⚠   PUBLIC SERVICE ANNOUNCEMENT   /   THIS WEBSITE IS A WARNING, NOT AN INVITATIONNO INSTRUCTIONS   •   NO WARRANTIES   •   NO ADULT SUPERVISION

✳ ITERVENTIONS INDUSTRIES / DEPARTMENT OF UNFORCED ERRORS

DO NOT.TRY THIS.AT HOME.

I did. That is why there is a website.

Some people see a problem and buy a solution. I see a solution and wonder whether it would be more interesting with three microcontrollers, a custom enclosure, and the possibility of fire.

Welcome to the ongoing investigation into what happens when technical curiosity is permitted to operate without a functioning adult in the room. The experiments are real. The measurements are occasionally real. The financial decisions are between me and my creditors.

This is not a how-to site. It's a how-did-it-come-to-this site.

ABILITY TO BUILD IT
CONCERNINGLY HIGH
REASON TO BUILD IT
NOT LOCATED
ADULT SUPERVISION
SELF-REPORTED
FIG. 01 / EVIDENCE SEIZED FROM WORKBENCHUNALTERED PHOTOGRAPH
A three-stage electromagnetic apparatus on a crowded workbench with oscilloscope, controller, capacitor enclosure, and monitor.EXHIBIT A
THE ACCUSED
WHY?
REDACTED

At some point a perfectly ordinary hobby became a laboratory with a documentation problem.

VIEW INCIDENT REPORT #001 →
OFFICIAL POSITION

“Just because you can” is not a reason.
Unfortunately, it is a very effective starting point.

// THE EVIDENCE LOCKER / OPEN CASES

Exhibit A: the consequences.

03 DOCUMENTED / 00 LESSONS LEARNED

I present three exhibits in the case against my own judgment. Every photograph is real. Every version number represents another opportunity to have walked away.

The coil gun and its touchscreen on a blue-lit bench beside a shelf of controller revisions.ACTIVE / V04
CASE #001 / APPLIED ELECTROMAGNETIC REGRET

Gauss Sequencer

The tiny projectile that required an entire laboratory.

A perfectly serviceable electromagnetic question somehow acquired three coils, optical gates, a touchscreen, an oscilloscope, and a logging platform. Eleven runs returned no velocity reading. At least the disappointment was automated.

ORIGINAL QUESTION
Where did the velocity go?
COMPLICATION
The coil was also a brake.
5
VERSIONS
4
FAILURE REPORTS
9
PHOTOGRAPHS
Read the confession →
Color-coded printed magazines marked JK-47 holding foam pool-noodle segments beside the launcher.MOTHBALLED / V06
CASE #002 / MUNICIPAL CONSTRAINT ENGINEERING

Residential Ordinance Platform

Somehow, the ammunition got softer while the engineering got harder.

A motorized contraption for pointing things at other things. Seven versions later, it had four payloads, increasingly polite ammunition, and a camera that was also wrong. The controller eventually escaped into another project.

ORIGINAL QUESTION
Is it pointing where it thinks it is?
COMPLICATION
The camera was an accomplice.
7
VERSIONS
3
FAILURE REPORTS
23
PHOTOGRAPHS
Read the confession →
Open white doorway with blue lighting around the frame, the door swung inward, and part of the actuator visible near the floor.PHOTOGRAPHIC RECORD
CASE #003 / DOMESTIC ACCESS CONTROL

Evil Lab Secret Door

A perfectly innocent door was recruited into supervillain cosplay.

An actuator. A control cabinet. A wall display. Blue lighting. An actual EVIL LABORATORY sign. The word “secret” has been asked to leave the room.

THE BRIEF
The proprietor called it a secret door.
OPEN QUESTION
What happens when its powered controls are unavailable?
10
PHOTOGRAPHS
2025
JANUARY – FEBRUARY
Inspect the installation →
LEGAL DEPARTMENT
(ALSO ME)

The fact that something worked is not evidence that it was a good idea. It is merely evidence that I will probably do it again.

“I will happily spend a month building something I don't need & could buy on Amazon for 1/10th the build price.”

THE PROPRIETOR / WATER METER INVESTIGATION / VOLUNTARY STATEMENT

// AN EXHAUSTIVE LIST WOULD REQUIRE ITS OWN SERVER

Other incidents of recreational escalation.

ADDITIONAL EVIDENCE / OPEN FILES

Not every bad idea earns its own project page. Some merely consume a weekend, several browser tabs, and the goodwill of the household. Click to unseal the records.

SEMICONDUCTORSThe skull stickers are not test results.

A gate-drive discussion reached the arrangement of skull stickers on the hardware. They looked like achievement badges. I confirmed the assessment and requested another “Hold my beer.” The stickers were apparently easier to standardize than the terminology.

The wiring acquired modular connectors, identified coils, and labels on the scope probes themselves. Even the equipment used to question the equipment needed an identity.

No test outcome is recorded in these conversations. The stickers do not establish a successful test or protect anyone from stored electrical energy.

