WINDROSE ATLAS · USE CASES · REAL RUNS

One tool, ten decisions — really run.

Ten industry decisions, put through the engine — not illustrations. Each card is a real SIM-RANK run on a documented synthetic population, with its stability, its segment split, and what drove the read.

These are runs on synthetic, documented reference populations (we supply the population). RUNG = SIMULATION, CALIBRATION = NONE — so no hard percentages, only direction and ranking. Not a live panel: a simulation on an explicitly described population.

Honesty over narrative: these real rankings differ from the illustrative ones. Several reads are near-ties, and in a few the textbook mechanism did not win — we say so. That is the product: a compass, not a verdict.

Mechanisms are hypotheses, not guarantees — behavioural effects vary by population and context. Atlas ranks which honest message to test first; it does not promise the effect, and it is not a manipulation kit.

Industry
Mechanism

SAAS · PRICING

Which of three pricing pages do we ship?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Three independent tiers > Annual framing > Dominated middle / decoy (0.298 · 0.254 · 0.210)
STABILITY
94/100 · HIGH (282/300 Monte Carlo)
SEGMENTS
Unanimous — all four buyer segments rank independent tiers #1 (robust read)
TOP HYPOTHESIS
The decoy did NOT make Pro obvious — it ranked last. The textbook decoy/anchoring effect is a hypothesis this segment didn't reward; ship-first candidate is the clean three-tier page.
VERIFY NEXT
2-week A/B on the live pricing page; decision metric: paid conversions, not clicks.
DRIVERS · SIM
Buyer traits barely differentiate (top trait gain ≈0.007). The finding: this decision is message-driven, not segment-driven.

Mechanism (hypothesis) Anchoring + decoy

A compass, not a verdict.

FINTECH · IN-APP MESSAGE

Which message turns on auto-saving?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Social norm (8%) ≈ Loss (don't lose 2,400) > Instruction (save 200) (0.306 · 0.304 · 0.295) — near-tie at the top (Δ0.002)
STABILITY
59/100 — fragile order, treat as a tie
SEGMENTS
Split
TOP HYPOTHESIS
Norm ≈ loss; the plain instruction loses. Test norm vs loss first (both framings must be TRUE).
VERIFY NEXT
2-week in-app A/B; metric: activated auto-saving after 30 days, not taps.
DRIVERS · SIM
Strongest trait: mobile banking (gain 0.39), then social comparison — the norm lands on mobile, comparison-prone users.

Mechanism (hypothesis) Loss aversion / framing · Social proof

A compass, not a verdict.

INSURANCE · OFFER ARCHITECTURE

Which cover as the pre-selected default?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Recommended-cover default > Minimal-cover default > Forced choice (0.208 · 0.200 · 0.196)
STABILITY
100/100 — stable order (Δ0.008, narrow)
SEGMENTS
Split
TOP HYPOTHESIS
Defaults carry implicit advice; forced choice does worst (default effect / status quo) — consistent with the textbook.
VERIFY NEXT
Split test in the quote funnel; watch completion AND downstream complaints.
DRIVERS · SIM
Need-for-protection dominates (gain 0.45); trust-in-defaults follows — the recommended default works on those who feel the need to be covered.

Mechanism (hypothesis) Default / status quo

A compass, not a verdict.

RETAIL · CHECKOUT OFFER

“Free shipping” or “10% off” — which moves the cart?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Free-returns badge > 10% discount > Free shipping over threshold (0.289 · 0.247 · 0.223)
STABILITY
98/100 · HIGH (293/300)
SEGMENTS
Split — the deal-hunter segment prefers the 10% discount; the other three prefer the free-returns badge. No single winner for everyone.
TOP HYPOTHESIS
Reassurance (free returns) reduces perceived risk more than a raw discount for most segments; deal-hunters still chase the number. Which message to run depends on who you target.
VERIFY NEXT
3-week storewide test; metric: contribution margin per order, not conversion alone.
DRIVERS · SIM
Strongest driver: reads-reviews (gain 0.23) — review-readers are the most movable and respond to risk-reduction; mobile-share and cart-abandonment follow.

Mechanism (hypothesis) Loss aversion / framing · Zero-price effect

A compass, not a verdict.

HEALTH · PUBLIC MESSAGE · 5 VARIANTS

Which screening-invitation letter goes to print?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Gain ≈ Social norm ≈ Implementation-intention > Loss > Authority (0.313 · 0.312 · 0.309 · 0.305 · 0.302) — a 5-variant near-tie (~1pp spread)
STABILITY
100/100 — order stable, but magnitudes are almost flat
SEGMENTS
Split — four different segment winners
TOP HYPOTHESIS
Here the variants are close: a low-separation decision with high segment-dependence — this is where multi-variant genuinely beats a coin toss per segment.
VERIFY NEXT
Regional pilot with the health provider; metric: booked appointments.
DRIVERS · SIM
Procrastination is the top driver (gain 0.36) — planning variants (implementation intention) matter for procrastinators; education follows.

Mechanism (hypothesis) Social proof · Commitment & consistency · Loss aversion / framing · Authority

A compass, not a verdict.

