The Dead
Documentation
Graveyard

A memorial for docs that were dead on arrival. Browse the graves. Mourn the wasted sprint points. Submit your own.

Enter the graveyard

Here lie docs that nobody reads

Real examples of documentation found in production dbt projects. Names changed. Pain preserved.

"This column contains the status"
- name: status description: "The status of the record" tests: - not_null
"TODO: add description"
- name: dim_customers description: "TODO: add description" columns: - name: customer_id description: "TODO" - name: segment description: "TODO" - name: ltv_score description: "TODO"
"Ask Brian, he knows"
-- The logic for churn calculation -- is complicated. Ask Brian in -- the revenue team for details. -- -- Update: Brian left in Q3 2024.
"Updated: 2021-03-14"
## Data Dictionary Last updated: 2021-03-14 | Column | Type | Description | |----------|--------|-----------------| | amount | FLOAT | The amount | | type | STRING | 1=good, 2=bad | Note: We added 15 new columns since this was written.
"Don't touch this"
-- DO NOT MODIFY THIS QUERY -- I don't know why it works but -- if you change anything the -- numbers will be wrong. -- Trust me. I spent 3 days on this. -- -- - Someone, probably 2022
"The auto-generated masterpiece"
- name: crt_dt description: "Created date" - name: upd_dt description: "Updated date" - name: crt_by description: "Created by" - name: del_flg description: "Delete flag"
"The model that documented itself"
- name: stg_payments__v2_new_final description: "Staging model for payments. Replaces stg_payments__v2_new which replaced stg_payments__v2 which replaced stg_payments." # All four models are still running

Now AI agents write dead docs too

Everyone's building AI agents for data engineering. Nobody's asking why those agents produce the same dead documentation humans do.

"AI-generated, human-ignored"
- name: dim_customers description: > This table stores customer data. customer_id is a unique identifier for each customer. created_at is the timestamp when the record was created. updated_at is the timestamp when the record was last updated. # 200 words. Zero business context.
"The agent that confidently lied"
- name: status description: > Customer lifecycle status. 1 = Active, 2 = Churned, 3 = Suspended, 4 = VIP # Actual values: 1-17. # Nobody knows what 5-17 mean. # The agent guessed from column name.
"Generated by Copilot, reviewed by nobody"
- name: fct_orders description: "Orders fact table" - name: fct_payments description: "Payments fact table" - name: fct_shipments description: "Shipments fact table" - name: fct_returns description: "Returns fact table" # 47 models. Same pattern. Ship it.
"Your AI agent has no memory"
Session 1: "status=3 means churned" Session 2: "what does status=3 mean?" Session 3: "status=3 means churned" Session 4: "what does status=3 mean?" Session 5: "status=3 means churned" # Every session starts from zero. # The agent learns nothing.

Submit a documentation grave

Got dead docs in your codebase? Anonymize them and submit. The worst examples get immortalized here.

Why does documentation die?

Because it's written once and forgotten forever. Because it describes what but never why. Because the person who understood the business logic left six months ago.

Every dead doc is a symptom of the same disease: documentation is treated as a deliverable, not as living knowledge.

0
Graves submitted
0
Causes of death
0
% had "TODO"
0
Docs resurrected

We're building something that keeps documentation alive

An AI assistant that builds persistent business context for your data team. It doesn't just write docs — it remembers why things are the way they are.

No spam. One email when we launch. Unsubscribe anytime. Privacy Policy