Reviving Praderas (Day 24) — Phase 5 covers and agent-facing search

We resumed the agreed plan after a short break: keep Tier A retrofit at roughly two ES/EN pairs per day and advance Phase 6 without a second server or breaking Pico’s flat-file model.

Wall clock (order of magnitude)

• Retrofit Days 8 and 9 (Comfy + WebP + --translation-key): ~1 h (two SDXL passes + front matter audit).
• Phase 6 v1.1 (search.json + listing fields): ~40 min agent time.
• This ES/EN log: same session.

Block A — Tier A retrofit: Days 8 and 9

Rows 8 and 9 in .agents/retrofit-cover-queue.md still used Picsum: the Phase 5 multilingual announcement and the post that formalizes the batch migration plan. Both are core series chapters; they deserved the same treatment as Days 1–7.

Decision: one WebP file per ES/EN pair (identical Image: in both .md files), consistent with the rest of the retrofit. We did not reuse another day’s cover: each Translation_Key gets its own raster under assets/images/.

Reproduction (Day 8)

python3 scripts/comfyui/export_cover.py \
  --output assets/images/day08-comfyui-sdxl-phase5-multilingual-hero.png \
  --positive "Wide cinematic editorial illustration for a Spanish tech blog named Praderas, soft golden meadow light, abstract paired language glyphs and gentle bridge shapes suggesting bilingual content without readable text, subtle hreflang link motifs as soft curves, warm grass-green and neutral tones, professional quiet atmosphere, no logos, no watermarks, high detail, tasteful color grading" \
  --seed 08052026 \
  --prefix praderas_day08_export \
  --webp --webp-delete-png \
  --translation-key praderas-phase-5-multilingual

Reproduction (Day 9)

python3 scripts/comfyui/export_cover.py \
  --output assets/images/day09-comfyui-sdxl-translation-migration-batch1-hero.png \
  --positive "Wide cinematic editorial illustration for a Spanish tech blog named Praderas, soft dawn meadow light, abstract stacked document lanes and gentle batch queue shapes suggesting translation migration in orderly waves, subtle parallel paths converging without readable text, calm grass-green accents on warm neutrals, professional quiet atmosphere, no logos, no watermarks, high detail, tasteful color grading" \
  --seed 09052026 \
  --prefix praderas_day09_export \
  --webp --webp-delete-png \
  --translation-key praderas-day-9-translation-migration-batch-1

Queue tick: rows 8–9 → done. Next usual target: Days 10–11 (~two pairs on the next working day).

Block B — Phase 6 v1.1: JSON search and agent fields

On Day 23 we shipped /blog.json and siblings at schema 1.0. Today we move to 1.1 with two items from .agents/phase-5-6-plan.md:

Why a plugin, not a microservice

Same rationale as v1: 70-BlogJson.php handles dedicated routes in Pico’s request cycle, reuses loaded pages, and delegates ranking to 40-PicoSearch.php via searchBlogPosts(). Agents get the same relevance order as humans on /search/<term> without scraping HTML or running a second deploy.

What shipped

estimated_tokens is a rough budget (ceil(strlen(body) / 4)) for RAG context windows — not a tokenizer from the model that consumes the JSON.

Out of scope (v1.2): public /for-ai-agents page, tag filters on JSON, JSON URLs in the sitemap.

Quick check (local stack)

curl -sS 'http://localhost:8080/search.json?q=multilingual' | jq '.meta'
curl -sS 'http://localhost:8080/blog.json' | jq '.posts[0] | {title, word_count, estimated_tokens, modified_at}'

Full contract: .agents/blog-json-api.md.

What’s next

• Tier A — rows 10–16 (~2 pairs/day).
• Phase 6 v1.2 — /for-ai-agents discovery page for tools that do not read .agents/ in git.
• This log’s cover — generated at PR close (day24-comfyui-sdxl-tier-a-search-json-hero.webp, seed 24052026).