Notes from production

Creastra Digest

• —1C pays back in 3 months if routine work runs 20+ hrs/wk
• —Skip 1C if the team is under 3 people or processes are chaotic
• —Ask the vendor about fixed-price, warranty and code ownership

Creastra Digest

• —47 artifacts span 6 blocks — legal, financial, ops, brand, training, tech
• —Below 35 of 47 you'll collapse within six months of scaling
• —Honest packaging starts at 450K RUB; cheaper offers are just a PDF deck

Creastra Digest

• —Edge cuts 40–60% TTFB for geo-spread audiences; 5–10% if you serve one country
• —Banned Node APIs: fs, child_process, native modules — rewrite or keep on Node
• —Cost: at 1M req/day, edge runs 18–35% pricier than Node

Creastra Digest

• —27 items in four loops: code, infrastructure, process, people
• —Full rollout takes six weeks and ~380,000 ₽ in infrastructure
• —Once in place, the client's audit cycle shrinks from 8 weeks to 2

Creastra Digest

• —Liquid Glass is a WebGL shader, not a backdrop-filter
• —60 fps on mid-range Android with proper optimization
• —+15-30% time on page, +8-18% leads — measured across 6 projects

Creastra Digest

• —3D drives +62% leads on complex, high-ticket products
• —Doesn't work on mass-market apparel or cosmetics — adds friction
• —Tech rules: glTF+Draco, KTX2, 2-3 LOD levels, photo fallback

Creastra Digest

• —Links in 2026 only work alongside behavioral signals
• —Top 12 factors: CTR, dwell, CWV, E-E-A-T, freshness, IKS, schema
• —Honest SEO budget — 80-150K RUB/mo; anyone selling top for 30K is lying

Creastra Digest

• —An agent = tools + memory + checklists; a prompt alone is not an agent
• —You spend 80% of effort on guardrails and observability, not the model
• —Quality measurement: a 50-case golden set, refreshed each sprint

Creastra Digest

• —View Transitions shed 70–120 KB of JS per page versus Framer Motion
• —Works on top of Next 15 useRouter without Suspense gymnastics
• —Safari < 18.2 — fall back to no animation; do not try to polyfill

Creastra Digest

• —The main culprit is endless requestAnimationFrame; switch to frameloop="demand"
• —Cap DPR at 1.5 instead of Math.min(window.devicePixelRatio, 2)
• —IntersectionObserver pauses out-of-viewport scenes — that alone takes 60% off

Creastra Digest

• —Exactly one primary CTA — in hero and footer
• —Social proof = a case with a number, not "trusted by"
• —Form ≤ 3 fields; email after, not before

Creastra Digest

• —Up to 2M vectors, pgvector with HNSW matches specialised stores on latency
• —pgvector's edge: JOINs with relational data in one query — Pinecone and Weaviate cannot
• —Above 5M vectors or with >5 metadata filters — move to Qdrant

Creastra Digest

• —Build LTV on gross margin, not revenue — in services that is a 40–60% gap
• —Include sales salaries and software in CAC or you understate it by ~35%
• —A healthy LTV/CAC band for services is 3.5–5; below 3 you lose money within 12 months

Creastra Digest

• —Concession + ops + brand + LMS — this is the baseline
• —Reporting and BI can wait, but no later than partner #3
• —Re-cut the financial model every six months — not once-and-done

Creastra Digest

• —The day runs on four blocks: artefacts, antagonists, lexicon, validation
• —An antagonist (a brand we deliberately do NOT want to resemble) yields 70% of the contrast
• —The guide passes when a copywriter can produce a social post from it in 12 minutes

Creastra Digest

• —A TMA pays back in 6–10 weeks; a native app in 8–14 months
• —TMA limits: offline, background tasks, heavy graphics, non-Telegram push
• —Payments via Telegram Stars or an in-app provider — legal and free of store fees

Creastra Digest

• —Seller unit economics splits into 11 lines — half of sellers track only four
• —Meeting two: a per-SKU «keep / raise / drop» decision model
• —The biggest margin killer is in-platform ads without a ROAS cap

Creastra Digest

• —Tokens live in @theme, not in JS
• —OKLCH/oklab — for color-mix() with stable lightness
• —Themes via CSS variables on :root, no recompile

Creastra Digest

• —The real EDO risk is not IT — it is partner-side buyers refusing. A 14-day parallel loop fixes it
• —We run a single adapter for Diadoc and SBIS, so switching operators does not stall work
• —Payback for a mid-sized manufacturer: 4–6 months

Creastra Digest

• —−60% production cost at comparable quality
• —×4 iterations — material and angle changes in minutes
• —AR Quick Look — partners also get .usdz at no extra cost

Creastra Digest

• —Ozon Seller API covers 90% of the routine: prices, stock, FBO/FBS via idempotent jobs
• —Redis queues plus Node workers beat cron scripts on a warehouse server
• —The main risk is rate limits: batch 1,000 SKUs per request and add jittered backoff

Creastra Digest

• —14 days split into 5 modules: operations, marketing, hiring, finance, crisis
• —Each module: async video + a 1-hour call + a graded homework
• —Worst anti-pattern: the «3-day intensive» — by day three the franchisee retains nothing

Creastra Digest

• —Four slots: Trigger, Internal conflict, Alternatives, «Good-enough» criterion
• —Each slot gets a verbatim customer quote, not a marketer's paraphrase
• —A segment = the «Alternatives» slot. The others are shades of the same fight

Creastra Digest

• —A bus in the middle, not a direct p2p call
• —Idempotent messages — a retry won't double an order
• —JSON-schema contract, versioned in git

Creastra Digest

• —Half the revisions trace back to unclear requirements, not the client's taste
• —An eight-point brief contract with references cuts taste revisions by 60%
• —Two checkpoint reviews with minutes kill «let's redo everything» in 90% of cases

Creastra Digest

• —Infographic mandatory from photo #2
• —A video review lifts CTR 12–18%
• —100% spec fill or you lose ranking

Creastra Digest

• —One tech lead per 5 engineers, no more
• —Weekly sync mandatory; long sync harmful
• —A "how we work" doc beats spoken rules

Creastra Digest

• —Resistance is a symptom, not a cause — 9 times out of 10 it is a clunky deal form
• —Tie commissions to CRM data from day one or sabotage is inevitable
• —A two-week «quiet period» with paired piloting cuts fear by 3×

Creastra Digest

• —T-shirt sizes for discovery, story points for build
• —Fixed price only if scope < 4 weeks and 100% clear
• —Reserve 25% for unknown unknowns

Creastra Digest

• —All artefacts in the cloud, not on a designer's laptop
• —One channel per project, not five chats
• —A cap on video calls — or you burn out in a month