# Changelog

All notable changes to `tidb-gtm-brain` are documented here.

Format follows [Keep a Changelog](https://keepachangelog.com/en/1.0.0/). Version numbers follow [Semantic Versioning](https://semver.org/): MAJOR.MINOR.PATCH.

- **MAJOR** — structural reorganization or complete knowledge overhaul
- **MINOR** — new knowledge sections added, or significant fact updates (new tier names, new competitors, ICP shifts)
- **PATCH** — corrections, proof point updates, wording fixes

> **Update rule:** Never update `playbooks/messaging.md` without a `tidb-messaging-review` PASS. Never add a proof point without written approval for external use. Never add a metric without citing its source.

---

## [Unreleased]

---

## [beta v0.13] — 2026-07-24

### Added
- **`knowledge/product.md`** — `TiDB Cloud Premium` tier (Preview; enterprise workloads needing unlimited growth, advanced security, instant elasticity) — confirmed live on pingcap.com/tidb/cloud/, was missing from the tier table
- **`knowledge/product.md`** — new AI/Agentic proof points: **Kimi** ("millions of agent-created databases, provisioned in under a second"; 10M+ tenants; zero per-site idle cost) and **Plaud** (MySQL+S3 → TiDB Cloud; 10x QPS improvement; 2M+ users/170 countries; 13 clusters in production) — both confirmed live case studies
- **`knowledge/product.md`** — internal-roadmap-watch note under Agent State Stack: flags an in-progress, not-yet-public internal proposal (tidb_family_knowledge, 2026-06/07 drafts) to reposition Starter as a broader "unified storage layer" ahead of a "TiDB Cloud FS" Public Preview (~Oct 2026), fold drive9 into a "TiDB Drive" capability, and stop new investment in mem9 as a standalone product. Explicitly not adopted into external-facing knowledge — names are working/TBD and unconfirmed live as of this update
- **`knowledge/icp.md`** — Segment 1 (AI Agent Builders): added Kimi and Plaud as validated public customer examples; updated Manus stat to "1M+ database tenants" with case-study link

### Changed
- Updated `Last updated` dates on `knowledge/product.md` and `knowledge/icp.md` to 2026-07-24 (re-verified against pingcap.com)
- **`brain/README.md`** — added explicit source-of-truth statement: pingcap.com (and subdomains) is the sole authoritative source for this knowledge layer; internal GTM drafts are signal, never citable fact until live
- **`knowledge/competitors.md`** — added **Redis** and **Qdrant** to the AI-agent-builder competitive set (confirmed on pingcap.com/compare/best-database-for-ai-agents/); flagged as needing a validated when-wins/when-loses section before use in detailed comparisons
- Ran the first pass of the new weekly `tidb-gtm-brain-sync` check manually (scheduled task's first real run hadn't fired yet) — spot-checked pingcap.com/customers/ (50+ case studies exist beyond the curated proof-point tables; intentionally not mass-imported, see Maintenance rule on curated proof points) and pingcap.com/compare/best-database-for-ai-agents/

---

## [beta v0.9] — 2026-04-16

### Added
- **`knowledge/product.md`** — official TiDB product positioning, tier definitions (Cloud Starter, Cloud Essential, Cloud Dedicated, Self-Managed), TiDB X engine note, validated customer metrics, capability claims with do/don't guidance
- **`knowledge/icp.md`** — 5 validated ICP segments with customer evidence and deciding language:
  1. AI Agent Builders (Manus, Dify, Rengage, Toprism)
  2. AI SaaS / Multi-Tenant Platform Builders (Atlassian, Pinterest, Conga)
  3. Scale-Out MySQL (LINE, Shopee, Rakuten, Flipkart)
  4. HTAP / ETL Elimination (Catalyst, Conga, ZTO Express)
  5. Fintech / High-Stakes OLTP (Plaid, Chainbase, SB Payment Service)
- **`knowledge/competitors.md`** — 10 competitive scenarios covering 15 products:
  - Vector databases: Pinecone, Weaviate, Milvus, Chroma
  - MongoDB Atlas
  - Amazon Aurora (MySQL / PostgreSQL) — 35-second multi-region failover, blocking DDL
  - CockroachDB — SERIALIZABLE isolation distinction
  - YugabyteDB — hash-based sharding default
  - Snowflake / ClickHouse
  - Neon (now part of Databricks) — vertical-only scaling, fragmented stack
  - PlanetScale — $3k/mo enterprise gate for auto-scaling
  - Supabase — BaaS distinction (Auth/Storage/APIs), not just a database
  - Siloed stack: RDS + ClickHouse + Elasticsearch
- **`playbooks/messaging.md`** — approved messaging snapshot: positioning, segment value props, competitive angles, AI integration messaging, proof points, banned phrases. Status: Draft (pending formal `tidb-messaging-review` PASS)
- **`playbooks/benchmark-standards.md`** — methodology requirements for OLTP, HTAP, and vector benchmarks; required disclosure fields; when NOT to publish rules
- `README.md` with quick start, two-repo system diagram, skills → brain mapping table, maintenance ownership guide
- `LICENSE` (MIT)
- `CHANGELOG.md` (this file)