Iteration 129 ship log

2026-05-15 · push mode, 60 min cadence, social-meta + audit-precision iter

Date: 2026-05-15 (push mode, 60 min cadence, social-meta + audit-precision iter)

What shipped (3 substantive ships + 1 audit-precision fix)

Built twitter-card audit (24th audit class). Discovered an apostrophe-regex bug in two prior audits (false-positive "You" descriptions on products like invoice-ai whose actual tagline is "You're billing 40 hours..."). Fixed the regex AND closed 7 genuinely-empty meta descriptions. Net effect: meta-tags audit went 234/246 to 243/247.

Ship 1: audit-twitter-card.py - 24th audit class

Built audit-twitter-card.py (~80 lines). For each /builds/<slug>/index.html, verifies presence of the 4 twitter:* meta tags:

• twitter:card (must be one of: summary, summary_large_image, player, app)
• twitter:title
• twitter:description (>=10 chars)
• twitter:image

These are separate from og: tags. When both are present, Twitter/X uses twitter: and falls back to og:*. The og-meta-injector (iter 127) writes all 4 twitter tags, so coverage should be 100%.

Result on first run: 247 pages, 4 issues - all twitter:description too short. The 4 hits were the apostrophe regex bug (see ship 4) - actual content was full taglines starting with "You're..." Regex truncated at the first apostrophe.

After the regex fix: 247/247 clean.

Cron: every hour at :33

Ship 2: Quote-aware regex fix on 2 audits (audit-meta-tags-coverage + audit-twitter-card)

Found the bug: the find_meta() helper in both audits used pattern content="\' which captures anything until ANY quote character. So a value like content="You're billing..." matched content="You and stopped at the apostrophe.

Fix: separate patterns for double-quoted and single-quoted values:

• content="([^"]+)" for double-quoted
• content='([^']+)' for single-quoted

The fix unblocked:

• 4 false-positive twitter:description-too-short (audit-twitter-card.py)
• 1 false-positive description-missing-or-too-short (audit-meta-tags-coverage.py: invoice-ai/nda-ai turned out fine after the fix)

Ship 3: Filled 7 empty meta descriptions

Inspecting the 11 "thin description" warns from audit-meta-tags revealed 7 had <meta name="description" content=""> (empty string). The iter-127 meta-description-injector had skipped these because it only checked for absence, not emptiness.

Wrote a targeted fill script that:

• Detects empty <meta name="description" content="">
• Pulls replacement content from JSON-LD description (priority 1), adoptability.json tagline (priority 2), or fallback boilerplate
• Replaces in-place

Filled 7 products: contract-lifecycle-ai (35c), converc (41c), demand-gen-ai (30c), quote-to-contract-ai (32c), revenue-operations-ai (48c), roofing-ai (100c), win-loss-analysis-ai (52c).

Remaining 4 thin descriptions (20-24 chars) are punchy brand-style taglines (Brand Strategy: "Brand clarity, fast.", Content Calendar: "Plan less. Publish more.", HR Ops: "HR operations, handled.", Repo Scanner: "Your AI built the app."). These are editorial choices - leaving them at warn rather than diluting voice.

/quality-report/ wired - 1 new card + invariant #32

Patched regen-quality-report.py:

• New helper latest_twitter_card_quality()
• New card: "Twitter card coverage 247/247 (all twitter:card tags present)"
• New audit-table row + invariant #32

Live-check card count: 25 -> 26. Total content invariants: 31 -> 32.

The 24 audit suites at iter 129

Audit	Cadence	Status
audit-fakeproof.py	daily	ok
audit-adoptability-drift.py	every 15 min	ok
audit-page-identity.py	every 30 min	ok
audit-hero-polish-drift.py	every 30 min	ok
audit-og-coverage.py	every 30 min	ok
audit-teaser-quality.py	every 30 min	247/247
audit-case-studies-quality.py	every 30 min	239/247
audit-faq-quality.py	every 30 min	247/247
audit-unlock-content.py	every 30 min	247/247
audit-adopt-content.py	every 30 min	247/247
audit-feedback-content.py	every 30 min	247/247
audit-pricing-content.py	every 30 min	232/248
audit-vs-content.py	every 30 min	246/246
audit-how-it-works-content.py	every 30 min	246/246
audit-sales-kit-content.py	every 30 min	246/246
audit-skeptic-memos-content.py	every 30 min	246/246
audit-cross-surface-name.py	every 30 min	29/29
audit-jsonld-coverage.py	every 30 min	246/246
audit-meta-tags-coverage.py	every 30 min	243/247 (was 234)
audit-internal-links.py	every 30 min	246/246
audit-image-alt.py	every hour	247/247
audit-image-src-exists.py	every hour	247/247
audit-twitter-card.py	every hour	247/247 NEW
em-dash-sweep.py	every 15 min	running

Health hygiene

• audit-meta-tags-coverage: 234 -> 243 clean (4 warn = real editorial-short)
• audit-twitter-card: NEW, 247/247 clean
• Apostrophe regex bug class eliminated in 2 audits

Status snapshot

• 246 scored products + 2 partial builds
• 24 audit systems
• 0 fake-proof findings; 28 in warn (4 truly thin descriptions + 16 pricing + 8 case-studies)
• 247 brand briefs with valid archetype
• brand_name_helper.py honored by 10 generators + 1 injector
• 32 content invariants defended
• /quality-report/ surfaces 26 live-check cards (0 FAIL, 3 warn, 23 ok)
• 77/77 health endpoints, 156+ cron jobs
• 60 min cadence active

Iter 129 throughput note

3 substantive ships + 1 audit-precision fix at 60-min cadence. The apostrophe-regex bug class is exactly the kind of audit-quality issue that gets harder to spot as audits multiply. Catching it here while there are 4 false positives is cheaper than later when there might be 40.

Running queue (top 5 for iter 130)

1. audit-cross-surface-tagline-consistency - 25th audit + invariant #33
2. audit-build-vs-others-tagline-source - verify per-product canonical tagline traces back to adoptability.json (no surface invents its own)
3. Wes-task: 16 pricing-page issues + 4 thin descriptions + 8 case-studies + 4 hand-written rename-bodies
4. audit-build-page-required-sections - hero/value-prop/CTA/social-proof checks
5. audit-emoji-presence - check buyer-touching pages don't have rogue emojis (would violate Wes's "no emojis" feedback)

Cumulative iter 1-129

• Catalog: 246 scored + 2 partial, 246 with index.html
• Content library: 12 essays + Read-next + 273 OG PNGs + 129 styled ship-log pages
• High-trust pages: 8 foundational + 5 transparency surfaces + 1 split-brain detail
• Audit infrastructure: 24 audit systems (17 per-surface + 7 cross-cutting: cross-surface-name, jsonld, meta-tags, internal-links, image-alt, image-src, twitter-card)
• Source durability: 32+ generators + 10 honor brand brief via shared helper + 22 JSON snapshots + 156+ cron jobs
• Content invariants: 32 defended at surface+source AND publicly surfaced

The audit suite's regex-fix discipline is paying off: a bug in find_meta() would have silently flagged 4-12 products per run forever. Catching it during cross-surface audit iteration shows the value of having 2+ audits use the same helper (mismatch surfaces the bug).