Split a big job across several AI workers without getting mush back

I have a big job I want to split across several parallel AI workers instead of doing it in one long conversation.

The job: [WHAT YOU'RE TRYING TO PRODUCE]
What "done" looks like: [THE FINISHED ARTEFACT]
What I have to start with: [SOURCES, DATA, CONSTRAINTS]

Design the split before any work starts:

1. IS THIS ACTUALLY PARALLELISABLE? Say no if the answer is no. Work where step two needs step one's output is sequential, and forcing it into parallel produces confident nonsense. This is the most useful thing you can tell me.

2. THE SPLIT. Divide it into pieces that genuinely don't need each other. For each: what it gets, what it returns, and the exact instruction it runs. Each piece must be answerable without knowing what the others found.

3. THE SHARED CONTEXT. What every worker needs to know regardless — definitions, format, tone, constraints. Inconsistency between parallel outputs almost always traces back to something obvious that only one of them was told.

4. THE JOIN. How the pieces come together, and specifically what to do when two of them contradict each other. "Use judgement" is not a plan. Tell me which one wins and why, or that it needs a human.

5. THE VERIFICATION PASS. A separate step that checks the assembled result for gaps, duplication and contradictions. This is the step people skip, and it's the one that catches parallel work having drifted apart.

6. WHAT WILL GO WRONG. Given this split specifically, where does it break? Usually a piece that turns out to need another's output after all.

Then give me the exact prompt text for each worker and for the verification pass, ready to paste.