Winner Patterns — What Actually Gets to Houston
Evidence base: data/winners.jsonl — 59 records across 2024, 2025, 2026 cycles. The 2026 cycle is the deep sample: 35 records, 31 with full innovation descriptions (the complete Innovation Summit finalist roster). 2025 and 2024 are name-level only.
Read the honest limitation first: the strong claims below are drawn from one deep cohort (2026). Treat them as well-grounded hypotheses about what the rubric rewards, cross-checked against the rubric text and the Chief Judge's stated reasoning — not as multi-year statistics.
1. The one-sentence finding
Conrad finalists do not win with advanced technology. They win with a bounded problem, a named incumbent, a quantified delta, and a moat.
Nearly every 2026 finalist description contains the same four moves. Most losing briefs (see 02-judging-playbook.md) contain none of them.
2. The finalist sentence template
Reverse-engineered from all 31 described 2026 finalists. It is startlingly consistent:
[Specific artifact] for [named, bounded user]
that [does X] via [named mechanism],
achieving [quantified delta] versus [named incumbent],
priced at [number], defended by [specific moat].
Worked examples, unmodified from the official finalist descriptions:
| Team | Artifact + user | Mechanism | Quantified delta | Incumbent named | Moat |
|---|---|---|---|---|---|
| Vigil | vital-sign monitor for unmonitored hospital beds | radar + thermal + motion | ICU-level at <$100/bed | ICU monitoring, consumable-based | no consumables |
| Echo/Pulse | speech headset for 7.5M Americans with speech disorders | facial EMG + LLM | $2,500 | AAC tablets 10, 500 * *, implants * *50–100k | EMG-LLM pipeline |
| Clarity | microplastic detector for households | optical scattering + AI | <$250 | lab equipment $10,000+ | 40× price collapse + crowd dataset |
| BeeGuard | hive diagnostic for 10M beekeepers | temporal multimodal AI | 94.7% in <1 s, zero hardware cost | manual inspection | detects weeks before symptoms |
| Aphelion | attitude control for small satellites | spherical motor | −20% mass, −61.8% volume, −71% power | gimbaled mechanisms | proprietary control algorithms |
| Root and Renew | seeder for 480M smallholder farms | no-till precision | −62% seed, +52% yield | ploughing | $499, 1–2 season payback |
Notice what is absent: none of these is a scientific discovery. Several use entirely off-the-shelf components (Clarity: optical scattering; BeeGuard: a phone camera and microphone).
3. The seven repeatable winning archetypes
Each is a strategy for generating a defensible innovation, not a topic. These drive the topic engine in topic-engine/METHOD.md.
A. Price collapse
Take a capability that exists only at institutional cost and deliver it at consumer cost.
- Clarity — $10,000 lab → $250 consumer microplastic detection
- Vigil — ICU monitoring → <$100/bed
- Echo/Pulse — $50k–100k implant → $2,500 wearable
Why it wins: the delta is a single number, instantly checkable, and the incumbent is named for you. Innovation scores on Ambition; Marketing scores on Differentiation. Requirement: you must name the expensive incumbent and its real price.
B. Zero-resource constraint
Deliver the function with one critical input removed — no power, no hardware, no water, no connectivity.
- BlueFlow HydroSpines — atmospheric water with zero energy input
- ExFire Brick — fire suppression with no electricity, water, or intervention
- BeeGuard — diagnosis at zero hardware cost
- BeCure — runs offline on basic phones
- EcoClave — passive countercurrent heat exchange, no controls
Why it wins: the constraint is the innovation, and it is self-evidently hard. Also scores Impact Potential because it unlocks underserved geographies. Requirement: name the input you removed and why competitors need it.
C. Retrofit into installed base
Don't replace the expensive thing; clip onto it.
- Kipos SubStation — retrofit onto aging subway cars, 200+ metro systems
- AeroLattice — flap retrofit for existing 737/A320 fleets, 18,000+ aircraft
- EcoClave — external install on existing autoclaves, workflow unchanged
Why it wins: market size is a fleet count you can look up, adoption friction is near zero, and it neutralises the "who would rip out their equipment?" objection before it's asked. Requirement: state the installed-base number and prove the retrofit changes nothing else. EcoClave's "sterilization procedures unchanged" is doing enormous work — it kills a regulatory objection in four words.
