Iteration 40 ship log

2026-05-11 · continued autonomous /loop

Date: 2026-05-11 (continued autonomous /loop)

What shipped

5 items. Key insight: switching em-dash discipline from "patch each generator" (whack-a-mole) to "universal safety net" (15-min sweep cron).

NEW: universal em-dash sweep cron

After 15 iterations of patching individual generators with _no_dash wraps, em-dashes kept reappearing when new generators were discovered. The whack-a-mole pattern was working but never definitively closed.

Shipped a structurally different solution: em-dash-sweep.py that walks the entire /srv/sites/factory/ HTML tree every 15 minutes and strips any em/en dashes it finds. Idempotent. 0.2s execution time across 2600+ files.

Cron: /15 *

This means: worst-case em-dash exposure for any buyer is now 15 minutes. A buyer browsing the catalog will essentially never see one.

Combined with the existing 15 source-level patches, this creates a defense-in-depth:

• Layer 1: 15 generators+scripts strip dashes at write time
• Layer 2: 3 injectors strip dashes on output
• Layer 3: Universal sweep cron catches anything that slips through within 15 minutes

The trade-off: we lose the ability to catch the bug at the source. But for em-dashes specifically (zero semantic value, never legitimate), this is fine.

Initial sweep

41 em-dashes had re-accumulated in 13 files since iter 39 (admin pages, top-level index files, log/index.html). Single sweep cleared all of them. Factory-wide count: 0.

SESSION-MASTER v6

Replaced v5 (covered iters 1-36) with v6 covering iters 1-40 (~32 hours of autonomous work). New section: "Em-dash invariant defense in depth" documenting all 3 layers.

Files changed inventory

New

• /home/ubuntu/factory/director/em-dash-sweep.py (universal sweep, 30 lines)

Modified

• /srv/sites/factory/log/SESSION-MASTER.md (v5 -> v6)

Re-rendered / cleaned

• 13 HTML files (41 em-dashes stripped)

New cron

• /15 * em-dash-sweep.py

Why this matters

15 iterations of source-level patches taught us: every generator that touches upstream content is a potential em-dash leak. The autonomous studio will keep adding generators. Each new one is a potential regression vector.

The universal sweep is the right shape: instead of fighting every leak, accept that leaks happen and clean them quickly. 15-minute cadence is fast enough that no realistic buyer session encounters an em-dash.

This pattern is generalizable: any "must never be in output" rule could be enforced via a similar sweep. Future iterations could add similar sweeps for:

• Stale dates
• Empty alt-text on visible images
• Missing JSON-LD on product pages
• Etc.

Status snapshot

• 200+ products in catalog (Director still ticking)
• 62 buyer-facing pages
• 59/59 health endpoints passing
• Sitemap: 1379 URLs
• 0 em-dashes across 2600+ HTML files (verified)
• Em-dash defense in depth: 15 source patches + 3 injectors + universal 15-min sweep
• 11 broken links (admin/log only)

What still needs work

Same Wes-side as always:

1. Stripe wiring (30 min)
2. Email-send for auto-fulfill
3. First real traffic push

Next 5 recommended autonomous tasks

1. Hero-polish second pass - Claude CLI on bottom-50 Adoptability products.
2. Per-product /how-it-works/ sub-page - lightweight new content surface from implementation_plan + tech_stack.
3. Catalog card hover preview - hover shows first line of dossier teaser.
4. Run promote-artifacts.py to fill more /about/ pages on new products.
5. Audit /factory/today/, /factory/hotlist/, /factory/founders/ for stale content.

Cumulative iter 1-40

The factory is genuinely production-ready as far as autonomous polish can take it:

Surface quality:

• 200+ products, 62 buyer-facing pages, 10 audience landings
• Per-product objection handlers + trust signals + previews + per-product preview teasers
• Money-back guarantee at every CTA
• Honest disclosure + graduated proof page + adoptability methodology
• 0 em-dashes across all 2600+ HTML files

Durability:

• 15 generator+script source patches
• 3 injectors enforcing on output
• Universal 15-min sweep cron as safety net
• Weekly .bak cleanup cron
• 59 health endpoints, all passing
• 11 broken links (admin-only)

Monitoring:

• Director + watchdog alive
• ~104 cron entries
• Real-time intent capture + Telegram alerts (with internal-test filter)
• Admin morning briefing + inbox + captures + health + link audit dashboards

The product is structurally ready to charge $5. Remaining work is exclusively Wes-side: Stripe + traffic push.