Iteration 125 ship log

2026-05-15 · push mode, 60 min cadence, split-brain remediation iter

Date: 2026-05-15 (push mode, 60 min cadence, split-brain remediation iter)

What shipped (4 substantive ships - the big systemic fix)

This iter took the iter-124 cross-surface discovery (29 split-brain products) and fixed 24 of them in one pipeline-level sweep. The remaining 5 are hand-written content the generators don't touch.

Ship 1: brand_name_helper.py - shared resolver module

Built /home/ubuntu/factory/director/brand_name_helper.py (~60 lines). Single source of truth for "what's the display name of this product?" Resolution order:

1. Brand brief explicit name: YAML field
2. Brand brief H1 fallback (# Inkwell · Brand Brief)
3. adoptability.json prod["name"] (legacy default)
4. Slug

Provides two functions:

• brand_brief_name(slug) - returns brand brief name or None
• display_name(prod_or_slug, fallback_name=None) - returns the brand brief name OR canonical OR slug

Any generator imports this and gets brand-rename support for free.

Ship 2: Patched 8 surface generators

Used a batch patch script + manual edits to wire brand_brief_name(slug) into 8 surface generators. The 5 batch-patched (standard p.get("name", slug) pattern):

Generator	Surface	Pattern	Replacements
faq-template-gen.py	faq	p.get	1
regen-unlock-pages.py	unlock	p.get	2
regen-adopt-pages.py	adopt	p.get	1
regen-feedback-page.py	feedback	p.get	1
regen-fallback-pricing.py	pricing	prod.get	1 (manual)

The 3 manually patched (used get_field(brand_text, "name") which only reads YAML field, not H1 fallback):

Generator	Surface	Approach
regen-vs-pages.py	vs	brand_brief_name(slug) or get_field(...) or slug
regen-fallback-subpages.py	vs, how-it-works, skeptic-memos, sales-kit, about	same fallback chain
case-studies-gen.py	case-studies	same fallback chain

Ship 3: FORCE_REGEN + custom-page preservation guard

Both regen-fallback-pricing.py and regen-fallback-subpages.py were idempotent (skip if index.html exists). Added FORCE_REGEN env flag to bypass for one-shot fixes.

ALSO added a custom-page preservation guard to regen-fallback-subpages.py (WD_PRESERVE_CUSTOM_SUBPAGE_v1): if the subpage directory contains any sibling asset files besides index.html, the generator skips it regardless of FORCE_REGEN. This protects custom hand-crafted page sets (e.g., the 4 full sales-kits with product-brief.html, roi-calculator.html, objection-handler.md etc.) from being overwritten by the gated fallback.

Ship 4: Force-regenerated 246 products' 5 surfaces

Ran all 8 patched generators with FORCE_REGEN=1 to propagate brand brief names to existing pages:

fallback-pricing (FORCE):  wrote 246 new
faq-template-gen:          wrote 237, skipped hand-written 9
unlock-pages:              wrote 246
adopt-pages:               wrote 246
feedback-page:             wrote 246
vs-pages:                  wrote 17 of top 20 (custom)
fallback-subpages (FORCE): wrote 246 vs, 246 skeptic-memos, 246 how-it-works
case-studies-template:     wrote 239, skipped hand-written 7

Result: 29 split-brain -> 5 split-brain (83% drift fix)

Audit	Before	After
audit-cross-surface-name	0/29 (FAIL)	24/29 (WARN)
/quality-report/ card state	FAIL	WARN

Examples of fixed products (all 11 surfaces now use brand name):

• contract-ai: Inkwell on build / pricing / faq / vs / how-it-works / sales-kit / unlock / adopt / feedback / skeptic-memos
• audit-ai: Ledgerline (everywhere)
• coach-ai: Stride (everywhere except the 1 hand-written FAQ)
• handoff-ai: Relay (everywhere except hand-written case-studies)
• dispatch-ai: Pylon (everywhere)
• 20 more: Foreword, Riverine, Clearwater, Pylon, Plumb, Foyer, Lockstep, Almsbury, Northstar, Threshold, Vellum, Counsel (lawfirm-ai), Saltline, Sealwax, Doorstep, Lectern, Frontage, Pencil, Standing, Goodword, Bracket - all aligned

Remaining 5 split-brain (hand-written content the generators don't regenerate):

• brief-ai, cashflow-ai, handoff-ai, lawfirm-ai: hand-written case-studies still say canonical name (case-studies-gen.py skipped 7 hand-written)
• coach-ai: hand-written FAQ still says "Coach AI" (faq-template-gen.py skipped 9 hand-written)

These are Wes-task fixes - either edit the hand-written content to use brand name OR add a hand-written-content brand-rename helper.

