Tsoden
Public audit for poissonlabs.ai

What we found

Mapping where a quadruped RL policy fails, with a live probe is 7 fixes away from more research partnerships.

7 things a visitor needs are missing. The ones costing you the most: no obvious button to start, Search engines can’t tell what you sell and no easy way to get in touch, plus 4 more below.

Read from 5 live pages · metadata, crawlability, trust, CTA, schema, sitemap, robots, llms.txt

Questions about this audit? Reply to the email or write to us — a human answers same day.

7 fixable leaks

The complete, prioritized plan to fix all 7 leaks above and win more research partnerships — written for Mapping where a quadruped RL policy fails, with a live probe, delivered instantly.

Book a free 30-min call See what's leaking ↓

What buyers and AI/search engines see right now

Pages checked55 read in full
How we read itRead as crawlers doSame view buyers & bots get
Response time822msUnder the 2.5s target
Checks verified6/137 need fixing

Diagnostics

What we checked, and what we found

40 checks across six areas — search visibility, AI & answer engines, performance, security, conversion and trust — each measured against Mapping where a quadruped RL policy fails, with a live probe's live site.

17 pass 7 warn 16 fail

Of the 40 checks we ran on your page

Anything we couldn’t read is its own segment — a page that blocks our reader is our limitation, not a mark against you, and it never lowers your score.

Fine as-is — 17Needs fixing — 16Couldn’t read — 7
See the numbers
StateCountShare
On your page1743%
Missing1640%
Couldn’t read718%

Where your page is strong, and where it isn’t

Each area is scored 0–100 against what a customer needs to find before they act.

strong (75+)close (50–74)needs work (under 50)
Can they take the next step 36
Reasons to believe you 0
Can people find you 69
Can AI assistants read it 0
Speed on a phone 67
A safe, trusted connection 69
See the numbers
Part of your pageScoreHow that reads
Can they take the next step36/100needs work
Reasons to believe you0/100needs work
Can people find you69/100close
Can AI assistants read it0/100needs work
Speed on a phone67/100close
A safe, trusted connection69/100close

Priorities

Fix these first

Ranked by revenue impact. Each one is something labs and startups — or an AI answer engine — need to see and currently can't. Fixing the top few moves your score the most.

01

No obvious button to start

Add a clear call-to-action for requesting consultations on our research services.

02

Search engines can’t tell what you sell

Add structured data (LocalBusiness / Organization schema: address, hours, phone, ratings) so search engines show you with rich results.

03

No easy way to get in touch

Streamline the contact process to facilitate easier inquiries from interested labs and startups.

04

AI assistants have nothing to read

Publish /llms.txt with a concise machine-readable product summary, audience, pricing, and docs links.

05

Nothing on the page proves you’re a real business

Add the trust signals labs and startups look for — reviews, real cases or testimonials, credentials, and a privacy page.

06

Your page title doesn’t say what you do

Make the page title more descriptive of our research focus.

07

Search engines have no map of your site

Publish /sitemap.xml and submit it to search engines.

What we read on your page

What a customer finds — and doesn’t

3 of the 8 things we could check are already on your page. We found only 3 of the 8 things a customer looks for before they act. Most visitors are leaving without enough reason to call, book or buy.

Below is the breakdown: what's already working (with the exact text quoted from your page as proof), and the highest-impact additions to win more research partnerships — in priority order.

Scope: an automated scan of the 5 pages we could read across your site. Every “found” item quotes the exact text on your page so you can verify it. It doesn't judge copy quality or design.

Already on your page

  • A headline that says what you do — found: “10,924x: The Instability Bomb at 1.7B Scale” (on /research/mhc-reproduction-part-2)
  • A concrete number, not an adjective — found: “2.0+” (on /research/mhc-reproduction-part-2)
  • A price, or at least a from-figure — found: “$42” (on /research/map-the-failure-boundary)

What to add, in order of impact

  • An obvious button that starts the thing — Put one clear action button on the first screen (e.g. Start free, Book a demo).
  • Reviews or proof from real customers — Add a testimonial, customer logo, star rating, or an 'N customers' proof point near the top.
  • Something that removes the risk of trying you — Lower perceived risk: free trial, money-back guarantee, no credit card, or cancel-anytime.
  • An easy way to get in touch — Give an easy capture path: short form, booking link, or contact route from the first screen.
  • Answers to the questions people ask before buying — Answer the top 3-5 buyer objections in an FAQ to remove hesitation before the CTA.

