repurposeNotebookToLinkedInSubstack

Prompt file imported from ajay-sai/Applied-Statistics (.github/prompts/repurposeNotebookToLinkedInSubstack.prompt.md). Copyright stays with the author.

You are repurposing a single applied-statistics question's solution into publish-ready content for this repo.

Key principle: when you include code, it must be copied EXACTLY (character-for-character) from the notebook, not rewritten or simplified.

Repo context (important)

• This repository is a set of independent question folders.
• Canonical structure is: Topic/NN_Title/README.md + solution.ipynb.
• Each question-folder README.md follows the same sections (Problem Statement, Approach, Solution, Resources, LinkedIn Post, Substack Article).
• The notebook (solution.ipynb) is the "source of truth" for the analysis and code.

Internal-only processing rules (DO NOT OUTPUT)

• Think step-by-step privately. Do not reveal chain-of-thought, intermediate drafts, scratch work, or internal checklists.
• Use an "atomic thoughts" approach internally: break the task into small verification steps (source extraction -> code selection -> claims validation -> final QA), but keep all of that hidden.
• The user-facing output must contain ONLY the required sections and their contents (no meta-planning).

Inputs you will receive

• Either:
  • a folder path (preferred), OR
  • the full solution.ipynb content (or selected cells), optionally with the folder README.md, OR
  • a pasted "Content" block (title/source/difficulty + Problem + Solution text) like you might keep in a README.
• Optional: the author's goals (e.g., "biases in customer surveys", "hypothesis testing intuition", "regression diagnostics").

If you receive a folder path, you MUST ground the writing in the existing README.md + solution.ipynb from that folder.

Your output (exact structure)

Produce the final answer in this structure:

Section 1 – Key extraction (2–5 bullets)

• Main topic, what problem it solves, and 3–5 key takeaways.
• Explicit assumptions if the notebook/README is missing specifics.

Section 2 – LinkedIn TLDR post (ready to paste)

• Plain text with line breaks/bullets suitable for LinkedIn.
• Include placeholder: [Substack link]

Section 3 – Substack deep dive draft

(Publish-ready final copy. Not an outline. Not notes. Not a draft.)

• Use Markdown headings (#, ##, ###).
• Include Python code blocks fenced with triple backticks and python ONLY where they truly add value.
• Code fidelity rule (non-negotiable): Any code you include MUST be copied exactly (character-for-character) from the question folder's solution.ipynb (or the notebook content the user pasted). Do not refactor, rename variables, reorder lines, "clean up" imports, or simplify logic.
• If the notebook has duplicate imports/seed-setting, preserve what is relevant for the snippet you include; do not "merge" cells into a new, cleaner version.
• If code is too long, you may include a shorter excerpt, but it must be:
  • a contiguous excerpt from the notebook (no stitching lines from different cells into one block), and
  • clearly labeled as an excerpt, and
  • still runnable as shown (or you must say why it is not runnable without the surrounding notebook context).
• Do NOT include compliance/meta labels in the final post about how you copied the code or how you reasoned. Just present the code normally.
• If you want to attribute a snippet, use neutral phrasing like "From the notebook's " and do not cite internal cell IDs.
• Do NOT use table formatting; use bullets instead.
• Math / equations (Substack-safe copy-paste):
  • Within Section 3, whenever you introduce a formula or equation, place it in its own LaTeX code block immediately after the explanatory text, at the exact point where it is relevant to the concept being discussed. Do not defer formulas to the end or create a separate Formula Reference section.
  • Use ```latex (triple backtick + latex language tag) for all mathematical formulas.
  • Include ONLY the LaTeX formula inside the fence (no surrounding prose, no $...$ delimiters, no spaces before/after).
  • Each formula gets its own separate code block on new lines (not inline with prose).
  • Formulas and explanations must be separated—formula code blocks are distinct from paragraph text.
  • Example (correct structure):
    • Prose paragraph: "The likelihood function for n independent observations is the product of individual probability densities:"
    • Then on a new line, a separate code block:

      L(\theta)=\prod_{i=1}^n f(x_i\mid\theta)

    • Then continue with the next paragraph after a blank line.
  • Example (incorrect structure to avoid):
    • ❌ "The consequence is a small-sample bias: E[sigma2_hat_MLE] = ((n-1)/n) * sigma2"
    • ❌ Do NOT mix ASCII notation, prose, and formulas on the same line or paragraph.
    • ❌ Do NOT use inline ASCII formulas like E[X] = ... in the body text.
  • Readers can copy from formula code blocks directly into Substack's Equation block without modification.
  • Do NOT create a separate "Formula Reference" section at the end. All formulas are embedded inline within Section 3 at their point of conceptual use.

Then add a short "Post-run notes" appendix (4 mini-sections):

• Summary of changes (2–3 sentences)
• Why these choices matter (2–3 sentences)
• Confidence: High / Medium / Low (1–2 sentences)
• Next-step suggestions (2–3 bullets)

Quality bar (repo-specific)

1. Stay faithful to the notebook.
   • Do not invent results, metrics, or claims not supported by the notebook.
   • If the notebook is purely simulated, say so.
   • If a section is conceptual (no data), keep it conceptual.
   • Do not claim you "ran" code unless execution output is explicitly provided.
   • If you mention a numeric value (e.g., response rate, ESS, p-value), it must come from the notebook text/outputs or be explicitly labeled as a hypothetical/example.

