Date: 2026-05-14 (push mode, 30-min cadence, polish + discoverability)
Small but high-leverage discoverability fix. The methodology page's FAQ on "What happens when a generator silently produces broken output?" already discusses the 15-min em-dash sweep cron. Added a callout linking to the new transparency page:
See it run live: /factory/cron-status/ shows all 124 cron jobs with last-run timestamps. The page regenerates every 15 minutes so the timestamps visibly shift if you reload.
A buyer reading the methodology page now has a one-click path to verify the autonomous-studio claim. The transparency page from iter 73 becomes discoverable.
Audit of /factory/catalog/ revealed it ALREADY has a comprehensive client-side search + filter + shortlist + compare system:
/ shortcut to focus (line 547)The full search + filter behavior is wired in a single inline <script> block (~50 lines) that hides cards based on combined filter state. This was built in earlier iterations.
Why this matters: the iter 73 ship log queued this as "build a small client-side search filter" without checking if one already existed. The audit-first pattern from iter 54 applies: check before polishing. Saved an iter by NOT rebuilding what existed.
Updated queue: /catalog/ search is feature-complete. The catalog navigation work is done. Next adjacent improvement is /playbooks/ filter (overkill for 9 essays) or per-product subpage navigation (e.g., between /faq/, /vs/, /how-it-works/) - both lower priority than per-product polish.
Audit on top 3 untouched 72-tier products found engagement-ai had a fake-customer claim in the hero lede:
Before:
Your customers stop coming back and you have no idea when it happened. Engagement AI keeps them connected through email, SMS, and in-app messages that work from one playbook. Businesses using it catch at-risk customers before they disappear, not after. Try it free for a week, no contract...
The bolded sentence implies "businesses are using it" - fake-customer claim. The other 2 audited products (sales-collaboration-ai, renewal-ai) were clean.
After:
Your customers stop coming back and you have no idea when it happened. Engagement AI keeps them connected through email, SMS, and in-app messages that work from one playbook. The honest design intent: flag at-risk customers from behavioral signals (drop in opens, no recent purchase, declining session frequency) before they disappear, instead of running a post-mortem after the cancel email. Try it free for a week, no contract...
The new framing is operator-honest about the DESIGN INTENT (we built it to do X) rather than claiming results from non-existent customers.
Fixed at source (placeholders.json) and re-rendered. The cron audit won't catch it because it's a vague phrasing, not a specific numeric claim - but the polish program catches what the regex audit misses.
Ran audit-fakeproof.py at start of iter:
/srv/sites/factory/methodology/index.html (cross-link to /cron-status/ added)/Users/wes/factory-templates/engagement-ai-placeholders.json (lede polished)/srv/sites/factory/builds/engagement-ai/index.htmlSmaller iter than recent ones because the catalog search task turned out to be already complete (saved an iter of work). 3 substantive ships still represents real progress, but the iter is roughly 60% the size of iter 73. That's the right behavior - the discovery WAS the work. Re-building existing features would have been worse value.
The factory is now in maintenance + polish mode. The big-bug arc is closed. Iteration value is incremental polish on remaining mid-tier products + Wes-blocker prep.