CHAIN Framework: Chain-of-Thought Prompting Guide

Our Node.js Express API runs on 4 AWS EC2 instances behind an ALB (Application
Load Balancer). It connects to a PostgreSQL RDS instance (db.r5.xlarge). The API
handles 3,000 requests per minute during business hours. The 504 errors started
3 days ago. No code was deployed in the past 2 weeks. CloudWatch shows CPU usage
on the EC2 instances at 45% and RDS CPU at 78%. The ALB timeout is set to 30
seconds. Average response time was 200ms last week; it is now 1,800ms at the p95.

Hypothesis: The 504 errors are caused by PostgreSQL query degradation, likely
from table bloat or missing index usage on a query plan change, which is pushing
response times past the ALB's 30-second timeout for the slowest requests.

Analysis:
1. Compare RDS query execution times from last week vs. this week for the top 10
   slowest queries
2. Check if PostgreSQL autovacuum has run recently and whether any large tables
   have significant dead tuple counts
3. Examine the ALB access logs for which specific endpoints are generating 504s
4. Review the EC2 connection pool configuration and check for pool exhaustion
   during peak load
5. Verify that no RDS maintenance events or failovers occurred in the past week

Inference: Based on the analysis, determine whether the root cause is database-side
(supporting the hypothesis) or infrastructure-side (refuting it). If multiple factors
contribute, rank them by impact. Identify the single change that would resolve the
majority of 504 errors.

Narration: Present the findings as a root cause analysis with: (1) executive summary
stating the root cause in one sentence, (2) evidence chain showing how each analysis
step supports or contradicts the hypothesis, (3) recommended fix with specific
commands or configuration changes, (4) a verification plan to confirm the fix worked.

Context: A factory produces 3 types of widgets. Type A costs $4 to make and sells
for $10. Type B costs $7 to make and sells for $18. Type C costs $12 to make and
sells for $28. The factory has a daily budget of $2,400 for materials and can
produce a maximum of 400 widgets per day across all types. Customer demand requires
at least 50 Type A, 30 Type B, and 20 Type C widgets daily.

Hypothesis: Maximizing Type B production (after meeting minimums) yields the highest
daily profit because it has the best profit-to-cost ratio.

Analysis:
1. Calculate the profit margin and profit-to-cost ratio for each widget type
2. Determine the material cost of meeting minimum demand requirements
3. Calculate remaining budget and capacity after fulfilling minimums
4. Model three allocation strategies: maximize A, maximize B, maximize C
5. Compare total daily profit under each strategy

Inference: Determine whether the hypothesis holds or whether a different allocation
maximizes profit. Identify if the budget constraint or the capacity constraint is
the binding factor.

Narration: Present the optimal production plan as a table showing units per type,
total cost, total revenue, and total profit. Include the mathematical reasoning
for each step.

Context: A Python Flask application uses SQLAlchemy with a PostgreSQL database.
Users report that the /api/reports endpoint takes 45 seconds to respond when
generating reports for date ranges longer than 90 days. The same endpoint returns
in under 2 seconds for 30-day ranges. The reports table has 12 million rows and
is indexed on created_at. The query uses a JOIN with a 500K-row customers table.

Hypothesis: The query planner switches from an index scan to a sequential scan
when the date range exceeds a threshold, because PostgreSQL estimates that scanning
the full table is cheaper than using the index for large result sets.

Analysis:
1. Run EXPLAIN ANALYZE on the query for a 30-day range and a 120-day range
2. Compare the estimated vs. actual row counts in both plans
3. Check table statistics: when was the last ANALYZE run on the reports table?
4. Examine if the JOIN strategy changes between the two ranges (nested loop vs.
   hash join)
5. Calculate what percentage of the table a 90-day range represents

Inference: Confirm whether the plan switch is the root cause or whether the
slowdown comes from the JOIN, data volume, or memory pressure. Recommend the
fix that addresses the actual bottleneck.

Narration: Provide the root cause with EXPLAIN ANALYZE output comparison, the
specific fix (partial index, query hint, or statistics update), and the SQL
commands to implement it.

Context: Five houses in a row are painted different colors (red, blue, green,
yellow, white). Each owner has a different nationality (English, Spanish, Japanese,
Norwegian, Italian), drinks a different beverage (coffee, tea, milk, juice, water),
and has a different pet (dog, cat, bird, fish, horse). Clues: The English person
lives in the red house. The Spanish person has a dog. Coffee is drunk in the green
house. The Italian drinks tea. The green house is directly to the right of the white
house. The bird owner drinks juice. The yellow house owner drinks water. Milk is
drunk in the middle house. The Norwegian lives in the first house. The cat owner
lives next to the juice drinker. The horse owner lives next to the yellow house.
The Norwegian lives next to the blue house.

Hypothesis: The Norwegian lives in the yellow house (position 1), and the fish
belongs to someone in the green or white house.

Analysis:
1. Start from the fixed facts: Norwegian in position 1, milk in position 3
2. Determine house colors using the adjacency constraints
3. Assign nationalities based on house colors and remaining constraints
4. Assign beverages based on house colors and nationalities
5. Assign pets using the remaining constraints and adjacency rules

Inference: Solve the puzzle step by step. Confirm or correct the hypothesis about
the Norwegian's house color and the fish owner's location.

Narration: Present the complete solution as a table (position, color, nationality,
beverage, pet) with the logical reasoning chain that produced each assignment.

