Structures that report themselves.

A measurement platform for materials that sense their own load. We built it on self-sensing concrete, because concrete is the hardest case and the biggest market. The pipeline does not care what the material is.

BEDROCK ANALYTICA

DNHacks 2026 · Energy and Industrialization · Washington, D.C.

THE PROBLEM

We are pouring more infrastructure than at any time�in a generation, and measuring it the way we did in 1930.

$77.7B

US data centre construction starts, 2025

up 190% year on year

86 GW

New generating capacity planned for 2026

a record year

~40%

of 2026 data centre projects tracked as delayed

operators dispute the figure

ConstructConnect Feb 2026 · EIA 2026 · FT / SynMax 2026

THE PROBLEM

The test we rely on was called unsafe in 1981,�and the sensors we install keep dying.

1981

A federal investigation after eleven workers died at Harbour Cay found lab-cured cylinders should not schedule formwork removal. Surfside failed the same way in 2021.

NBSIR 81-2374

5 ratings

is what the same bridge drew from different inspectors. Only 68% landed within one point of each other.

FHWA-RD-01-020

19 days

is how long cracks forty times the acceptable width grew at FIU, with the road open.

NTSB HAR-19/02

9.5%

of vibrating wire gauges died during the pour on one MnDOT bridge. The corrosion subsystem produced no usable data in a decade.

MnDOT 2020-19

You cannot industrialize�a process you cannot measure.

In manufacturing every part is measured as it is made. The built environment has no equivalent, for any material.

MANUFACTURING

Every part

measured as it is made

THE BUILT ENVIRONMENT

A few samples

tested in a lab, weeks later

WHAT WE MEASURE

Continuously

from the pour onward, in place

SOLUTION · THE MECHANISM

The material is the sensor.

Put a conductive network inside a material and it becomes its own strain gauge. Load it, contacts close, resistance falls. We read that, and the reading is the measurement.

Carbon fibre in concrete is our first case. The pipeline reads any material whose resistance moves with load.

WHAT IS ACTUALLY NEW

Not the chemistry. The path from a raw reading to a decision on a member of a real structure.

Our model of the specimen under load. Damage initiates at the interfacial zones.

LIVE DEMO · 01

Tap the cube. The bridge responds.

Real signal

Live resistance from the specimen on the table, not a simulation.

One member, not a category

2.6 million surveyed points segmented into 28 addressable members. The cube maps to one. It goes critical on screen.

Hazard logged

The threshold crossing writes an event and the inspection record updates.

TECHNICAL DEPTH · 01

We did not fit a curve. We solved the geometry�and it landed on the published law.

0.129

pp

worst disagreement, over the full 0 to 4000 µε range

AT 4000 µε

Published gauge-factor law −9.200 %

Our resistor lattice −9.071 %

Every voxel of the 32³ specimen is a node. 32,768 nodes, conductance set by phase, one volt across opposite faces, solved as a sparse graph Laplacian.

One free parameter — how fast a tunnelling gap closes under load.

The geometry and the literature arrive at the same answer independently.

Gauge factor 23 from the carbon-fibre cement literature (Chung et al.). Not fitted here.

TECHNICAL DEPTH · 02

At a realistic fibre dosage our own model says�the network does not conduct. We shipped that.

zero

conductive paths span the specimen

21 separate clusters at 1.59% fibre by volume, inside the range used in practice

SAME MIX, SAME DOSAGE

52 111 62 94

conducting voxels, four independent pours

A 2.1× spread. None of them spanning.

We swept all 260 dosages and found spanning starts at 3.84%.

A binary voxel model cannot represent tunnelling. Two fibres a micron apart land in different voxels and read as disconnected, so the model understates connectivity.

That is why the gauge factor comes from the literature and not from this lattice.

LIVE DEMO · 02

Same signal, same physics, three asset classes.

3,826,902

surveyed points

273

instrumented members

Bridge 2.6M, laser scan · Dam 755K, photogrammetry · Data hall 470K, 213 members over three levels

Ten level-three slab bays are critical and the three footings beneath them are warning. That is a load path failing top down, not ten unrelated sensors.

TECHNICAL DEPTH · 03

Every monitoring product reports a number.�Ours reports what it cannot see.

64×

One 120 mm electrode pair stands in for an 8 m member. Computed from the model, not asserted.

5 of 10

ontology classes carry no gauge at all. We publish the list rather than implying full coverage.

null

is what two data-hall members publish instead of a strain figure, because their temperature channel is dead.

Carbon-fibre cement is also a thermistor: 2,290 µε of apparent strain per degree, so 1.75 °C with no load reads full scale. An uncompensated reading there is not a degraded measurement. It is not a measurement.

FEASIBILITY · 01

A state DOT already funded this�at our exact dosage.

40.61

gauge factor, at 0.5 wt% carbon fibre

Highest compressive strength of every mix they tested. Reduced chloride ingress, low rebar-corrosion risk. Commercial milled fibre.

WHAT THEY DID NOT DO

Their specimens were mortar, not structural concrete, and their resistance moved only 189 to 178 kilohms.

Nobody has taken a self-sensing pour through to a member-level decision on a real structure.

That is the gap we built into.

Louisiana DOTD / FHWA, report LA.24/24-4TIRE, July 2024

FEASIBILITY · 02

We are not asking for a new code.�ACI has allowed this since 1983.

“Since 1983, ACI 318 has permitted the use of in-place testing as an alternative to testing field-cured cylinders … provided there are sufficient correlation data.”

ACI 228.1R-19 §1.3

Seven

accepted in-place methods. None of them is electrical. That is the gap.

Colorado

already forbids acceptance cylinders for form removal and mandates maturity meters.

AASHTO

lets measured strain multiply a bridge's official load rating. A strain reading already changes a regulated number.

FEASIBILITY · 03

It ships inside processes that already run.

~0.5 wt%

conductive fibre by weight of binder. Nothing about how a structure is designed or built changes.

$7K + $4K

one-time per instrumented segment, then per structure per year.

[N]

companies interviewed — precast, 3D-printed construction, offshore manufacturing. Every one named the same problem.

WHO BUYS FIRST

Precast. Weeks-to-months production cycles, and 41 state DOTs already recognise NPCA plant certification — one door instead of 41.

Then data centres. $75.2B annualised, up 57.2% year on year, while general office construction fell 11.3%.

We add to a mix they already run. The integration is the process, not a retrofit.

America does not just need to build more.�It needs to know what it has built.

BEDROCK ANALYTICA