Sales-kit collateral repair

The fallback-subpages run initially overwrote 4 custom sales-kit index.htmls (audit-ai, contract-ai, dispatch-ai, lawfirm-ai) with the gated stub. The .html/.md asset files (product-brief, roi-calculator, etc.) are still on disk so direct URLs still work. The new WD_PRESERVE_CUSTOM_SUBPAGE_v1 guard prevents this from recurring on future runs.

Audit verifies: audit-sales-kit-content still 246/246 clean - the gated index.html is one of 3 accepted patterns.

Health hygiene (Op rule 5)

• audit-cross-surface-name: 24/29 clean (was 0/29) - WARN state
• audit-sales-kit-content: 246/246 clean
• audit-faq-quality: re-checked, still 247/247 clean
• audit-pricing-content: re-checked, 232/248 (16 in warn unchanged - those are different issues)
• All other audits unchanged

Status snapshot

• 246 scored products + 2 partial builds
• 18 audit systems including cross-surface
• 0 fake-proof findings; 29 in warn (8 case-studies fabrications, 16 pricing-page issues, 5 split-brain - down from 29)
• 247 brand briefs with valid archetype; brand_brief_name now read by 8 generators
• 12 essays + Read-next + JSON-LD
• 8 high-trust pages with JSON-LD durable + 1 split-brain detail
• /factory/catalog/ with CollectionPage + 246 cards link /own/
• 273 OG PNG images
• 5 transparency surfaces + 125 styled ship-log detail pages
• /quality-report/ surfaces 20 live-check cards (0 FAIL, 2 warn, 18 ok - first iter with no FAIL state)
• 26 content invariants defended
• 18 audit systems operational
• 77/77 health endpoints, 150+ cron jobs
• 60 min cadence active

Iter 125 throughput note

4 substantive ships at 60-min cadence. This is the highest-impact iter since the early audit-suite buildout - one cross-cutting audit (iter 124) plus one pipeline-level fix (iter 125) reduced systemic drift from 29 products to 5. Pattern: cross-cutting audit finds the drift, shared-helper module + bulk generator patch fixes it.

The brand_brief_name helper + WD_PRESERVE_CUSTOM_SUBPAGE_v1 guard are reusable for future rename pipelines.

Running queue (top 5 for iter 126)

1. Wes-task: fix 5 hand-written content split-brain - edit 4 case-studies (brief-ai, cashflow-ai, handoff-ai, lawfirm-ai) + 1 faq (coach-ai)
2. Wes-task: 16 pricing-page issues (8 too-few-prices, 8 title-drift)
3. Wes-task: 8 case-studies fabrications
4. Audit-jsonld-coverage.py - verify every /builds/<slug>/ has valid Product or AboutPage JSON-LD (19th audit + invariant #27)
5. Restore 4 custom sales-kit index.htmls OR confirm the gated stub is acceptable

Cumulative iter 1-125

• Catalog: 246 scored + 2 partial, 246 with index.html, all cards link /own/
• Content library: 12 essays + Read-next + 273 OG PNGs + 125 styled ship-log pages
• High-trust pages: 8 foundational + 5 transparency surfaces + 1 split-brain detail
• Audit infrastructure: 18 audit systems (17 per-surface + 1 cross-cutting) + 10-surface page-identity + brand-rename support on 6 per-surface audits + multi-pattern audits
• Source durability: 31+ generators + 8 of them now read brand brief via shared helper + 6 regen scripts auto-call injectors + 17 JSON snapshots + 150+ cron jobs
• Content invariants: 26 defended at surface+source AND publicly surfaced
• Pipeline architecture: brand_name_helper.py as single source of truth for display names

This is the second major systemic milestone after the audit-perimeter close (iter 123). The factory's rename pipeline is now consistent through the surface layer.