Present to Kyaw Zin Oo and close the project

Step 1: Run the full pipeline end-to-end

Run the complete pipeline with several days of data, including one correction day. Verify the full chain:

• Watermark advances with each new day's data
• MERGE handles the correction re-load without creating duplicates
• Morning report numbers in fct_daily_operations are accurate
• Farmer tracking in dim_farmers includes all deliveries and advance payments
• No PII appears in files that should not contain it

This is the verification run before delivery. If something is wrong, it is better to find it now than to find it after Kyaw Zin Oo sees incorrect numbers in his morning report.

Step 2: Present to Kyaw Zin Oo

Message Kyaw Zin Oo with the morning report and a summary of what the pipeline does. He does not need to know about MERGE keys or watermark columns. He needs to know:

• Every morning, the system picks up new data from both mills automatically
• If a correction happens -- a supervisor discovers yesterday's numbers were wrong -- the fix is applied cleanly without creating duplicates or confusion
• The report shows both mills side by side: paddy received, grade distribution, yield, farmer deliveries

If you addressed the scope creep from Unit 5 (his daughter's request for year-over-year farmer comparison), include that in the presentation. If you scoped it for a future phase, say so clearly.

Step 3: Address final feedback

Kyaw Zin Oo may have follow-up questions. Answer them in terms he uses -- farmers, paddy, mills, grades -- not in pipeline terminology.

Step 4: Write the project README

Summarize what was built: the extraction strategy per source, the MERGE key design, the watermark monitoring approach, the hook-gated verification setup. How to run the pipeline. What to check if something looks wrong.

This is for the next engineer who works on this project. Or for you, three months from now, when you have forgotten the details.

Step 5: Final CLAUDE.md review

Open CLAUDE.md. Does it accurately describe the current state of the pipeline? Not the state from Unit 2 when you first wrote it -- the state now, after building the staging layer, the mart layer, the hooks, and the monitoring.

Check: extraction strategy per source. MERGE keys. Watermark monitoring configuration. Hook configuration (which hooks exist, what they catch, fail-closed behavior). Known quality issues (corrections, advance payments, format divergence). Data dictionary with unified field names.

A project memory file that has not been updated since the first session is worse than no project memory file -- it gives AI confident but outdated information.

Step 6: Push to GitHub

Commit everything. Verify the pre-commit hook passes with all tests green. Push to the remote repository.

The pipeline is live. Kyaw Zin Oo checks his morning report. The numbers are there. Corrections are handled. Two mills, one view, no confusion.