hypothesis unless evidence already supports it. 2. **Variants.** Specify A and B precisely. Change one element when the goal is to learn its effect. If several elements change, call it a package comparison and do not attribute the result to any one component. 3. **Comparison.** Use genuine random assignment only if an available tool supports it. Ordinary posts reach different audiences at different times; compare repeated, reasonably matched posts and label the result directional. Record differences in topic, offer, format, distribution and post age. A before/after change to a profile also has time and audience confounds. 4. **Measurement.** Pick one primary metric tied to the decision and any necessary guardrail. Define numerator, denominator, data source and reporting window. A hook test needs an available retention measure; raw views alone cannot isolate the hook. If a metric is unavailable, narrow the question or select a usable measure before testing. 5. **Scope and stopping.** Set a feasible number of comparable posts, a common observation age for each, and a review date. Define a practically useful improvement and a stop condition before seeing results. Without enough baseline data to justify sample size, call this a pilot; no fixed small number of posts guarantees a reliable answer. 6. **Decision.** State when to adopt, retain the current approach or collect more evidence. Do not keep extending a test only to obtain a preferred result or declare a winner after one unusually successful post. Do not schedule posts or initiate paid tests. Script scores from `hookscore.py` or `detect.py` can help edit a draft but are not audience experiment outcomes. ## Record observations Use a stable experiment ID and retain its original plan. Collect: ```text post ID/URL | variant | published at/timezone | measured at | observation age primary numerator | primary denominator | guardrail | relevant differences ``` Separate raw counts from rates. Missing data stays missing; a zero denominator produces an undefined rate. Do not convert unavailable values to zero. Compare posts at the agreed age, and label immature observations as pending. ## Evaluate - Show per-post observations and the planned summary, including sample size. State how rates were combined; a mean of post rates and a pooled rate answer different questions. A viral post must not silently dominate the conclusion. - For rates, distinguish percentage-point change from relative percentage change. Relative change from a zero baseline is undefined. - Keep the unit of comparison clear. Thousands of views across two posts do not become thousands of independently randomised tests of a creative choice. - Report uncertainty, inconsistent results and plausible confounds. Do not claim statistical significance without an appropriate design, assumptions and a calculation that was actually performed. - Return a result of promising, no practical improvement, inconclusive or invalid comparison, with the evidence and a concrete next decision. Describe organic comparisons as evidence for a next test, not proof of causation. ## Output and handoff For planning, return the test card and a blank observation table. For review, return the filled table, calculations, limitations and recommendation. Preserve other experiment entries when updating `~/.chatgpt/instagram/experiments.md`; if file access is unavailable, return its updated contents for the user to save. Only mark a test as run when actual observations have been supplied. Pass planned variants and measurement timing to [im-plan](../im-plan/SKILL.md). Use [im-audit](../im-audit/SKILL.md) for wider patterns and [im-strategy](../im-strategy/SKILL.md) when the user wants to revise direction after reviewing the evidence. Use these skills only if available and relevant to the requested work. The user runs the publishing and supplies the results.
Wersje
W bazie nie ma innych skilli o nazwie im-experiment.
Migawka bazy z 13.11.2026 · wersja paczki c35244090fed (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 90create-experiment-design
Projektuje rzetelny test A/B lub eksperyment, na prośbę o walidację hipotezy. - 100gtm-experimentation
Projektuje test przed uruchomieniem: hipotezę, metrykę i regułę decyzyjną na tle bazowego lejka, na przykład brief testu A/B. - 90ab-test-setup
Planuje, projektuje i wdraża test A/B lub eksperyment, z hipotezą, wariantami i sposobem pomiaru wyników. - 95ab-test-setup
Planuje, projektuje i wdraża test A/B lub eksperyment, z hipotezą, wariantami i sposobem pomiaru wyników. - 90a-b-test-design
Projektuje eksperyment A/B: hipotezę, warianty, metrykę główną i wielkość próby, gdy zmianę można mierzyć ilościowo w dużej skali. - 95ab-test-setup
Pomaga planować, projektować i wdrażać testy A/B oraz eksperymenty, w tym hipotezy, warianty i istotność statystyczną.