Research report · 29 October 2026
One A/B-test report, two local social-draft routes
A maintained template and a manually source-guided route each produced a LinkedIn draft and a four-post X thread from the same published report. Both needed corrections in an unblinded author review. This one exercise does not show a fidelity or reviewer-effort advantage for either route.
Fixed input and method
The sole substantive input was the 28 October A/B-test preflight report, not a customer article or live experiment. Both routes used the same audience, plain voice, report link as the only call to action, no performance promise, one LinkedIn text and four X posts. Each material claim was checked against a report passage.
The four-post format, passage trace, exact link, local-only output and separate review were proposed controls for this exercise. They are not requirements established for Social Content. Its post templates supplied format ideas only; no story, result or engagement prompt was filled with invented material. Scheduling, account monitoring and remote drafts were excluded.
What self-review corrected
| Initial wording or omission | Correction | Why it matters |
|---|---|---|
| Template route shortened “20% relative lift” to “20% lift.” | Restored relative. | A 1% baseline with 20% relative lift means 1.2%, not a 20-point increase. |
| Both routes used the approximate 43,000-per-arm illustration without its assumptions. | Added 1% baseline, 20% relative lift, two-sided α=.05 and 80% power. | This is a rounded source-table example, not a sample target, duration or feasibility result for a real page. |
| Source-guided copy called units “participants.” | Removed the label. | The independent assignment unit for a real page was not defined. |
| Source-guided X copy said “don’t peek” without a design qualifier. | Limited the caution to unplanned interim p-value looks in a fixed-horizon design. | A predeclared safety or sequential rule is a separate case. |
| Template wording implied a winner could be shipped. | Changed to deciding whether to ship or stop. | The source report contains no experiment result or winner. |
One further wording edit avoided implying the source table was verified for every use. These corrections were made by the author, who knew which route produced each draft.
Result and limits
After correction, the author's passage-level check found no remaining unsupported material claim, invented outcome, endorsement, customer context or extra link. All eight reviewed X texts met this exercise's under-280-character target; that is not a check of current X policy or live display. Both routes retained worksheet-and-analyst-first language and separate implementation, inference and action decisions. The source-guided route made the numerical example and multiple decision owners more prominent; the template included those facts after review too. Prominence here is an observation, not a measured quality gain.
This is one artifact, two manually written draft pairs and an unblinded self-review. Route choice and author expectations may have shaped the wording and corrections. No independent reviewer, reviewer-time measure, preference study, skill execution, account action, reach or engagement observation was available. A route-masked review of multiple authorized artifacts could test whether the guided route reduces claim corrections or review effort; this exercise cannot answer that question.
Source and provenance boundary
On 29 October, the mutable Social Content selected file displayed version 2.4.0. Only the brief, hooks/anti-bait and repurposing/scheduling passages, plus the directly used post-template reference, were revisited for this exercise. A crawl label is not an immutable source revision. No selected-file SHA/date, complete package or chosen-host permissions were established. The published report link was present in the local research collection, but live resolution of its canonical URL was not verified.
For any later skill comparison, first review an exact payload and effective host access, then use independent reviewers and multiple authorized sources. These are proposed next steps, not evidence of skill quality. The earlier Social Content source audit explains why account and scheduling actions stay outside the local-draft route.
Research library · Four conditional candidates · Original selected file