Where you work shapes how you work.
Why Perch exists
Perch started with a person driving to small towns around Orlando on weekends, ordering a coffee, opening a laptop, and quietly noting: how's the Wi-Fi, really? Can I hear myself think? Will anyone mind if I stay three hours? Where do I even park?
Nobody was collecting those answers. Reviews said the lattes were great. They never said whether you'd get an outlet.
So we started measuring: a speed test here, a decibel reading there, a note about the house rules. Perch exists because remote workers deserve better than guessing.
Here's the thing the coffee-shop boom quietly proved: it was never about the coffee. A cafe is a third place, not home, not the office, and people go for the ritual, the hum, the feeling of working among others. That's also why star ratings fail you there: the rating measures the emotion, and it can't tell you whether the Wi-Fi holds. Perch is the rational instrument inside that emotional ritual. We guide you to third places, and we bring the numbers the feeling can't.
Curious how the score is built? The full methodology is below ↓
What we believe
-
Measurement is a form of respect.
Opinions are cheap; readings are care. When we publish "387 Mbps, 60 dB at 5pm," we're respecting your time enough to have actually checked. We'd rather list 40 venues we've truly measured than 4,000 we've scraped.
-
Honesty over inventory.
A stale score is worse than no score. Our data shows its age, decays visibly, and admits what it doesn't know. "We're not sure anymore" is a feature.
-
The score is sacred.
No venue can buy a point. Not now, not at any revenue stage, not ever. Certification attests to quality; it cannot manufacture it. The day a Perch Score can be purchased is the day Perch stops meaning anything.
-
Time matters as much as place.
A cafe isn't good or bad: it's good Tuesday at 9am and impossible Saturday at noon. We score moments, not just addresses, because that's how reality works.
-
Venues are partners, not products.
Cafes tolerate, and sometimes welcome, people who occupy a table for four hours on a $6 spend. We honor that bargain: we surface the house rules, we encourage buying generously, we send venues their data freely, and we never treat a small business as inventory to be rated and discarded. A good perch survives only if the tree stays healthy.
-
Community data, community benefit.
The people who measure are the platform. Contributors get credit, reputation, and a voice in the methodology (public, criticizable, and versioned). We will never sell contributor data, and the core data stays free to everyone, forever.
-
Depth, then breadth.
We go to a new city only when we can keep our promises there. An empty Perch city is a broken promise wearing a map.
-
Calm software.
No engagement bait, no streaks-for-streaks'-sake, no notification spam. Perch helps you find a spot, then gets out of the way. The product succeeds when you close it and get to work.
The name
A perch is where a bird lands: settled, alert, at ease, ready. That's the feeling we're building toward: you walk in, the spot is exactly what the score promised, you open the laptop, and the day goes well. Land somewhere good.
What Perch refuses to become
- A pay-to-play directory.
- An ad farm wrapped around crowdsourced labor.
- A ghost town of 2019 reviews.
- A growth-at-all-costs map of cities we've never measured.
Each of these is a real failure mode of a real predecessor. We name them so we can refuse them deliberately.
Methodology v0.1
Measured. Not guessed.
The Perch Score is a 0–100 workability index built from instrumented readings (speed tests, decibel meters, outlet counts), not vibes. This section is the whole recipe. It's public, criticizable and versioned, because a score you can't interrogate is just a star rating with better typography.
Four design principles
- Measured beats opined. Instrumented readings (Mbps, dB) outrank star-ratings.
- Time-aware. Every venue gets daypart scores: weekday AM, weekday PM, evening, weekend. A cafe at 70 dB on a crowded Wednesday morning is a different venue at 3pm.
- Fresh or flagged. Data older than 90 days starts to decay; older than 12 months, the score greys out and asks to be re-verified.
- Never for sale. No paid product may ever move a score. Certification can attest to a score; it cannot raise one.
Six pillars, weighted
| Pillar | Weight | What it captures | Primary inputs |
|---|---|---|---|
| Wi-Fi | 25 | Speed, stability, access friction | Measured Mbps (speed test), password/cutoff policy, drops |
| Noise | 20 | Acoustic workability | dB reading + qualitative type (music vs. chatter vs. grinder) |
| Crowd & seating | 20 | Will you get a seat? Can you stay? | Observed occupancy %, seat count, table size, turnover pressure |
| Power & comfort | 15 | Sustainability of a 3-hr session | Outlet density, seat ergonomics, table height, AC, lighting |
| Parking & access | 10 | Can you get there? | Ease 1–5, cost, bike/transit options |
| Affordability | 10 | Cost of squatting politely | Min spend, cheapest "rent-a-seat" item, refill policy |
Weights sum to 100. Missing pillar data defaults to a neutral 60 and is flagged, never hidden.
