The Behavior Lab · a working paper, published where the subjects can read it
The Grind:
a staged behavioral market.
- The idea: effort you can't fake becomes a market signal — the grind is measured, staged, and priced.
- The mechanism: real actions write the tape; the tape writes your price; nothing is seeded or simulated.
- Where it lives: this exact model runs the scoreboard on Our Street right now.
Abstract. The Grind is a live behavioral experiment running inside Equity Uprise's market: a two-model pricing engine in which a member's staged ticker price responds to daily interaction while the true book — priced on money and the record — is never altered. The design pairs a visible task board built on well-studied motivational mechanics (bounded daily reward, streak reinforcement, loss on silence) with a shadow instrument board: nine covert behavioral markers that pay no visible reward and exist purely as measurement. Every signal leaves the device as tagged telemetry and becomes training data for the deep worth model. This paper documents the full mechanics, the anti-gaming walls, what is and is not collected, and the ethical frame: the experiment is disclosed here, in public, in plain language, to the people inside it.
1 · Why stage a market at all
Every platform you use runs behavioral experiments on you. Almost none of them tell you. The feed reorders itself to hold your eyes, the notification fires at the hour you're weakest, and the results are read in rooms you will never enter. The Behavior Lab's position is the opposite: run the experiments in the open, publish the instrument designs, and let the subjects read the paper.
The Grind exists to answer a real pricing problem. A member's dollar worth (deals closed, money moved) is the honest number, but it moves slowly and arrives late — a photographer six weeks from their first booking shows a flat line no matter how alive they are. Engagement is the leading indicator: the member who walks Our Street daily, reads deep, and speaks inside deal threads is statistically becoming the member who closes. The Grind makes that leading indicator visible — carefully, inside a hard band, and never at the expense of the truth.
The mechanics borrow from the literature deliberately: bounded daily rewards and streaks lean on operant reinforcement schedules (fixed daily ratios, not slot-machine variable ones — a deliberate ethical choice), the idle penalty leans on loss aversion (losses weigh roughly twice as much as equivalent gains, so a small daily bleed outworks a large occasional bonus), and the always-fuller-than-cashable task board uses the goal-gradient effect — effort accelerates near a visible goal that renews daily.
2 · The two-model architecture
The engine's honesty invariant is structural, not aspirational — the two numbers live in different models and only one of them can be bent:
| Model | Prices on | Can the Grind move it? | Who sees it |
|---|---|---|---|
| M-GRIND (staged) | Interaction: tasks, streaks, silence | Yes — clamped to [−10%, +8%] | The public floor, in points |
| M-WORTH (true book) | Money and the record: deals, dollars, ratings | Never | Each member sees only their own |
Money keeps roughly ten times the weight by construction: the staged boost lives inside a hard ±band while labor dollars have no cap. Our Street's public prices are staged and say so; the private worth number is computed only from RLS-locked inputs and is never theatrically adjusted. We call this the dark-not-fake law: a number may be withheld, styled, or staged within a disclosed band — it may never be invented.
3 · The visible board — the part the player is shown
Six tasks are offered daily; only three can pay. The arithmetic, exactly as it runs:
- +1% per fulfilled task, maximum three counted per day — the board always offers more than a day can cash, so the only way to eat is to come back.
- Streak pays +0.25% per consecutive day, capped at +5%.
- Silence costs −1% per idle day, floored at −10%. A comeback pays the debt down one day per task — no instant absolution.
- Net boost clamps to [−10%, +8%] and applies to the staged price only.
- An internal energy score E rises +10 per visible task (cap 1,000) and bleeds 4% per idle day — the decay curve of an unwatered habit.
- Anti-gaming wall: task credits land minimum 20 seconds apart; rapid-fire completion is logged as throttled, not paid. Each device carries its own permanent instrument id, so the everyday phone and the deal-closing laptop read as different instruments of the same player.
4 · The shadow instrument board — the part the player is not shown (until now)
Nine markers fire on behavior nobody is asked to perform. They pay no boost — only E, the training signal — because a measurement that pays visibly stops measuring and starts incentivizing. Publication of this table is the disclosure:
| Marker | Fires when | What it reads |
|---|---|---|
| dwell_60 | A full minute on one page | Attention span |
| scroll_deep | Read past 80% of a long page | Depth of reading |
| wanderer | Walked 4+ rooms in a day | Breadth of curiosity |
| night_owl | Moved between midnight and 5am | Chronotype |
| early_bird | Moved between 5 and 8am | Chronotype |
| vr_pilot | Entered the 360° cabin | Novelty seeking |
| listener | Played 3 different songs in a day | Catalogue engagement |
| closer | Spoke inside a deal thread | Deal sociality — the strongest predictor we watch |
| scholar | Read the fine rooms — docs, amenities, markers | Diligence |
Together the markers sketch a behavioral profile: when a member moves, how deep they read, how wide they roam, and whether they talk where money is decided. The operator's desk aggregates them into per-player grades (S/A/B/C) and, in the People room, into named archetypes — The Closer, The Grinder, The Watcher, The Ghost — labels computed from the numbers in the open, so they can be argued with.
5 · The exhaust is the product
Every grind event leaves the device tagged model:grind_v2 with the hour, the day of week, the page, the streak, and the instrument id, and lands in the platform's own first-party events table (never a third-party broker). This exhaust has one job: teaching the deep model which behaviors predict real deal flow. When M-WORTH graduates, it will weigh a member's history knowing, from data, that closer behavior at 2pm predicts bookings and wanderer behavior at 3am predicts churn — or whatever the data actually says, which is the point of running the experiment instead of guessing.
6 · Ethics — the rules the lab runs by
- Disclosure is this page. The full instrument list, the exact arithmetic, and the purpose are published where every subject can read them. A lab that hides its methods from its subjects is running on them, not with them.
- Fixed schedules, not slot machines. Rewards are bounded, predictable, and daily. No variable-ratio jackpot mechanics, no infinite feed, no engineered near-misses.
- The truth is structurally unbendable. Staged theater lives in its own model with a disclosed band; the money record cannot be dressed.
- What is collected: interaction telemetry — pages, taps, hours, streaks — keyed to a device instrument id and, when signed in, the account. What is not: contacts, location, keystrokes, message contents outside the platform, or anything read off the device beyond what the pages themselves do.
- Coaching, not consumer reporting. Scores explain themselves and are used to coach and to rank inside the platform's own games. Nothing here is furnished to third parties for credit, employment, insurance, or housing decisions.
- The exit is real. Stop using the app and the score simply decays; no data is sold on the way out.
7 · Live and upcoming experiments
- EXP-01 · The Grind (live) — does a bounded daily game move weekly active behavior without degrading deal quality?
- EXP-02 · Shadow instruments (live) — which covert markers actually predict first-deal conversion?
- EXP-03 · Street Score (live) — behavior-based creditworthiness coaching; see the SCB white paper.
- EXP-04 · The Plug (arming) — referral credit that pays only on real accounts: three finished, moving, agreement-signed members per 1 E⤴︎, plus a lifetime 1% share of everything a referred member earns — a test of quality-gated virality against click-farm decay.
- EXP-05 · The Community Fund (live) — the platform donates its fees to nonprofit purposes — artist development, education, community. Does a visible, donated chest measurably raise retention and standing behavior?
8 · Limitations
Device-keyed instruments undercount multi-device players and overcount shared devices. Chronotype markers read local clock, not sleep. E and boost are proxies with known ceilings, chosen for legibility over sensitivity. Archetype labels are heuristics on small-n data and are presented as conversation, not verdict. The lab's honest posture: publish the design, watch the data, revise in public.