Context: An e-commerce company's conversion rate dropped from 3.2% to 2.1% over
the past month. Traffic remained stable at 500K monthly visitors. The drop
coincided with a site redesign that changed the checkout flow from 3 steps to 5
steps, added a mandatory account creation screen, and introduced a new product
recommendation carousel on the cart page. Mobile traffic is 65% of total.

Hypothesis: The mandatory account creation screen is the primary cause of the
conversion drop, accounting for 60%+ of the lost conversions, because it adds
friction at the highest-intent moment in the funnel.

Analysis:
1. Compare funnel drop-off rates at each step: pre-redesign (3 steps) vs.
   post-redesign (5 steps)
2. Identify which specific step has the highest new abandonment rate
3. Segment the data by device type (mobile vs. desktop) to see if mobile users
   are disproportionately affected
4. Check the account creation step completion rate vs. the old guest checkout rate
5. Measure whether the product recommendation carousel increases cart
   modifications or just adds load time

Inference: Quantify the contribution of each change to the overall conversion drop.
Determine if the hypothesis about account creation is correct or if the additional
steps and carousel are larger factors. Identify whether the impact differs
significantly between mobile and desktop users.

Narration: Deliver the analysis as a conversion funnel report with: a waterfall
chart description showing where users drop off, the estimated revenue impact per
change, and a prioritized list of recommended reversions or fixes.

Context: A greenhouse experiment is testing whether LED grow lights with a higher
blue-to-red ratio (3:1 vs. the standard 1:2) affect tomato plant growth. After
6 weeks, the 3:1 blue-red group shows 15% shorter stems but 40% more leaf surface
area compared to the control group. Fruit production has not started in either
group. Both groups receive identical water, nutrients, and temperature conditions.

Hypothesis: The higher blue light ratio is causing compact vegetative growth with
increased photosynthetic capacity, which will translate to higher fruit yield once
flowering begins because the plants have more energy-producing leaf area.

Analysis:
1. Review existing research on blue light effects on tomato morphology and
   photosynthesis rates
2. Evaluate whether the shorter stem height indicates stress or efficient growth
3. Calculate the estimated photosynthetic capacity difference based on leaf
   surface area
4. Assess whether compact growth with more leaves is advantageous or disadvantageous
   for fruit production in greenhouse conditions
5. Identify potential confounding variables (light intensity differences, spectral
   overlap effects)

Inference: Determine whether the observed growth pattern is likely beneficial for
eventual fruit yield or whether it signals a problem. Assess confidence level in
the hypothesis given that fruit production data is not yet available.

Narration: Write a mid-experiment assessment memo with: current observations
summary, hypothesis evaluation with confidence level, recommended adjustments
to the experiment if any, and a timeline for when fruit production data should
confirm or refute the hypothesis.

Context: A marketing team with a $50K quarterly budget needs to choose between
three campaign strategies for a B2B SaaS product (ACV $15K, sales cycle 4 months).
Option A: LinkedIn ads targeting CTOs ($25 CPM, estimated 2% CTR). Option B:
Content marketing with SEO, producing 12 long-form articles per quarter (each
costing $2K to produce). Option C: Sponsoring 4 industry podcasts ($8K per
sponsorship, estimated 50K listeners each). Current organic traffic is 20K monthly
visits with a 0.5% conversion to demo requests.

Hypothesis: Content marketing (Option B) will generate the highest 12-month ROI
because the compounding nature of SEO traffic will outpace the linear returns of
ads and sponsorships after month 6.

Analysis:
1. Calculate the expected pipeline value per dollar spent for each option over
   3, 6, and 12 months
2. Model the traffic growth curve for content marketing assuming standard SEO
   ramp-up timelines
3. Estimate the lead quality difference: which channel produces prospects closest
   to purchase intent?
4. Assess the time-to-first-result for each option
5. Calculate the break-even point for each strategy

Inference: Determine whether the hypothesis holds at each time horizon. Identify
whether a blended approach (splitting budget across options) outperforms any
single strategy. Flag the assumptions most likely to be wrong.

Narration: Present a decision matrix comparing all three options across ROI at
3/6/12 months, time to first lead, lead quality score, and scalability. Include
a recommended allocation with quarterly milestones.

Context: A 200-person software company generates $20M ARR, growing 25% YoY. 80%
of revenue comes from a single product (project management tool). The CEO wants
to launch a second product (time tracking) to reduce concentration risk. The
engineering team has 60 developers, 45 of whom work on the core product. Competitors
in time tracking include two established players with 70% market share combined.

Hypothesis: Building a time tracking product internally will dilute engineering
focus and slow core product growth by 10-15% without reaching meaningful revenue
($2M+ ARR) within 18 months, making acquisition of a smaller time tracking
startup a better path to diversification.

Analysis:
1. Estimate the engineering allocation required to build a competitive time
   tracking MVP (based on feature parity with the simpler of the two competitors)
2. Model the impact on core product roadmap velocity if 15 developers are
   redirected
3. Research acquisition targets: small time tracking companies with $500K-2M ARR
   that could integrate with the existing product
4. Compare the 18-month total cost of build vs. buy (salaries, opportunity cost,
   integration effort)
5. Assess the go-to-market advantage of each approach (cross-sell to existing
   customers vs. acquiring an existing user base)

Inference: Determine whether build or buy better achieves the diversification goal
within the CEO's timeline. Identify the key risk for each path and the conditions
under which the other option becomes preferable.

Narration: Deliver an executive briefing with: a build vs. buy comparison table,
a recommended approach with 18-month timeline, the three biggest risks with
mitigation strategies, and the decision criteria for a go/no-go checkpoint at
month 6.