Normalization curves
Raw readings become 0–100 pillar scores through published curves, interpolated between these anchor points, log-scaled for Wi-Fi.
Wi-Fi (Mbps → 0–100)
| Mbps | Score |
|---|---|
| ≤2 | 5 |
| 5 | 25 |
| 10 | 45 |
| 25 | 70 |
| 50 | 85 |
| 100 | 95 |
| ≥200 | 100 |
A log curve with a plateau: work needs reliability, not headroom. Qualitative-only ratings (no speed test) cap at 90 and carry a confidence penalty. A drop during a session costs 15.
Noise (dBA → 0–100)
| dBA | Score |
|---|---|
| ≤50 | 100 |
| 55 | 85 |
| 60 | 70 |
| 65 | 50 |
| 70 | 35 |
| 75 | 15 |
| ≥80 | 0 |
Predictable noise is more workable than the number alone implies: steady ambient hum +5, music with lyrics at volume −5, loud TV −10.
Parking & access (ease 1–5)
| Ease | Score |
|---|---|
| 1 | 10 |
| 2 | 30 |
| 3 | 55 |
| 4 | 80 |
| 5 | 100 |
Paid-but-easy ≈ 4. Walkable cities will substitute an access score.
Affordability (min spend)
| Min spend | Score |
|---|---|
| $0 | 100 |
| $4 | 85 |
| $5 | 80 |
| $6 | 70 |
| $8 | 55 |
| $10 | 45 |
| $15 | 25 |
Counts the cheapest socially acceptable spend for a multi-hour stay.
Crowd & seating: under 40% occupancy scores 90–100; 40–70% scores 60–85; over 70% drops to 40 or less; "couldn't find a seat" caps at 20. Big venues absorb crowds, so seat count adjusts the result.
Power & comfort: outlets at more than half the seats scores 90+; about a quarter, 60; almost none, 30 or less, with ±10 for ergonomics (real tables vs. low coffee tables).
Policy modifiers
The house rules move the final number, after the weighted sum:
- No-laptop hours or a weekend ban → score capped at 49 for the affected dayparts
- Wi-Fi time cutoff under 2 hours → −8
- Explicitly laptop-hostile signage → −15
- Notably welcoming (laptop tables, posted Wi-Fi, "campers welcome") → +5, maximum
Confidence & freshness
Verified: instrumented readings (speed test + dB) within 90 days.
Reported: structured user submissions, no instruments.
Stale: last data over 12 months old. The score greys out and asks to be re-verified. A stale score is worse than no score.
Beyond 90 days since the last verified reading, the displayed score loses 1 point per 30 days. Re-verification restores it. Missing pillars default to a neutral 60 and cost confidence.
Worked examples: the founding field data
Three Orlando cafes, measured in person in June 2026. This table is the calibration set: if the model's verdicts don't match the experience of actually working in these rooms, the weights get tuned. Ground truth wins.
| Pillar (wt) | Haraz Coffee | Qreate Coffee | Lineage Coffee |
|---|---|---|---|
| Wi-Fi (25) | 5/5 qual. → 90 | 387 Mbps → 100 | 95 Mbps → 95 |
| Noise (20) | 70 dB → 35 | 60 dB → 70 | 70 dB → 35 |
| Crowd (20) | not logged → 60* | 5pm, light → 85 | crowded → 30 |
| Power/comfort (15) | not logged → 60* | not logged → 60* | not logged → 60* |
| Parking (10) | 2/5 → 30 | 3/5 → 55 | 5/5 → 100 |
| Affordability (10) | $8 → 55 | $6 → 70 | $5 → 80 |
| Perch Score | Reported | Verified | Verified |
* not logged → neutral 60 default. Sanity check: Qreate is the clear workhorse, Lineage is "great if you go off-peak", Haraz is "fine but loud and parking-annoying."
The field protocol (Scout kit)
Per visit, under 4 minutes:
- Speed test screenshot (fast.com)
- 60-second dB reading (NIOSH SLM or Decibel X), from a seat
- Occupancy estimate: seats taken / total
- Outlet count per 10 seats
- Minimum polite spend
- Parking ease, 1–5
- Timestamp, one photo, a one-line note on the house rules