Three engagements that show the method end to end — from a symptom nobody could explain to a fix backed by numbers. Each follows the same arc: reveal → scry → foretell → scroll.
Honest note: Emrysia is a new studio. These are realistic scenarios built with the actual workbench to show exactly how an engagement runs and what you receive. They'll be replaced with named client case studies as real work lands.
The curse. A 40-node soil-moisture mesh across a European farm. Every afternoon, roughly a third of the nodes fell off the map and came back overnight. The installer had already swapped antennas twice. "It's like they only work in the morning."
The reveal. Modeled as-built, the links were fine at dawn. The pattern was temporal, not spatial — so the audit looked at timing and regulation, not geometry.
Scry. The afternoon irrigation controller woke the far cluster into a burst-reporting mode. On EU 868 — capped at 1% duty — the extra traffic pushed those nodes to ~3% duty. The radios legally backed off; the network read it as range loss.
The fix, foretold. No new hardware. A firmware change staggered the reporting window and dropped duty back under 1%. Corvus flagged the exact nodes and the offending band.
Those nodes aren't weak — they're over their duty budget after noon. Fix the timing, not the antenna.
The curse. A point-to-point backhaul between two towers modeled fine in free space but never actually connected. On paper there was margin to spare; on site, nothing.
The reveal. The link crossed a low ridge exactly at midspan. A free-space calculation can't see it — but the terrain engine runs an ITU-R P.526 knife-edge profile along the path.
Scry. The ridge clipped ~60% of the first Fresnel zone: a 31 dB diffraction loss that turned a +14 dB budget into a −17 dB dead link. The cross-section made it undeniable.
The fix, foretold. The placement search found a spot on high ground with clear line of sight to both towers. A single repeater restored +9 dB of margin — cheaper than raising either main mast.
Free space says go. The ridge says no. Put a repeater at the knoll — clear LOS both ways, +9 dB, done.
The curse. A BLE wearable, first fab run scheduled. The team wanted range and had cranked the TX power table to the top — assuming more power is always better.
The reveal. Running the regulation check against the chosen antenna showed the top two power steps put total EIRP 4 dB over the FCC cap for the band. It would have failed certification after the boards were built.
Scry. The problem wasn't the radio — it was antenna gain plus an over-eager power table. The device library flagged a compliant chip antenna already close to the footprint.
The fix, foretold. Swap the antenna and clamp the power table two steps. EIRP lands under the cap, ~90% of the range is retained, and the re-spin never happens.
You're 4 dB over the line. Two clicks down on the power table and the antenna in the BOM keeps you legal — and keeps the range. Do it before the fab, not after.
The trickier the symptom, the more useful the reveal. Tell me what's dropping and I'll show you what I see.