Deliverable

The full write-up is ready

We've already done the full work-up for Mapping where a quadruped RL policy fails, with a live probe — every fix written out, with copy and code ready to paste. Here's what's inside:

Complete fix plan (every issue, step by step)

1. No obvious button to start

Add a clear call-to-action for requesting consultations on our research services.

  1. Put one primary button on the first screen with action text (e.g. the suggested CTA below).
  2. Repeat the same CTA after the proof section and in the footer.
  3. Make the button visually dominant — one accent colour used only for the primary action.

2. Search engines can’t tell what you sell

Add structured data (LocalBusiness / Organization schema: address, hours, phone, ratings) so search engines show you with rich results.

  1. Add JSON-LD using the schema snippet below, edited with your real details.
  2. Validate it in Google's Rich Results Test before publishing.
  3. Keep it in sync with the on-page facts (name, address, phone, offers).

3. No easy way to get in touch

Streamline the contact process to facilitate easier inquiries from interested labs and startups.

  1. Add a visible booking link, short form, or tap-to-call number on the first screen.
  2. Keep the form to 3 fields max; every extra field drops completion.
  3. Confirm submissions with a clear success state, not a silent reload.

4. AI assistants have nothing to read

Publish /llms.txt with a concise machine-readable product summary, audience, pricing, and docs links.

  1. Publish /llms.txt with a plain-text product summary, audience, pricing and key links.
  2. Keep it concise — answer engines reward clear, factual structure.
  3. Update it when your offer or pricing changes.

5. Nothing on the page proves you’re a real business

Add the trust signals labs and startups look for — reviews, real cases or testimonials, credentials, and a privacy page.

  1. Add 2–3 real testimonials or case results with names/logos near the CTA.
  2. Link a privacy policy and terms page in the footer.
  3. Show any credentials, guarantees, security or compliance badges you hold.

6. Your page title doesn’t say what you do

Make the page title more descriptive of our research focus.

  1. Write a 50–60 character <title> in the format: Primary outcome for [audience] — [Brand].
  2. Put the most-searched term first; keep the brand at the end.
  3. Make every important page's title unique — never reuse the homepage title.

7. Search engines have no map of your site

Publish /sitemap.xml and submit it to search engines.

  1. Publish /sitemap.xml listing every indexable URL with <lastmod>.
  2. Reference it from robots.txt and submit it in Google Search Console.
  3. Regenerate it automatically whenever pages change.

Done-for-you assets — copy & paste

SEO title
Where to paste: Replace the <title> tag inside your page <head>.
Poisson Labs - Advanced Research in AI Systems
Meta description
Where to paste: In <head>, set <meta name="description" content="…"> to this (replace the old one).
We specialize in multi-agent RL and adversarial simulations, helping teams innovate with robust AI systems. Explore our research today!
H1 — main page heading
Where to paste: Make this the single <h1> at the top of the homepage (real text, not an image).
Innovative Research in Multi-Agent Reinforcement Learning

— or one of these —
- Transforming AI Through Advanced Adversarial Simulations
- Pioneering Systems That Learn: From Research to Reality
Primary CTA button
Where to paste: Use as the main button on the first screen; repeat it after your proof/services section.
Explore Our Research
JSON-LD structured data (Organization)
Where to paste: Paste this entire block right before </head> (or before </body>).
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Mapping where a quadruped RL policy fails, with a live probe",
  "url": "https://poissonlabs.ai/",
  "description": "Poisson Labs is a research group focused on developing advanced AI systems through innovative methodologies.",
  "makesOffer": [
    {
      "@type": "Offer",
      "itemOffered": {
        "@type": "Service",
        "name": "Multi-Agent Reinforcement Learning Development"
      }
    },
    {
      "@type": "Offer",
      "itemOffered": {
        "@type": "Service",
        "name": "Adversarial Simulation for Robotics"
      }
    },
    {
      "@type": "Offer",
      "itemOffered": {
        "@type": "Service",
        "name": "Sim-to-Real Transfer Solutions"
      }
    },
    {
      "@type": "Offer",
      "itemOffered": {
        "@type": "Service",
        "name": "Robustness Evaluation for LLM Agents"
      }
    },
    {
      "@type": "Offer",
      "itemOffered": {
        "@type": "Service",
        "name": "Custom Research and Engineering Collaborations"
      }
    }
  ]
}
</script>
llms.txt (for AI answer engines)
Where to paste: Create a plain text file at poissonlabs.ai/llms.txt with this content.
# Mapping where a quadruped RL policy fails, with a live probe

> We specialize in multi-agent RL and adversarial simulations, helping teams innovate with robust AI systems. Explore our research today!

