Haga · Executive Deck
Independent verification for physical AI
Derived directly from published Lab test fixtures.
Haga · Overview
01 / 14
Independent verification for physical AI
We run your robot policy or generated world through a physics check in simulation, and report what breaks — privately, with the failures shown.
Stress-test. Break. Report.
Evidence figures from Lab artifacts. Financial figures from plan assumptions.
The problem
02 / 14
Physical AI is funded at scale, but the hard tasks remain unsolved.
* Sources: PitchBook, Q4 2025 Robotics & Physical AI VC Trends (pitchbook.com/news/reports/q4-2025-robotics-physical-ai-vc-trends); Stanford HAI, AI Index 2026 Ch. 2, Fig 2.7.2 (hai.stanford.edu/assets/files/ai_index_report_2026_chapter_2_technical.pdf).
Capital is real ($27.6B), difficulty is real (12.4% sim success), checking is still done by builders.
Takeaway: The gap between funding and verified capability is the opportunity.
How it works
03 / 14
Your policy goes in. A verdict with evidence comes out.
Three-state gate (FLAG/CLEAN/ABSTAIN) with per-clip measured support plane.
Takeaway: The checker refuses on thin evidence and does not assume the floor.
Proof · policy stress
04 / 14
The harness applies and measures the perturbation correctly, and the learned-policy version runs on a partner's checkpoint.
| Task | nominal | mild | moderate | severe |
|---|---|---|---|---|
| Stack | 0.900 | 0.840 | 0.840 | 0.820 |
| Door (non-monotonic) | 0.800 | 0.760 | 0.660 | 0.760 |
* Signal: Stack and Door. Lift and PickPlaceCan are insensitive controls — in the run tables, not on this slide.
Scripted OSC_POSE controller, four tasks, four tiers, 50 episodes per cell. Lift and PickPlaceCan hold at 1.000 across every tier, which is what a self-correcting controller should do. Stack moves 0.900 to 0.820. Door moves 0.800 to 0.760 with the moderate tier at 0.660, which is non-monotonic and is reported as such.
Takeaway: What this establishes: the ladder applies, the scoring measures it, and the harness reproduces. What it does not establish: policy robustness. That run needs a learned checkpoint, and it runs on a partner's own.
Proof · real footage
05 / 14
On real robot footage the gate refuses, and shows its working.
| Dataset | Tracks | Flag rate (clean) | Target | Verdict |
|---|---|---|---|---|
| towel | 67 | 0.045 | 0.020 | FAIL |
| candy | 83 | 0.012 | 0.020 | PASS |
| ziploc | 57 | 0.193 | 0.020 | FAIL |
* Run verdict: FAIL. Towel FAIL (0.045), Candy PASS (0.012), Ziploc FAIL (0.193).
15 of 207 tracks flagged across 3 datasets (target ≤ 0.020). Towel detection 1.000 on 134 injected violations from 67 clips.
Takeaway: Catching injected faults is the easy half. False-positive rate on real clean footage is the hard half — gate fails it on 2 of 3 datasets.
Proof · generative worlds
06 / 14
Generative world-model cohort
The frozen-threshold physics-consistency checker scored over three world models' held-out generations, against real control clips. Sensitivity and specificity are reported together; a specificity number is published only when the real-control cohort is large enough to have a usable denominator.
FAILThe real-control gate failed, so this cohort's flag rates carry no information
The real-control specificity is 0.000: 15 of 15 presumed-clean real clips were flagged. The generated clips' flag rates are therefore not a measurement of anything about the generated clips -- a checker that flags presumed-clean real footage as readily as it flags generated footage has not distinguished the two.
* Real-control specificity 0.000 on earlier binary gate. SAP v2 ready; re-scoring outstanding.
SAP v2 (the Scenario Assessment Protocol) is built and run on real footage: 3-state gate, per-clip support plane, gripper-derived mask, disjoint calib/test split.
Takeaway: Re-scoring this cohort is the only step left — it applies a protocol that has already run.
How an engagement works
07 / 14
Two weeks. Criterion agreed in advance. One private report.
Objection: unreleased policy is exposure. Answer: criterion agreed first, report private, method public.
Takeaway: You commission one independent check of a claim you framed, before committing hardware.
