GCP Enhancement Candidates

Status

held

Tier

Tier 2 — Platform

Owner

Ryan Colston

Started

2026-04-26

One-line description. Verified-but-unbuilt GCP capabilities to surface when an adjacent problem appears. Don't pitch unsolicited.

Why

I audited the actual GCP footprint on 2026-04-26 (525 APIs scanned, 50 Cloud Run services, 40 schedulers, 12 Pub/Sub topics, 26 BQ datasets).

That audit produced a real list of what's truly unused vs. what I assumed.

This file exists so future-me has a ranked menu when a problem shows up that one of these solves better than what I'd build from scratch. The discipline is: don't volunteer the whole list, surface the right one when the moment is right.

Current state

Held.

Parked after the Cloud Run audit on 2026-04-26. The candidates are catalogued in three tiers (direct fits, operational improvements, speculative). Confirmed dismissals are also captured so I don't waste time re-evaluating them.

What pulls a candidate forward: a project context that matches its trigger. Examples:

• Dataform → a BQ view breaks during 1x1 prep
• Vertex AI Embeddings + Vector Search → I manually search notes or ask "have I written about this"
• Speech-to-Text + Vertex AI on call recordings → see Custom Dialer
• Maps Platform suite → buyer-side work or neighborhood content
• Search Console + GA Data + Indexing API → SEO measurement on Ghost / market reports

Tier 1 has 5 candidates, Tier 2 has 15, Tier 3 has 11. Full list lives in the idea-inbox doc.

• Status: held
• Last update: 2026-04-26
• Blocked on: nothing. Held by design.

Next 3 actions

(none — surface when relevant)

Decisions log

• Portfolio organization — vault README is the front door, every project gets a hub from the template

Locked at audit:

• LLM for voice-sensitive content is Claude (Sonnet 4.6 / Opus 4.7). Not Gemini Flash. Flash is fine for classification and extraction. Wrong for voice-sensitive prose.
• Confirmed dismissals (don't recommend, even adjacent): Apigee, Cloud Workflows, Cloud Composer, Memorystore, Cloud Spanner, Cloud Armor, GKE/Anthos, ad-tech APIs, Healthcare/Retail/Blockchain APIs, AutoML, Earth Engine.

Open issues

• Verified gap: GCS lifecycle missing on pospj-480915-courted-screenshots and pospj-480915-file-archive. Probably small dollars. Worth setting on the next GCS cleanup pass.
• Cost finding (2026-07-06, verified from billing export): the Cloud Build staging bucket pospj-480915_cloudbuild costs ~$33.60/mo, of which ~$24 is the US-multi-region transfer/replication tax. The 3-day delete lifecycle is already in place, so retention is NOT the problem — it's the multi-region bucket taxing every source-archive write (~27 GB/day of deploy uploads). Recoverable by regionalizing the build staging bucket, NOT a lifecycle change. Conditional on confirming Cloud Build can point at a regional staging bucket without breaking deploys (there are no build triggers — deploys are gcloud run deploy/builds submit from CI + dev sessions, so the staging bucket is chosen per-invocation). ~$12/mo confidently recoverable (the replication SKU dies on single-region); up to ~$24 if the transfer SKU also drops.

References

• Idea-Inbox doc: ~/CCPJ/projects/Idea-Inbox/idea/2026-04-26-gcp-enhancement-candidates.md
• Related: Custom Dialer (Speech-to-Text + Vertex AI), Market Report Lead Conversion (Maps + RESO + reCAPTCHA), External API Platforms (the non-GCP companion catalog), Appointment Wallet Pass (Wallet Objects API)