Experiments
+ New experimentSuggestions · 6
Experiment hypotheses from Claude Opus. Each one is a falsifiable test — run it and the verdict engine tells you whether it held.
Drop post cadence to 1 post per day maximum for two weeks
We hypothesize that reducing posting frequency to a maximum of 1 post per day (from the current implied higher cadence) will increase top_decile_reach_multiple by at least 25%, because concentrating audience attention on a single daily post should consolidate engagement signals rather than splitting them across multiple lower-performing posts, which — on Threads, inferred from Meta's content-quality ranking emphasis — may improve per-post distribution scores.
Why · Top_decile_reach_multiple at 14.2x indicates your best posts already break out significantly; the hypothesis is that the floor is being dragged down by volume, and constraining to one post per day forces curation that could raise the average closer to the ceiling — this is a Threads-specific inference and should be treated as a hypothesis to falsify, not a prescription.
Publish recruitment friction posts on the English-speaking accountant search
We hypothesize that posts documenting specific friction points in hiring bilingual accountants for remote BPO roles (candidate drop-off rates, salary anchoring, language test failures) will increase reply_rate_per_view by at least 25%, because this topic is structurally absent from AI-adjacent Threads accounts and will attract a distinct audience segment — operators, HR leads, and Indonesian business owners — who reply with their own experiences.
Why · Your protect list and double_down list both flag recruitment posts as a real-business anchor that differentiates you from pure AI commentary accounts; testing whether this topic drives reply rates validates whether the audience overlap between BPO operators and your AI-agent followers is large enough to reward continued posting in this lane.
Reply to every commenter within 30 minutes on AI agent posts
We hypothesize that responding to all commenters within 30 minutes on posts tagged to AI agents and swarm automation will increase reach_rate by at least 25% over the 14-day window, because early reply velocity is inferred by Threads' ranking signals (consistent with Meta's stated engagement-weighting logic) to indicate a high-quality conversation thread worth distributing further.
Why · Reach_rate at 1.67 is the most actionable distribution lever right now; Meta's open-sourced ranking work on Instagram/Threads suggests early comment velocity influences feed amplification (this is an inference for Threads specifically — treat as hypothesis, not fact), and your conversational posting style means replies are already happening, just potentially too slowly to catch the distribution window.
Test OVERRATED vs UNDERRATED format on model+workflow combos
We hypothesize that applying the OVERRATED / UNDERRATED binary frame to specific model-plus-workflow combinations (e.g., 'Claude Code + swarm: OVERRATED for X, UNDERRATED for Y') will increase reply_to_like_ratio by at least 25%, because the oppositional structure forces readers to pick a side and defend it, converting passive likes into active replies.
Why · Your reply_to_like_ratio is already a strong 0.498, suggesting your audience leans conversational; your double_down list names the OVERRATED/UNDERRATED format on niche combos explicitly, so this experiment tests whether formalizing that frame into the hook itself pushes the ratio even higher.
Post mid-migration confession threads with before/after tool names
We hypothesize that posts structured as 'was using X, switched to Y because Z (with specific failure reason)' will reduce zero_reply_fraction by at least 25% relative to baseline, because operational switching stories create a natural question surface — readers want to know if the switch held, what broke, and whether their own setup is at risk.
Why · Zero_reply_fraction at 16.7% means roughly 1 in 6 posts gets no engagement; your double_down list specifically calls out mid-migration confessions with named tools as your most distinctive content type, suggesting this format reliably opens conversation loops that generic model commentary does not.
Lead with exact token/cost figures in the first line
We hypothesize that opening posts with a specific numeric data point (e.g., token counts, cost figures, quota burn percentages) in the first line will increase reply_rate_per_view by at least 25% compared to posts where the figure appears mid-thread or not at all, because concrete numbers create an immediate credibility hook that invites correction, comparison, and conversation.
Why · Your protect list explicitly names token counts and cost figures as credibility anchors, and your double_down list calls out exact terminal counts and quota burn as unique data nobody else has. Starting with the number rather than burying it should front-load the curiosity trigger that drives replies.
Active · 0
No active experiments.
Completed · 0
No completed experiments yet.