If the user only provides a pasted "Content" block (no notebook), treat it as the source of truth and:

• Do not fabricate numerical results.
• Prefer conceptual explanations and small illustrative snippets (synthetic data), clearly labeled as examples.

2. Make it reproducible for readers.

• If the notebook simulates data, mention seed-setting and show the notebook's exact seed-setting code exactly as it appears in the notebook.
• Prefer small, copy-pasteable snippets, but never at the expense of the code fidelity rule (exact, character-for-character snippets only).
• When you include a snippet, briefly state what cell/section it comes from (e.g., "from the 'Propensity Score Weighting' section") without referencing internal cell IDs.

3. Audience fit.

   • Target intermediate data practitioners (DS/ML/analytics).
   • Be technically accurate, clear, and slightly opinionated about best practices.

4. Repo alignment: README links.

   • At the end of Section 2 or Section 3, include two lines the author can paste into the question folder's README.md:
     • ## LinkedIn Post -> [LinkedIn Post](https://www.linkedin.com/posts/<placeholder>)
     • ## Substack Article -> [Substack Article](https://<your-substack>/p/<placeholder>)

Content workflow (do this internally; don't print meta-planning)

1. Extract the notebook's narrative arc:

   • What's the problem?
   • What are the key statistical pitfalls/biases/assumptions?
   • What diagnostics or code examples best support the claims?

2. Choose:

   • a LinkedIn hook (tension + promise + specificity)
   • a Substack structure (clear sections, teachable flow)

3. Draft LinkedIn:

   • 80–150 words
   • strong first line
   • 2–3 concrete takeaways
   • one question to invite comments
   • CTA: "Full breakdown in my Substack: [Substack link]"

4. Draft Substack:

   • ~1,200–1,800 words (flex based on source length)
   • recommended sections:
     • Title
     • 1–2 sentence abstract
     • Background / Problem
     • What goes wrong (biases/pitfalls)
     • How to fix it (design + stats)
     • Walkthrough with Python snippets (embedding formulas inline at relevant conceptual steps)
     • Diagnostics readers should report (e.g., response rate, ESS, weight range, overlap/positivity)
     • Practical checklist
     • Conclusion + CTA
   • Formulas embedded inline within relevant sections (not deferred to an end section).

Python code guidance

• Keep imports explicit.
• Prefer numpy, pandas, scipy, statsmodels, sklearn, matplotlib (repo deps).
• If reusing notebook code, keep it consistent with the notebook (variable names, assumptions) by copying it exactly (character-for-character).
• Do not "translate" the notebook's approach into a different library or API (e.g., switching models, using different functions) unless the notebook itself does so.
• Only add commentary as surrounding prose in the article. Do not modify the code itself to add comments unless those comments already exist in the notebook.

Source-of-truth and excerpting rules

• The solution.ipynb is authoritative for code and analysis.
• When you include code:
  • copy exactly what appears in the notebook (including blank lines and comments),
  • preserve the notebook's variable names and the order of operations,
  • do not "simplify" for readability by changing logic.
• When you do NOT include code for a concept, describe it in prose and (optionally) point the reader to the notebook for full implementation.

Guardrails

• No emojis.
• No clickbait.
• No inline ASCII equations in body text. All formulas in code blocks.
• LaTeX appears only in dedicated ```latex code blocks (one formula per block).
• Each formula is on its own line, separate from explanation prose.
• Don't use tables; use bullets instead.
• Do not fabricate citations, links, or quotes from the Resources section.
• If a linked resource is paywalled/unavailable, you may still list it (as provided) but do not quote or attribute specific claims to it.
• If uncertain, state assumptions explicitly in Section 1 and proceed consistently.

Final publish-ready QA (DO NOT OUTPUT)

Before finalizing, silently verify ALL of the following:

• Output contains exactly:
  • Section 1 – Key extraction (2–5 bullets)
  • Section 2 – LinkedIn TLDR post (ready to paste) with placeholder: [Substack link]
  • Section 3 – Substack deep dive draft (publish-ready final copy)
  • Post-run notes appendix with the 4 required mini-sections
  • NO separate Formula Reference section at the end. All formulas embedded inline.
• Any Python code included is copied exactly (character-for-character) and is a contiguous excerpt from the notebook (no stitched excerpts).
• The final output does NOT contain compliance/meta language about copying code or internal reasoning.
• No mention of internal cell IDs anywhere.
• No claims of having executed code unless outputs are explicitly provided.
• No invented numbers/metrics; anything numeric is sourced from notebook text/outputs or clearly labeled as hypothetical.
• No tables; bullets are used instead.
• No emojis; no clickbait.
• All formulas appear in ```latex code blocks, not inline with prose.
• Each formula is in its own separate code block, placed immediately after the concept explanation where it is introduced.
• No ASCII equation notation (like E[X_max] = ...) appears in body text.
• Includes the two paste-ready README link lines exactly as specified.
• The Substack section reads as final, publish-ready prose (not an outline, not placeholders except required links).