## Services
- Multi-Agent Reinforcement Learning Development
- Adversarial Simulation for Robotics
- Sim-to-Real Transfer Solutions
- Robustness Evaluation for LLM Agents
- Custom Research and Engineering Collaborations

## Links
- https://poissonlabs.ai/

Page-by-page findings

PageWhat to fix
/looks healthy
/research/mhc-reproduction-part-2looks healthy
/research/mhc-reproductionlooks healthy
/research/vjepa-2-1-robustnesslooks healthy
/research/map-the-failure-boundarylooks healthy

Action order

48-Hour Fix Plan

  • Add a clear call-to-action for requesting consultations on our research services.
  • Add structured data (LocalBusiness / Organization schema: address, hours, phone, ratings) so search engines show you with rich results.
  • Streamline the contact process to facilitate easier inquiries from interested labs and startups.
  • Publish /llms.txt with a concise machine-readable product summary, audience, pricing, and docs links.
  • Add the trust signals labs and startups look for — reviews, real cases or testimonials, credentials, and a privacy page.

Why it matters

The fastest labs and startups do not read every page. They scan the first screen, trust signals, pricing cues and how easy it is to request a consultation. Missing signals make the offer harder to understand and easier to ignore.

Packages

Turn this into more research partnerships

You've seen the leaks. Most owners start with the €49 full report — the same fixes we'd implement for €199, written out so you can do them yourself. Want it hands-off? Step up a tier.

Do it yourselfFree

The complete evidence report and the prioritized, step-by-step plan for Mapping where a quadruped RL policy fails, with a live probe — every fix written out with copy & code ready to paste. It is on this page. Nothing to buy, nothing to unlock.

Read the full plan ↑
Done-for-you — 48h€199

We implement the highest-impact fixes for Mapping where a quadruped RL policy fails, with a live probe — metadata, schema, CTA, llms.txt, trust signals — in 48 hours. You keep every bit of the lift, no lock-in.

Have us do it — €199
Ongoing monitor€499/mo

Weekly re-checks of Mapping where a quadruped RL policy fails, with a live probe + drift alerts so the fixes compound and never silently slip back.

Start monitoring

Not sure which fits? Book a free 30-min call and we'll tell you straight. · 48-hour turnaround · no lock-in · every claim verified.

Verification

The exact wording we found

Every signal below was read from your live site — nothing is estimated. We checked 5 pages; “couldn’t verify” means a signal needs JavaScript to read and we won’t guess.

What we read

TitlePoisson Labs
Meta description168 chars
H1 heading Found 5/5
Structured data Missing

Crawl & AI files

robots.txt Found
sitemap.xml Missing
llms.txt (AI) Missing
Response time822ms

Pages we checked

/
Poisson Labs
Read
/research/mhc-reproduction-part-2
10,924x: The Instability Bomb at 1.7B Scale — Poisson Labs
Read
/research/mhc-reproduction
DeepSeek's mHC: When Residual Connections Explode — Poisson Labs
Read
/research/vjepa-2-1-robustness
Where V-JEPA 2.1's Dense Features Hold Up (and Where They Don't) — Poi
Read
/research/map-the-failure-boundary
Map the Failure Boundary — Poisson Labs
Read

Questions about anything in here? Reply to the email, or reach us directly — we answer the same day.

Email

Who we are

Tsoden — a full-service web studio

We don't just audit — we build and rank websites end-to-end. Web development & design, plus SEO & AEO, so your business shows up in Google and in AI answers like ChatGPT, Perplexity and Gemini. This report is where we start; we can implement every fix below and grow your visibility on a full-service basis.

Web development & redesign — fast, conversion-first sites
SEO — rank in Google for what your buyers actually search
AEO — get found & cited by AI answer engines (ChatGPT, Perplexity, Gemini)
Ongoing monitoring & growth — we keep it improving
Work with Tsoden →or email support@tsoden.ai

Curious how a competitor scores?

Run the same free check on any site — the same buying signals, the same evidence quoted from the page itself. About a minute, no email needed to see the result.

Disclaimer

This service is a marketing and informational tool based on public website signals. It is not legal, financial, tax, investment, security, or professional advice. Scores and recommendations are estimates, not guarantees of revenue, rankings, AI visibility, deliverability, or business outcomes. You remain responsible for reviewing, approving, and implementing any changes and for complying with laws that apply to your business.

Full disclaimer · Terms · Privacy

This audit is from 20 Aug 2026 — we re-run it fresh when you order Mapping where a quadruped RL policy fails, with a live probe · Grade F · 37/100
Done-for-you — €199 Book a free 30-min call