D. Workflow replacement
Attack the process, not the device. Collapse sample→transport→lab→report into in-situ.
- MicroSeek Ideon — on-site microplastic analysis replacing lab transport
- BeCure — behavioural analysis replacing physical hive inspection
- Dravix — ML screening upstream of physical fire testing
Why it wins: differentiation is structural rather than incremental, so Ambition Level scores high. Requirement: diagram the old workflow and the new one, with time/cost per step.
E. Data as the second moat
The device is the wedge; the accumulating dataset is the business.
- SafeSat — debris shield plus monetizable in-situ debris measurement datasets
- Clarity — detector plus crowd-sourced contamination maps for researchers and policymakers
- Schoolace — platform plus student knowledge graphs
- Back Tracker — patient exercise app plus objective movement data for clinicians
Why it wins: directly answers IP Defensibility without a patent. A dataset that grows with adoption is a textbook moat and reads as commercially sophisticated. Requirement: name who buys the data and why they can't collect it themselves.
F. Razor / blade and recurring revenue
Put the defensible IP in the consumable or the service, not the hardware.
- VORTA GEOOSE — robots deploy patent-pending SAP-encapsulated capsules (IP is in the capsule)
- Ocean Energy (non-finalist, but strong on this) — units plus maintenance contracts plus licensing
- OctoScope — monitoring service with hardware
- Dravix — cloud subscription
Why it wins: Finances scores on Revenue Projections and Financial Viability. Recurring revenue is the single easiest way to look like you understand business.
G. Hyper-specific geography or failure mode
Refuse the generic problem. Solve one population's exact version of it.
- Soaring Seeds — safe water specifically for western plateau pastoral nomadic households, addressing filter clogging, secondary contamination, and freeze damage — and holds a granted utility model patent
- ExFire — fire suppression specifically for overcrowded urban settlements
- Root and Renew — seeders specifically for Asian smallholders, locally manufacturable
- StomaLock — ostomy care, an underserved niche, with discreet alerts as a design requirement
Why it wins: specificity defeats the originality search. "Water purifier" has 10,000 competitors; "freeze-resistant backwashing purifier for high-altitude pastoralists" has approximately zero. This is the highest-yield move available to a team with no lab and no budget.
4. Quantified findings from the 2026 cohort (n=31 described)
| Signal | Frequency | Interpretation |
|---|---|---|
| Names a specific incumbent, product, or competing technology | ~26/31 | Near-universal. Its absence is the strongest negative signal. |
| States at least one quantified performance delta | ~28/31 | Percentages, dB, litres/day, accuracy, seconds. |
| States a price or unit cost | ~18/31 | Not universal — but every price-collapse archetype has one. |
| States a sized market with a number | ~22/31 | Often nested ($216M device market inside $12.4B tourism). |
| Explicit IP claim (patent granted/pending) | 3/31 | Rare. Soaring Seeds (granted), VORTA and DeepTrust (pending). |
| Moat via data, price, retrofit, or specificity rather than patent | ~majority | The dominant defensibility strategy. |
| Uses AI/ML inside the product | ~12/31 | Fully permitted and common — see §6. |
| Team size 2 | 11/31 | Two-person teams are the single most common size. |
| Team size 5 | 7/31 | Also common. Size is not a predictor. |
| Cross-border team (2+ countries/states) | ~9/31 | Explicitly encouraged; not required. |
Most important number here: only 3 of 31 finalists claimed a patent. Teams routinely believe Conrad requires patentable IP. It does not. The rubric's own wording lists "first mover, contracts, ecosystem capture" alongside patents.
