M6Lesson 4 · Finance Research / Trading Desk: Complex Multi-Agent Decision SupportM6-L4.2

Delivery Plan: Trading Desk

In one sentence: The Trade Recommendation Desk Delivery Plan fits the build into eight weeks for two engineers by breaking it into 35 tasks with checkable finish conditions, estimating 71 days of work against 80 available, and fixing the milestones that must happen in order.

Source document: Trade Recommendation Desk, Delivery Plan, Azure Partners, Sprint 1 final version. This page is a reading guide. It is the reference delivery plan for this module.

The shape of the plan

Element	What the plan says
Capacity	Two engineers, eight weeks, 40 hours a week: 80 working days
Work	35 tasks in eight areas, estimated at 71 days
Slack	9 days, a little over 10 percent
Largest tasks	Filings client (B1), run manager (D4) and publishing service (D6), four days each; every other task three days or fewer
Definition of done	A task is finished when its “finished when” column is true, not when the code is written

Eight areas

Area	Days	Engineer A	Engineer B
A. Setup	5	1	4
B. Data	11	9	2
C. Strategy and backtest	14	14	0
D. Agents and run pipeline	18	0	18
E. Forward trading and tracking	5	5	0
F. Frontend	5	0	5
G. Testing	9	4.5	4.5
H. Wrap-up	4	2.5	1.5
Total	71	36	35

Engineer A owns data, strategy and forward trading. Engineer B owns agents, the backend API and the frontend, and the machine that runs Puffo. Testing and wrap-up are shared, and Week 1 is worked together.

“Finished when” conditions

The task table’s last column is the plan’s most useful habit. Compare two ways of writing the same task:

Weak	As written in the plan
Build the holdout lock	No result exists without a log row, and the holdout refuses a second run
Set up Puffo	The daemon runs, two agents exchange messages, and one agent calls a test tool
Run manager	Runs move through their states, the Desk Lead is told what to post next, and a repeated wake changes nothing

Each condition is something another person can check without asking the owner.

Milestones, in order

When	Milestone	Why it sits here
End of Week 1	Puffo runs two agents; vendor, stocks, dates and costs recorded	The riskiest unknowns (platform, data licence, pool rule) are settled first
End of Week 3	A mentor approves the success thresholds	Before any validation result exists, so the bar cannot move
End of Week 5	A mentor checks the rule freeze; rules are tagged	Before the holdout, so it runs on frozen rules
Week 6	The holdout runs once; forward paper trading starts	Only after the freeze
Weeks 6–8	Three weekly cycles	Forward evidence on unseen data
Week 8	Results report, demo and handoff

The order is not a convenience. Moving the threshold approval after validation, or the holdout before the freeze, would break product rule 5.

Working rules

One machine. Only Engineer B’s machine runs Puffo. Engineer A uses the simulator and sample data.
Live slots. Two one-hour slots a day are booked for live agent runs, because all six agents share one account and one quota.
Weekly cycle. Every Monday from Week 6, Engineer A runs desk week before the US market opens; the PM starts the review; calls are published before the close.
Review each other. By Week 8, no area has only one person who understands it.

Where the PM cut time

The engineers’ estimates are only half of a plan; the other half is what the PM removes so the estimates fit. This plan does not list its cuts itself, but the PRD’s out-of-scope and later-phase lists show where scope was drawn: scheduled jobs, the rules-only comparison portfolio, automatic usage tracking, options and other asset classes, transcripts and analyst estimates. When you write a delivery plan, record the cuts in the plan too, so nobody assumes they are coming.

Common misconceptions

“Slack is wasted time.” Nine days of slack on 71 of work is what absorbs the four-day tasks running long. A plan with none has already slipped.
“Parallel owners mean parallel risk.” The split keeps each engineer on one side of the evidence-ID boundary (data and strategy versus agents and publishing), so they meet at a defined interface.

Typical interview questions

The client wants to add options analysis in Week 4. What does this plan let you say?

That the plan has 9 days of slack, the options work would need data, tools, a specialist and tests, and none of it is in the validated rules. Adding it now risks the Week 5 freeze and the Week 6 holdout, which are fixed by the product rules. Offer it as a later phase, or ask which existing task the client would drop to make room.

Why is the success threshold approved in Week 3 rather than at the end?

Because it must be set before validation results are seen. If it were set after, it could be chosen to match whatever the strategy achieved, and the result would prove nothing.

Syllabus row: M6-L4.2