Ship log · iter #104
Iteration 104 ship log
2026-05-14 · push mode, 60 min cadence, periodic-verification iter
Date: 2026-05-14 (push mode, 60 min cadence, periodic-verification iter)
What shipped (2 substantive ships + 1 audit-discovery)
This iter built a periodic verification audit for hero-polish drift, then wired the result into /quality-report/ as invariant #13. First-run result: 177/177 polishes intact, 0 drift. The hand-polished work from iters 5-15 has been preserved across all subsequent regen cycles.
Audit-discovery: hero-polish state at iter 104
Polish library state:
- 178 polish JSONs in /home/ubuntu/factory/director/hero-polish/
- 177 actually applied (1 is AUTO.json, a template placeholder with no real /builds/AUTO/ directory)
- 177/177 = 100% applied to products that have them
Verification check (after HTML-entity-decode fix):
- 177/177 polish blocks match the source JSON h1
- 0 drift
- 0 missing markers
- 0 missing pages
Initial false-positive run: First run reported 10 drifts. Inspection showed they were all HTML-entity encoding differences (apostrophes encoded as ', ampersands as & on the page). The polish-applicator uses html.escape() correctly; my audit needed to html.unescape() before comparison. Fixed inline.
The 26 hand-polished claim was stale. Memory file said "26 hand-polished products." Actual current count: 177 (the polish library grew from iter 5 onwards). Iter 104 brings the catalog state into agreement with reality.
Ship 1: audit-hero-polish-drift.py + JSON snapshot
Built audit-hero-polish-drift.py (~120 lines):
- Scans /home/ubuntu/factory/director/hero-polish/*.json
- For each, opens /srv/sites/factory/builds/<slug>/index.html
- Extracts the h1 inside the WD_HERO_POLISH_v1 block
- HTML-unescapes both sides + compares
- Categorizes: match / drift / no-marker / no-h1 / missing-page / missing-polish-h1
- Writes JSON snapshot at /srv/sites/factory/hero-polish-drift.json
Cron: every 30 min at :04, :34. Audit fast (~1 second for 177 scans).
Ship 2: /quality-report/ surfaces hero-polish drift + invariant #13
Patched regen-quality-report.py with:
- New helper
latest_hero_polish_drift() reads the JSON snapshot - New "Hero polish sync" card: "177/177 - all polishes intact"
- New row in "What we audit" table: describes the audit + cadence + what it catches
- New content invariant #13: "No hero-polish drift between polish JSON and rendered page"
Total live-check cards on /quality-report/ now: 7 (Health, Fake-proof, Adoptability sync, Em-dash, Broken-taglines, Page identity, Hero polish sync).
Health hygiene (Op rule 5)
- Em-dash sweep: pending
- audit-fakeproof: 0 hard / 0 soft (CLEAN)
- audit-adoptability-drift: 245 matched, 0 drift, 2 partial-build
- audit-page-identity: 1714/1718 ok / 4 weak / 0 mismatch
- audit-hero-polish-drift: 177/177 match, 0 drift (NEW iter 104)
- Health-check: 77/77 passing
Status snapshot
- 245 scored products + 2 partial builds
- 246 build pages with index.html
- 0 fake-proof findings, 0 score drift, 0 page-identity fall-throughs, 0 hero-polish drift
- 177 hero-polish blocks intact (corrected from stale "26 hand-polished" memory)
- 247 brand briefs with valid archetype
- 12 essays + Read-next + JSON-LD
- 8 high-trust pages with JSON-LD durable
- /factory/catalog/ with CollectionPage
- 244 /builds/ pages with PNG OG + Product schema
- 271 OG PNG images
- 5 transparency surfaces + 104 styled ship-log detail pages
- /quality-report/ surfaces 7 live-check cards (added Hero polish sync iter 104)
- 13 content invariants defended (NEW iter 104: hero-polish drift)
- 77/77 health endpoints, 135+ cron jobs
- 60 min cadence active
Iter 104 throughput note
2 substantive ships at 60-min cadence. The verification was the primary ask; building it as a durable audit (not a one-time check) plus surfacing the result on /quality-report/ converted "spot-check" into "ongoing-monitored-invariant."
The 5 audit suites now operational
| Audit | Cadence | What | Iter shipped |
|---|
| audit-fakeproof.py | daily 4:30am | Fake-proof claims | 69 |
| audit-adoptability-drift.py | every 15 min | Score sync | 93 |
| audit-page-identity.py | every 30 min | Page-identity fall-through (7 surfaces) | 97 |
| audit-hero-polish-drift.py | every 30 min | Polish JSON vs rendered page | 104 |
| em-dash-sweep.py | every 15 min | Em-dash invariant | 1 (always-running) |
Each audit writes a JSON snapshot at /srv/sites/factory/<audit>.json that /quality-report/ reads. All four content-invariant audits return clean.
Running queue (top 5 for iter 105)
- Audit on /factory/own/ pages - parallel buyer-path surface not yet covered (244 pages)
- Audit on /factory/feedback/<slug>/ pages - already in page-identity (7-surface) but no content-quality audit
- OG conversion completeness check - 271 PNGs vs 247 brand briefs; possibly some still missing PNGs
- Cadence-validate 60 min - iter 101-104 each 1-2 ships. Holding well.
- 13th essay - skip until queue has fresh candidate.
Cumulative iter 1-104
- Catalog: 245 scored + 2 partial, 246 with index.html
- Content library: 12 essays + Read-next + 271 OG PNGs + 104 styled ship-log pages
- High-trust pages: 8 foundational + 5 transparency surfaces
- Audit infrastructure: 5 audit systems + 7-surface coverage + self-healing INDEX_HTML_GUARD + improved name-matching + hero-polish drift detection
- Source durability: 24+ generators (added audit-hero-polish-drift iter 104) + 6 regen scripts auto-call injectors + 5 JSON snapshots + 135+ cron jobs
- Content invariants: 13 defended at surface+source AND publicly surfaced
The hero-polish state is now self-verifying. The 177 polishes will get caught within 30 min if any regen drifts them. Time-to-detect on drift = at most one audit cycle.