Marketing science, reproduced. Not just quoted.
Everyone cites the research. We rebuild it. Our agents read 8 landmark marketing papers, turn each one into working code, and check it against the paper's own math, its published data, or a second paper, live in your browser. What passes goes to work on your numbers. What can't be checked is labeled in review and never goes into a forecast.
Investigation feed
Simulator · your numbers
Method workspacethe code running now
Shared scene / difference field
Source / extraction routes
| PAPER | EXTRACT | CODE | CHECK |
|---|
Validation / sample buffer
Build crew / dispatch
3 roles per paper: read, build, review. Review = the live checks.
Method / reference index
| METHOD | CHECKS (hover for detail) | STATE | |||
|---|---|---|---|---|---|
Pick a paper, slide the numbers, and watch every panel follow: the code that's running, the paper-vs-code difference map, the checks, and what it means for your business. Hit Re-run every check to watch the build crew do it again. Nothing is sent anywhere.
Not an animation. A lab.
The usual "AI research lab" demo
- Counters typed in by hand
- Waveforms that loop forever
- Every result glows green
- You can watch it, not use it
The Paper Lab
- Every counter and pass/fail computed on your device at load
- Paper vs. our code, side by side, with the gap measured
- One paper honestly left in review, with the reason
- Every model takes your numbers and answers a business question
The results, in plain English
The same numbers the lab computes above, written down. 24 of 26 checks pass; the two open checks belong to the one paper in review.
| Paper | What it tells you | Checks passed | Status |
|---|---|---|---|
| Word-of-mouth growthBass, F. M. (1969). A New Product Growth for Model Consumer Durables. Management Science 15(5), 215-227. | How a new offer spreads: a few people try it because of your marketing (p), most try it because someone told them (q). It predicts the month word of mouth peaks. | 4 of 4RK4 solver vs paper's closed form; Adopters never exceed the market; Peak lands at t* = ln(q/p)/(p+q); Meta-analysis averages give a peak near period 6.2 | Reproduced |
| Ad carry-overBroadbent, S. (1979). One Way TV Advertisements Work. Journal of the Market Research Society 21(3), 139-166. | An ad keeps working after it runs, then fades. λ is how much of last week's effect carries into this week. It tells you what pausing really costs. | 4 of 4Constant spend settles at x/(1-λ); One burst's total effect = 1/(1-λ); Half the effect is gone at h = ln 0.5/ln λ; Stays bounded when λ < 1 | Reproduced |
| Trade area shareHuff, D. L. (1964). Defining and Estimating a Trading Area. Journal of Marketing 28(3), 34-38. | Shoppers weigh how big or good a place is against how far it is. It maps the share of your town each business wins, block by block. | 3 of 3Every shopper's choices add to 100%; Two equal stores split the midpoint 50/50; More pull never loses share | Reproduced |
| Town-to-town pullReilly, W. J. (1931). The Law of Retail Gravitation. Breakpoint form: Converse, P. D. (1949). New Laws of Retail Gravitation. Journal of Marketing 14(3), 379-384. | Between two towns, where do shoppers switch from one to the other? The bigger town pulls the line toward the smaller one. Useful for deciding how far out to advertise. | 3 of 31931 law agrees with the 1964 model; Worked case: 40 mi, 100k vs 25k people; Equal towns split the road in half | Reproduced |
| Memory fadeEbbinghaus, H. (1885). Über das Gedächtnis (Memory: A Contribution to Experimental Psychology). | Memory drops fast in the first hours, then slowly for weeks. That's why the first follow-up should come the same day, and later ones can be spaced out. | 3 of 3Power law reproduces the 1885 data; Simple exponential decay does not; Published savings only ever fall | Reproduced |
| Front-desk queueLittle, J. D. C. (1961). A Proof for the Queuing Formula: L = λW. Operations Research 9(3), 383-387. | Calls in line equal calls per hour times how long each one waits. When one person answers the phone, waits explode as they get busy. Rudy answers every text at once, so there is no line. | 3 of 3Simulated L equals λ·W; Matches the M/M/1 closed form; Flags overload when ρ >= 1 | Reproduced |
| Customer lifetime valueGupta, S. & Lehmann, D. R. (2003). Customers as Assets. Journal of Interactive Marketing 17(1), 9-24. | What one customer is worth over time, given the profit per period, how many come back, and the value of money later. Small retention gains are worth a lot. | 3 of 3Closed form equals the year-by-year sum; Paper's margin multiple r/(1+i-r); Value rises with retention | Reproduced |
| Share of voiceJones, J. P. (1990). Ad Spending: Maintaining Market Share. Harvard Business Review 68(1), 38-42. Growth rule as reported by Binet & Field for the IPA. | Brands that out-shout their market share tend to grow. It's a real pattern across many brands, but it's an average from private case databases, so we can't re-run it. It stays in review. | 1 of 3ESOV is computed by definition; Growth rule has an equation we can re-run; Underlying case data is public | In review |
Eight business questions it answers
Each model is a decision tool once it's verified. Here's what each one is for.
When will word of mouth peak?
Plug in your market size and see the month referrals take over from ads.
What does a pause really cost?
See how much of last month's ads are still working, week by week.
How much of town do I win?
Drop your location on the map and watch your trade area grow or shrink.
How far out should I advertise?
Find the line where shoppers switch to the bigger town.
When should I follow up?
Memory falls fastest on day one. Text the same day, then space it out.
How long are callers waiting?
Calls per hour times minutes per call shows when the front desk breaks.
What is one customer worth?
Profit, return rate and time give you a real number to spend against.
Should I out-shout my share?
A real pattern, but we couldn't re-run it, so it stays out of forecasts.
How the build crew works
- ReadPull the governing equation, settings, domain and page references out of the paper, and flag anything missing.
- BuildTurn it into runnable code with controls, so it can take real numbers instead of the paper's example.
- ReviewCheck the code against the paper's closed form, its published data, or a second paper. No check, no "reproduced."
Questions
What is the Paper Lab?
It's a live lab where eight landmark marketing-science papers are rebuilt as working code and checked against the paper's own math, its published data, or a second paper. Every check runs in your browser when the page loads, and you can plug your own numbers into each model.
What does reproduced mean here?
It means our code gives the same answer as the paper: the solver matches the paper's closed-form equation, the fit matches the published data, or two papers agree with each other. If we can't check it, it isn't marked reproduced.
Why is one paper in review?
The share-of-voice growth rule is a real, widely cited pattern, but it's an average across private case databases with no equation to re-run. We show it, label it, and keep it out of any forecast until it can be checked.
Is my data sent anywhere?
No. The simulators run entirely in your browser and the page has no form. Nothing you type or slide leaves your device.
Can you run this on my business?
Yes. Text Rudy your numbers (calls per hour, return rate, ad spend, locations) and we run them through the lab and send back what it means in plain English. We can also build a lab-style calculator for your own site.
Are these predictions guaranteed?
No. Each model is a simplified, well-tested way of thinking about a decision, not a promise. We show the assumptions on screen so you can see exactly what drives each number.
Run your numbers through the lab
Text Rudy your calls per hour, return rate, ad spend or locations. We'll run them through the verified models and send back what they mean for your business, in plain English.
Text Rudy: run my numbers Opens your messaging app with a draft. You choose whether to send. Msg & data rates may apply. Reply STOP to opt out, HELP for help.