Research report · 16 October 2026
Copywriting: what changed in the selected source?
The mutable Copywriting source displayed 2.3.1 on 16 October. It displayed 2.2.0 in a separate 15 October observation. The two obtainable complete files support a 2.1.0→2.3.1 endpoint comparison, not an exact full 2.1.0→2.2.0 delta.
What was actually compared
| Source point | Observation | What it establishes |
|---|---|---|
| 2 October file-attributable commit | Full 2.1.0 file, 260 lines; broker marked it crawled last week. | Historical full-text baseline. The commit page shows the selected file changed, not that this is the latest main modifier. |
| 15 October observation | Mutable raw main displayed 2.2.0; only decision passages were retained. | Dated version observation. Its complete file is not available here, so the exact full 2.1.0→2.2.0 delta is partial. |
16 October raw main | Full selected text displayed 2.3.1, 297 lines; broker marked it crawled today. | Current observed endpoint, not a pinned latest selected-file SHA/date. A GitHub search result still indexed 2.2.0 at a two-day crawl age, but opening the page displayed 2.3.1. That is a crawl-time discrepancy, not two pinned current versions. |
No history or API endpoint was inspected. No package, account, brief, CLI, host or page was run.
Six-area full-endpoint comparison
Each area separates the full 2.1.0 baseline and observed 2.3.1 endpoint from the decision boundary. The complete intermediate 2.2.0 file was not retained.
Context and diagnosis
Source comparison: Both ask for page action, audience, offer/proof and traffic context, and point to three possible local product-marketing context paths. The observed 2.3.1 text also requests a context file’s Learnings (lines 13–14) and distinguishes blank brief, real-voice draft and narrow point edit (36–50).
Decision boundary: A source read instruction does not authorize a local file read. Exact data, file access and host permissions remain separately controlled.
Claims and proof
Source comparison: Both forbid invented statistics/testimonials and mark proof gaps. The observed 2.3.1 text adds line-level picturability/falsifiability/differentiation checks (54–59) and instructions to name an unsupported claim’s gap, closest supported alternative and evidence needed (82–89). The 15 October observation already contained the supportability/decline rule; its full surrounding section is unavailable.
Decision boundary: Source instructions are not the pool’s claim/proof register. They do not require approved use rights, denominators, qualifiers, exact CTA URL, named reviewer or signoff. The source’s conversion figures and cited 3× example had no checkable underlying study in the inspected passages of Run 22.
AI-tells
Source comparison: Both direct a pre-handoff ai-tells.md check; observed 2.3.1 additionally points full rewrites toward a build-a-writer subsection and second-generation-tells check (132).
Decision boundary: The intermediate timing of that pointer is unknown. Here it is optional style review, not a validated AI detector, buyer-response measure or conversion result.
Output
Source comparison: The old file describes sectioned page copy with annotations and headline/CTA options (229–250). Observed 2.3.1 keeps these for page drafting, adds unit-by-unit drafting and a leave-it-alone review (51–63), but makes a narrow point edit changed lines plus one-line reasons only (265).
Decision boundary: Describe output as page-copy draft or narrow edit depending on input. Do not promise a full page and alternatives for each point edit. The intermediate 2.2.0 output section was not retained, so this exception cannot be dated to that version.
Linked resources
Source comparison: Both name copy-frameworks.md, ai-tells.md and natural-transitions.md. Observed 2.3.1 attributes its diagnosis/drafting technique to Profound’s external improved-copywriting skill in skills-marketing-library (63).
Decision boundary: Attribution is not an install instruction; that external skill was not inspected. Named links do not prove complete directory, ZIP/CLI payload or host resolution.
Publication/account action
Source comparison: Neither complete selected file has a CMS publish, account connection, deployment or send step. Their selected-file outputs are textual copy or edits and related-skill pointers.
Decision boundary: A narrow text observation, not package-safety proof. Rendered-page review and separate publication approval are pool controls, not source-enforced gates.
How to use this finding
- Set an authorized inputUse an approved non-personal brief or provided draft. Do not infer permission from the source’s context-file instruction.
- Keep the output in scopeA blank brief may lead to a page draft; a point edit asks for changed lines and brief reasons. Neither is a quality result.
- Hold unsupported claimsRecord each material claim, evidence, qualifiers and rights. Hold numbers, testimonials, comparisons, guarantees and CTA links until a named marketing/brand/legal reviewer verifies them.
- Decide publication separatelyReview the rendered page, then obtain an explicit publication decision. Source copy directions do not clear that gate.
Skill catalogue · Review a skill · Dated 2.2.0 observation · Full 2.1.0 source · Mutable source observed as 2.3.1