Your machine passed every test and failed anyway. Find out why.

A fixed-fee audit of your vibration or acoustic data. The decision rule is frozen and dated before your data arrives. The fee is the same for either answer. The verdict can be rerun on your own machine, bit for bit. The question it answers: does the phase, timing, or relational channel carry value on your machines, yes or no, and how much.

The offer

A Phase/Timing Ambiguity Audit. Send us your vibration or acoustic data, one messy diagnostic dataset or known failure case. You get back a signed verdict certificate that answers one question: does the phase/relation channel carry value on your machines, yes or no, and how much.

The certificate is a physical record of the discipline, not just the answer. It states the question asked, the decision criteria and the date they were frozen, the verdict, what the verdict buys or saves you, the conditions under which the answer would change, and a reproduction pack: code, seeds, and data spec, so your own engineer can rerun the whole audit and get the same result.

The phase lane is recommended only when the full audited procedure confirms it on data the analysis never saw while it was being built (held-out, leakage-guarded data). A preliminary screen is used for triage only and is never the verdict. If the procedure says no, the report says no, in writing, and you stop spending money on that question.

Fixed fee, $2,000 to $10,000 depending on scope. About one week. Same fee for yes or no.

Not ready to commit? The $100 sample. Send a small slice of your data. We check whether it can support a full audit at all: capture length, channel count, session structure, sample rate. You get back a short written answer: usable as is, usable with a specific different capture (we tell you exactly what to record), or not a fit. It comes with a real sample verdict certificate from a public benchmark dataset, so you hold the exact deliverable before spending real money. The $100 is credited toward the full audit if you proceed. If you walk away, you walk away.

The sample tells you whether your data can support the audit. It is not the audit, and it does not predict what the verdict will be.

Start a $100 sample

The degree audit (included)

Every audit also reports where the useful signal in your data actually lives: in single sensor channels on their own, in pairs of channels compared against each other, or only in the way three or more channels move together (what a statistician would call the interaction order). This tells you what class of analysis, and what class of sensor, your problem actually needs, before you buy either.

Why this is different

The incentives are built so you do not have to trust us. The fee is flat and identical for either answer, so nothing rides on which answer comes out. The decision criteria are frozen and dated before your data arrives, so the goalposts cannot move after the fact. And our misses are published alongside our hits, with causes named.

On every public benchmark dataset tested to date, the audit has never recommended paying for a phase lane that was not there. Zero false positives on that record.

Where our current feature set cannot see, we say so: the known blind spot is named in every report as a limitation, and the fix for it is pre-registered, not promised. Our no is proven hard. Our yes is deliberately conservative.

Most condition-monitoring vendors make money when you buy more monitoring: sensors, subscriptions, dashboards. An audit is the opposite. It is a one-time answer, and most of the time the answer is no. A screen that usually says no is the only kind whose yes is worth paying for.

ARIES was founded in 2026 and has no client case studies to show yet. Early clients are buying a constrained, reproducible procedure, not a track record of logos, and the price reflects that.

Our track record, including the misses

This is the section most vendors would leave off their website. Everything below is from public benchmark datasets, so you can check it.

The screen usually says no

We reran a public pump and motor fault dataset through the grouped audit path. Three condition pairs were tested. All three client answers were no: do not pay for the extra phase/relation lane here.

NO NO NO CONTROL normal vs misalignment normal vs looseness misalignment vs looseness known yes/no checks "standard tools enough" "standard tools enough" "standard tools enough" "audit behaves both ways"
The grouped public dataset rerun gave three no verdicts. Synthetic controls checked that the procedure says no on a magnitude-only case and yes on a known phase case.

What a no looks like

A public dataset of pump and motor faults: normal running, misalignment, and mechanical looseness. We ran standard analysis and the extra phase/relation channel side by side. To keep the test honest, all data from one recording session stayed on one side of the fence, so the analysis was never graded on data it had already seen during setup (grouped by capture session, the standard guard against leakage).

Standard analysis 0.746 Phase / relation 0.908 a coin flip scores 0.50 The report still said no. The locked rule read the useful information as signal strength, not phase. One tall bar on a small dataset does not overrule it.
Public benchmark, normal vs looseness: 100 data windows from 10 recording sessions, scored with sessions kept separate. Scrambled copies of the data with the timing wiped out (phase-destroyed surrogates) scored 0.579 to 0.755, so a score in that band can happen with no real phase information at all. Small public dataset, so exact scores carry uncertainty; the client verdict is no.

Why the no stands even though the red bar is taller: before we recommend paying for the lane, the result has to hold up on data the analysis never saw while it was being built, under criteria frozen in advance. On this data it did not. A single flattering number does not clear that bar, and it is not supposed to.

The misses are on the record too

The audit's current feature set has produced known false negatives on public benchmarks: cases where value existed and the procedure did not see it. Each one is recorded with its cause named, and the fix is pre-registered. We would rather show you that record than pretend it is empty.

We audit ourselves the same way

In one recent week, this same process caught five errors in our own work before anything shipped: an overstated claim, a windowing artifact that changed a headline number, a contaminated validation set, an undefined scoring criterion, and an incomplete internal ledger. Each is documented and dated. A process that cannot catch its own errors has no business ruling on yours.

A validation program is underway: a pre-registered bar of independent dataset pairs, run and published as they complete. Progress is stated as it stands. It is not stated as finished, because it is not finished.

Who it is for

Motor and pump rebuild shops with repeat failures that never show a clean cause. Maintenance and condition-monitoring teams sitting on ambiguous calls. Integrators who inherit machines the sensor data never quite explains. Acoustic teams with data that will not give a straight answer.

How it works

Standard magnitude analysis Phase / relation analysis Same statistical null tests Signed verdict yes / no / how much
How an audit runs. Both pipelines face the same null tests; the difference between them is measured, not asserted. Process only, no result shown.
  1. Before your data arrives: the decision criteria are written, frozen, and dated. They do not move afterward.
  2. Send one dataset or describe the failure case. A short call scopes it.
  3. The audit runs standard magnitude analysis and phase/timing analysis side by side, with the same honesty checks (statistical null tests) on both.
  4. You get a signed verdict certificate: what was asked, what was found, what was tested, what the data cannot prove, what would change the answer, and the reproduction pack. Then you decide with evidence.

Both analyses face the same honesty checks. The key one: scrambled copies of your data that keep the signal strength but wipe out the timing (surrogate-data tests). If the phase analysis cannot beat the scrambled copies, whatever it found was never phase. Any difference between the two pipelines is therefore attributable and quantified, not asserted.

Every report states its limits. The audit is decision support, not a safety system, and it will tell you when the honest answer is that nothing is there.

How your data is handled

An NDA is available before anything is shared. Your data is used for your audit and nothing else: no resale, no benchmarking, no training datasets. On completion, your data is deleted; the signed report is yours.

The audit letter

An occasional email when something real happens: a new public-benchmark result goes on the track record, the validation program advances, or the audit changes. Misses included, same as this page. No marketing cadence, no resale of your address, unsubscribe in one click.

Or just email liam@ariesphotonics.com and say "add me."

About

Liam Sparling, founder of ARIES Photonics

ARIES Photonics is a one-person measurement company in Salt Lake City, founded by Liam Sparling in 2026.

He is a self-taught optical physicist, artist, author, and inventor. Physics has been a lifelong pursuit, and ARIES Photonics is its direct result.

ARIES exists to see what isn't there: a tool for finding the answers hidden in phase, the part of a signal that standard analysis throws away.

Contact

Liam Sparling, ARIES Photonics. Salt Lake City, Utah.

liam@ariesphotonics.com