TBM Brain
Searchable index of ~440 posts from "The Beautiful Mess" newsletter
(2020-2026). Three-tier retrieval keeps token usage low.
Retrieval flow
Tier 1: Route via clusters
Read data/clusters.json (~600 tokens).
Each entry has id, label, and post_ids.
Match the user's query to 1-3 clusters by label.
Cluster quick reference:
| ID |
Label |
| 4 |
product management, discovery, experimentation |
| 16 |
product ops, org effectiveness, agile delivery |
| 14 |
product operating model, org transformation, SVPG |
| 1 |
roadmaps, planning, convergence, playbooks |
| 10 |
org design, leadership, accountability |
| 2 |
communication, psychological safety, trust |
| 5 |
WIP limits, dependencies, capacity, flow |
| 3 |
organizational behavior, systems thinking |
| 7 |
change management, continuous improvement |
| 19 |
strategy execution, enabling constraints |
| 17 |
metrics, measurement, analytics |
| 11 |
prioritization, strategy, sequencing |
| 15 |
strategy, alignment, decisiveness |
| 20 |
overthinking, simplification, recovery |
| 8 |
problem framing, decision making, assumptions |
| 12 |
goals, OKRs, north star metrics |
| 0 |
operating models, shared language |
| 22 |
frameworks, mental models |
| 9 |
north star, workshops, facilitation |
| 18 |
AI, context trap, glue work, overload |
| 6 |
narrative, complexity, change framing |
Tier 2: Scan cluster detail
Read data/clusters/<id>.jsonl for each matched cluster.
Each line is one post with title, date, url, questions,
search_terms, and summary.
Match the user's intent against questions and search_terms.
Return the best 3-5 posts.
Tier 3: Related posts
If the user wants to explore further, read
data/content-graph.json.
Look up the post ID to find nearest neighbors (ranked by similarity
score). Load neighbors from their cluster files as needed.
Response format
- Lead with the most relevant post and a one-sentence reason it matches.
- Always include the Substack URL.
- Group multiple results logically, not by date.
- Search multiple clusters when the query spans topics.
- Say so if nothing matches well; do not force weak results.
Example:
TBM 399: 10 Prioritization Traps
https://cutlefish.substack.com/p/tbm-399-10-prioritization-traps
Covers the most common anti-patterns teams hit when prioritizing.
Related: TBM 381, TBM 374
Freshness
Read data/version.json for built_at and
post_count. If the data is more than 30 days old, mention the
index may not include the latest posts.
1---2name: tbm-brain3description: Search and explore John Cutler's "The Beautiful Mess" (TBM) newsletter archive (~440 posts). Covers product management, org design, strategy, prioritization, delivery, metrics, leadership, and systems thinking. Use when the user mentions TBM, John Cutler, "Beautiful Mess," or asks about product strategy, roadmaps, OKRs, discovery, team health, prioritization, operating models, or organizational change.4license: MIT5---67# TBM Brain89Searchable index of ~440 posts from "The Beautiful Mess" newsletter10(2020-2026). Three-tier retrieval keeps token usage low.1112## Retrieval flow1314### Tier 1: Route via clusters1516Read [data/clusters.json](data/clusters.json) (~600 tokens).17Each entry has `id`, `label`, and `post_ids`.18Match the user's query to 1-3 clusters by label.1920Cluster quick reference:2122| ID | Label |23|----|-------|24| 4 | product management, discovery, experimentation |25| 16 | product ops, org effectiveness, agile delivery |26| 14 | product operating model, org transformation, SVPG |27| 1 | roadmaps, planning, convergence, playbooks |28| 10 | org design, leadership, accountability |29| 2 | communication, psychological safety, trust |30| 5 | WIP limits, dependencies, capacity, flow |31| 3 | organizational behavior, systems thinking |32| 7 | change management, continuous improvement |33| 19 | strategy execution, enabling constraints |34| 17 | metrics, measurement, analytics |35| 11 | prioritization, strategy, sequencing |36| 15 | strategy, alignment, decisiveness |37| 20 | overthinking, simplification, recovery |38| 8 | problem framing, decision making, assumptions |39| 12 | goals, OKRs, north star metrics |40| 0 | operating models, shared language |41| 22 | frameworks, mental models |42| 9 | north star, workshops, facilitation |43| 18 | AI, context trap, glue work, overload |44| 6 | narrative, complexity, change framing |4546### Tier 2: Scan cluster detail4748Read `data/clusters/<id>.jsonl` for each matched cluster.49Each line is one post with `title`, `date`, `url`, `questions`,50`search_terms`, and `summary`.5152Match the user's intent against `questions` and `search_terms`.53Return the best 3-5 posts.5455### Tier 3: Related posts5657If the user wants to explore further, read58[data/content-graph.json](data/content-graph.json).59Look up the post ID to find nearest neighbors (ranked by similarity60score). Load neighbors from their cluster files as needed.6162## Response format6364- Lead with the most relevant post and a one-sentence reason it matches.65- Always include the Substack URL.66- Group multiple results logically, not by date.67- Search multiple clusters when the query spans topics.68- Say so if nothing matches well; do not force weak results.6970Example:7172> **TBM 399: 10 Prioritization Traps**73> https://cutlefish.substack.com/p/tbm-399-10-prioritization-traps74> Covers the most common anti-patterns teams hit when prioritizing.75>76> **Related:** TBM 381, TBM 3747778## Freshness7980Read [data/version.json](data/version.json) for `built_at` and81`post_count`. If the data is more than 30 days old, mention the82index may not include the latest posts.
johnpcutler