Pricing
08 / 14
Three ways to buy: an audit, a pilot, a run.
| Offer | Scope | Price |
|---|---|---|
| Design-partner pilot | Two weeks: scope and criterion frozen, the run, one private report | Set on the published ladder and agreed on a call |
| Fixed-scope audit | Foundation: 6 clips, 5 days, short report • Standard: 6 clips, 5 days, full report | Foundation $1,900 • Standard $3,500 |
| Per engagement | Policy stress or world-model physics run, criterion frozen before run | Fixed fee, quoted per engagement |
Gross margin 78–84% (one-off), 70% (continuous), computed from published variable costs.
Takeaway: Price covers compute cost; margin is arithmetic, not a hope.
Market
09 / 14
A thin verification layer under a large physical-AI build-out.
Each layer = accounts × account value. TAM 1,080×$36K, SAM 430×$18K, SOM 25–35×$25K — about one year of this plan's $70,083/mo average burn.
Takeaway: Stated plainly: sized as a small, capital-efficient business that reaches rough break-even on the accounts it can name in year three. TAM and SAM are the upside if the verification market builds out — not a share of anyone else's forecast.
Competition
10 / 14
Robocurve publishes benchmarks. Titus consults. Prismatical is solo. Haga referees.
| Player | What they sell | What Haga does that they don't |
|---|---|---|
| Robocurve — $10M seed | Open-source benchmarks and public result pages: CIs on charts, no verdicts, no price. | Freezes the criterion before the run, refuses on thin evidence, keeps the run private. |
| Titus | Evaluation as a service for industrial operators; pre-release site, no product. | Ships a product with published artifacts — failures included. |
| Prismatical | Solo thesis: sim policy evaluation plus synthetic data; personal site. | Operating company with a funded pipeline across both lanes. |
Eight players verified in the survey; five sell anything. Robocurve raised $10M from Initialized Capital.
Takeaway: Criterion-first, private, and willing to refuse to certify.
Roadmap
11 / 14
2 design-partner reports by month 4. World-model depth by month 6.
Plan spends $997,000 of $1,000,000 raised over 14 months. Remaining $3,000 unspent.
Takeaway: Round buys the milestones above and stops. Last months need larger close or smaller scope.
Team
12 / 14
One founder, self-funded, every published run carries its failures.
- IP assignment signed 2026-07-17. Equity & IP with company.
- Seeking hardware/robotics co-founder: equity only, 4-yr vest, 1-yr cliff.
- First hire month 4: ML/robotics-eval engineer.
* North America incorporation on first close; entity/jurisdiction with lead investor.
Team budget 14 months: $503,500. Headcount: 3 (founder + 2 contractors) → 4 (month 4 hire) → 5 (month 9 co-founder).
Takeaway: Money buys named people; headcount rises on published dates.
The ask
13 / 14
Raising $1.0M — burn, runway, and where it goes.
Avg burn (mo 1–12): Month 1 $53,500 • steady $78,000 from mo 10
Runway: Recomputed from burn, not asserted.
Team funded: Headcount: 3 people in months 1-3 (the founder + 2 contractors: physicist, robotics scientist), 4 from month 4 (the ML/robotics-eval engineer starts), 5 from month 9 (the co-founder on equity only — no cash added, the team line stays $38,000/mo).
Design-partner pilot: Price is set on the published ladder and agreed on a call.
Where the money goes
| Bucket | Share | Amount |
|---|---|---|
| Team (Founder + Hire + Contractors) | 50.4% | $503,500 |
| Compute (GPU) | 19.3% | $193,000 |
| Product | 9.4% | $93,800 |
| GTM | 12.1% | $121,300 |
| Legal + Buffer | 8.5% | $85,400 |
What the money buys
| # | Milestone | When |
|---|---|---|
| 1 | 2 live design-partner evaluation reports | ≤4 mo |
| 2 | World-model scoring at depth (n≥30 held out) | ≤6 mo |
| 3 | Continuous-scoring product design | ≤6 mo |
* Full 18-mo plan costs $1,309,000; last 4 months need larger close or smaller scope.
Plan spends $997,000 of $1,000,000 raised over 14 months. Remaining $3,000 unspent.
Takeaway: The allocation above is what the plan actually commits over the funded window; the rest is headroom.
The ask
14 / 14
To earn the right to raise a Series A.
Primary Contact
Mohamed Mushood Hanif · Founder