5. Category dynamics
| Category | 2026 finalists | Character |
|---|---|---|
| Health & Nutrition | 8 | Largest and most crowded. Assistive/diagnostic devices dominate. |
| Energy & Environment | 7 | Equinor-sponsored. Broadest interpretation of "environment". |
| The Water Challenge | 6 | Special category — best odds per entrant, see below. |
| Aerospace & Aviation | 6 | Most technically demanding; space debris is a saturated sub-theme. |
| Cyber-Technology & Security | 6 | Loosest boundaries — absorbs AI, software, robotics, wearables. |
Category arbitrage is real and observable
Three 2026 finalists competed outside the "natural" category for their technology:
- BeCure (bee-mite detection) → Health & Nutrition, not Energy & Environment
- Robo Therapy (medical tremor exoskeleton) → Cyber-Technology, not Health
- Echo/Pulse (speech restoration) → Cyber-Technology, not Health
And The Bee Initiative (BeeGuard) entered the same underlying problem — hive health — under Energy & Environment, where it won Pete Conrad Scholar. Two teams, one problem, two categories, both finalists.
Interpretation: you compete against the teams in your category, not against the field. A health device in Cyber-Technology is unusual, memorable, and faces a different comparison set. This is a legitimate, rules-compliant strategic lever. Choose the category where your project is surprising, provided the fit is honest.
The special category is systematically under-entered
The Water Challenge takes the same ~6 finalist slots as the standing categories, but is new each year and has no accumulated body of "obvious" projects. Highest expected value per entrant. Whatever the 2026–27 special category turns out to be, this logic should hold.
6. AI in winning projects — the permitted use
12 of 31 finalists build AI/ML into the product: BeeGuard, BeCure, DeepTrust, Dravix, OncoMap, Echo/Pulse, MicroSeek, Clarity, Schoolace, Back Tracker, Robo Therapy, VisionBridge.
This is explicitly allowed and rewarded, and is categorically different from using AI to write your brief (prohibited). See 01-competition-facts.md §8.
But the Chief Judge's warning governs: "it's not convincing for teams to write that AI will solve a problem or create action unexplainedly." Compare:
- ❌ "Our AI analyzes hive health." — asserted, scores low
- ✅ BeeGuard: "temporal multimodal" — computer vision + acoustic + behavioural tracking, 94.7% accuracy, <1 second, detects weeks before visible symptoms, zero hardware cost
The second names the modalities, the architecture property, the metric, the latency, the lead time, and the deployment cost. That is what an AI claim has to look like.
7. Coaches compound
Two observable signals that coaching quality persists across seasons:
- Karin Sempf — Teacher of the Year 2025 (Inno-Vanation Panama); coached TermoPaint, a 2026 Citizenship Award winner.
- Aashna Saraf — 2026 Excellence in Education Award, coaching two teams (ExFire, TagAlong), one of which reached the finalist roster.
- Several finalist coaches share surnames with team members (Kummer/Aphelion, Davay/AeroLattice, Gupta/Flux Orbit, Suravarjjala/Dravix, Polapally/Clarity) — parent-coaches are common and successful.
Implication for a mentoring program: the coach relationship is durable and the returns accrue over multiple seasons. Build for the second year from the start.
8. Anti-patterns — what does not appear among finalists
Cross-referenced against the five judged briefs in 02-judging-playbook.md:
| Anti-pattern | Why it fails |
|---|---|
| Generic sustainability framing without a specific user | Fails the originality search; Impact is unquantifiable |
| "We'll copyright our code so it can't be replicated" | Signals IP illiteracy (SUNSYNC). IP Defensibility is an explicit sub-criterion. |
| Competitive analysis only against weak alternatives | The judge picks the comparison set (Puppy WC) |
| Advanced-sounding technology with no differentiation argument | Sophistication ≠ innovation (SUNSYNC) |
| Impressive arithmetic that violates conservation of energy | The reliable ~75-point ceiling (P-Bump, Ocean Energy) |
| Prototype built but never measured | Proves access and non-curiosity (SUNSYNC) |
| Duplicated answers across questions | Reads as abandonment (Greencrete) |
9. How to use this file
- Topic generation →
topic-engine/METHOD.mdoperationalises §3 (seven archetypes) and §5 (arbitrage). - Brief writing → §2's template is the target sentence for Q1 and Q4.
- Self-assessment → §4's signal table is a checklist: how many do you hit?
- Refresh → after each Summit, append the new finalist roster to
data/winners.jsonl(procedure in01-competition-facts.md§12) and re-run the frequency counts in §4.