HEALTH · ADHERENCE

Which reminder keeps patients on schedule?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Neutral reminder > Habit-anchored > Streak (0.331 · 0.311 · 0.285) — the cleanest separation of the batch (Δ0.019)
STABILITY
100/100
SEGMENTS
Split
TOP HYPOTHESIS
Surprise vs textbook: a plain neutral reminder beat habit-stacking and the streak here. A hypothesis to probe — overload of “clever” messages?
VERIFY NEXT
IRB-cleared pilot with the clinic; metric: refill and self-reported adherence at 8 weeks.
DRIVERS · SIM
Present bias dominates (gain 0.70) — the most movable are present-biased; health literacy follows.

Mechanism (hypothesis) Present bias · Commitment & consistency

A compass, not a verdict.

FITNESS · RETENTION

What keeps January members past March?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Streaks & milestones ≈ Refundable deposit > 6-month discount (0.248 · 0.245 · 0.226) — near-tie at the top (Δ0.002)
STABILITY
75/100
SEGMENTS
Split
TOP HYPOTHESIS
Identity/consistency and a small deposit beat a price discount (commitment & consistency + loss framing) — consistent; the discount ranks last.
VERIFY NEXT
One-club cohort test for 90 days; metric: month-4 retention.
DRIVERS · SIM
Streak-motivation dominates (gain 0.53) — streaks/milestones land on the streak-motivated and goal-oriented.

Mechanism (hypothesis) Commitment & consistency · Goal gradient · Loss aversion / framing

A compass, not a verdict.

PSYCHOLOGY · PRACTICE OPS

How do we reduce first-session no-shows?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Message from the therapist by name > Standard SMS > “reply YES” micro-commitment (0.304 · 0.291 · 0.285) (Δ0.013)
STABILITY
61/100 — fragile
SEGMENTS
Split
TOP HYPOTHESIS
The personal, by-name message leads; the textbook micro-commitment (“reply YES”) ranked last — a hypothesis to test.
VERIFY NEXT
6-week practice pilot; metric: attended first sessions.
DRIVERS · SIM
Preference-for-personal-contact dominates (gain 0.35) — the by-name message lands on those who value personal contact; SMS-responsiveness follows.

Mechanism (hypothesis) Commitment & consistency

A compass, not a verdict.

PSYCHOLOGY · MODALITY UPTAKE

Which modality will patients most readily take up and stay with?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
In-person CBT > Psychodynamic ≈ Online ≈ Group (0.363 · 0.351 · 0.349 · 0.348) — CBT slightly ahead, the rest near-flat (Δ0.012)
STABILITY
98/100
SEGMENTS
Split
TOP HYPOTHESIS
This is PREFERENCE / uptake readiness (behavioural, simulable) — NOT clinical efficacy. What HEALS is decided by an RCT, not a panel. Ask Atlas about clinical efficacy and it REFUSES: WE DIDN'T KNOW · OUT OF CLASS.
VERIFY NEXT
A preference / uptake study, not a therapy-outcome study.
DRIVERS · SIM
Preference-for-personal-contact and SMS-responsiveness drive uptake (gain 0.35 / 0.31).

Mechanism (hypothesis) Refusal · out of class

A compass, not a verdict.

TELECOM · RETENTION OFFER

Which save-desk offer actually retains?

WINDROSE ATLAS · RUN
EVIDENCE LABEL
SIM-RANK · DIRECTIONAL
READING
Free upgrade (price unchanged) ≈ 20% discount > Simplified plan (0.262 · 0.259 · 0.252) — near-tie at the top (Δ0.003)
STABILITY
100/100 — stable order
SEGMENTS
Split
TOP HYPOTHESIS
Added value ≈ discount; the simplified plan (choice-overload) ranked last — against the “fewer options retain” prior. A hypothesis to test.
VERIFY NEXT
Save-desk script test in one region for 60 days; metric: 6-month survival, not same-day saves.
DRIVERS · SIM
Churn-risk dominates (gain 0.39) — offers move the already-at-risk; complexity-tolerance follows.

Mechanism (hypothesis) Choice overload · Default / status quo

A compass, not a verdict.

Reproducibility

Every read is one engine run; hash = sha256(model + config) from the run manifest; dry_run:false.

#use casepopulation (synthetic)stabilityhashdate
01SAASsaas_buyers · 600 · 4 seg.94/10091a403c2…2026-07-20
02FINTECHfintech_users · 600 · 4 seg.59/100f3c9f3c4…2026-07-20
03INSURANCEinsurance_applicants · 600 · 4 seg.100/100a27d7810…2026-07-20
04RETAILecommerce_eu · 610 · 4 seg.98/100611f3aef…2026-07-20
05HEALTHhealth_public_pl · 600 · 4 seg.100/100f261ffed…2026-07-20
06HEALTHpatients_pl · 600 · 4 seg.100/100e0583c89…2026-07-20
07FITNESSgym_members · 600 · 4 seg.75/100ff3b992e…2026-07-20
08PSYCHOLOGYclinic_clients · 600 · 4 seg.61/100dd190677…2026-07-20
09PSYCHOLOGYclinic_clients · 600 · 4 seg.98/10079e1c8f8…2026-07-20
10TELECOMtelecom_subs · 600 · 4 seg.100/100c96bb570…2026-07-20

Oracle model: Minitaur-8B (Centaur family), run locally (MLX). 6 paraphrases per scenario, k=4 persona clusters. Experiment design: Claude. Reproduce: bloop.runner configs/<slug>.json --stages 1-8.