AIRFLOWAir needed moving. A test facility happened.

A high-flow blower and improvised ducting became a DIY wind tunnel. The innocent part was wanting to test airflow. The escalation was building a place for the testing to live.

A testing setup is documented. Calibrated precision and performance results are not established here. Rotating machinery and improvised ducting carry real mechanical hazards.

WATERThe meter declined to incriminate itself.

An unexpectedly high water bill led to inspecting the municipal meter and contemplating a homemade observation workflow. About twelve minutes later, the visible cumulative reading was unchanged. The proposed instrumentation had been outperformed by waiting beside a meter.

The unchanged display describes that interval only. It does not rule out intermittent leaks or establish the cause of the bill. A monitoring system was contemplated; no finished one is claimed.

BARBECUEDinner acquired an interoperability dispute.

The ChefIQ wireless temperature probes were good hardware. App changes were less welcome. Replacing the thermometer was an available response. Exploring third-party integrations with a Raspberry Pi 5 was the response under consideration.

The investigation is documented. A finished replacement app is not.

SOFTWAREAdult supervision was delegated to more software.

A local-model-assisted development workflow grew a hub-and-spoke agent structure, inference routing, gates, shadow tests, and metrics. Windows, WSL, and Linux filesystems also needed to agree on where things were. A small bureaucracy was assembled so one human could continue making the decisions.

The coordination is real. A net improvement in speed or reliability has not been established here.

CAMERALooking left became a service dependency.

The Raspberry Pi pan-tilt work expanded into stepper drivers, travel limits, hall-effect calibration, a Flask API, browser controls, video streaming, and face-tracking experiments. The camera mount had acquired a software stack before the question of whether it needed one was settled.

A summary of the control work, not a separate claim of another finished platform. Moving mechanisms can pinch or move unexpectedly.

NOT YET CATALOGUED: the melted voltage regulator and the stepper jitter. The debris stays. The detailed accounts are still to come.

// A VERY LIMITED INTERNAL AFFAIRS DEPARTMENT

The smoke is part of the story.

Nobody needs another beauty shot of a finished gadget. The interesting part is the part that melted, jittered, brownouted, went silent, or revealed that the original question was hopelessly naive.

01 / RECORD
The debris is the exhibit.

Dead controllers, blank readings, moving cameras. A failure mechanism is more useful than a photograph of something briefly working.

02 / DISTINGUISH
A theory is not a measurement.

Recorded observations, working explanations, and unanswered questions keep their own labels. The AI notebook does not get an exemption.

03 / DO NOT REPEAT
These are confessions.

An account of a dangerous setup is not a safety review of that setup. Photographs are evidence of what happened, not permission to try it.

Examine the dated record →
← ALL PROJECTS
PROJECT #001 ACTIVE V04 CURRENT

Gauss Sequencer

It started with a tiny slug and a magnetic field. It now occupies a bench containing three stages, optical gates, a touchscreen, and custom logging software. Eleven tests produced no velocity reading, an outcome the software documented with remarkable confidence. Five versions in, the apparatus has become substantially better at explaining why it cannot answer the original question.

ORIGINAL QUESTION
Can additional stages explain where the velocity goes?
UNNECESSARY ESCALATION
A custom sequencer, an instrumented bench, and a shelf of controllers.
RECORDED OUTCOME
Eleven empty velocity readings. Dead sensor-side boards. The latest supply change is unverified.
01

The Question

A single-stage coil gun can throw a slug. Can three coils, fired in sequence off optical gates, throw it faster — or does every stage after the first just brake what the one before it built?
02

Evolution Timeline

v00 Proof of Concept — and a name that was wrong 2025-12-08
HYPOTHESIS
A single electromagnetic stage would produce enough motion to make the question interesting. This was the hypothesis, not a statement about whether the apparatus was sensible.
WHAT ACTUALLY HAPPENED
It moved. Barely — about as hard as a strong exhale. The bank stored roughly 6 J and most of that went into heating copper rather than accelerating iron.
UNEXPECTED DISCOVERY
The build is labelled RAIL GUN. There is not a rail in it. A rail gun needs two conducting rails and an armature that shorts across them; this is a coil and a barrel, which is a coil gun. Getting the name wrong was the cheapest mistake in this project and the only one that cost nothing to fix.
WHAT CHANGED
The energy-storage subsystem was revised. The illuminated FIRE button stayed. Its contribution to the engineering was not measured.
Small first-build launcher on a white printed base, with a coil, driver board, capacitors, illuminated button, and keycaps incorrectly labelled RAIL GUN, lit by colored LEDs.
v01 Kickback — and a breadboard that had no business being there 2026-01-24
HYPOTHESIS
The switching failures might be explained by the coil releasing stored magnetic energy. The protection hardware became its own investigation.
WHAT ACTUALLY HAPPENED
The clamp worked. It worked on a solderless breadboard, six inches from cables I had already labelled FIRING CURRENT!!! Both facts are in the same photograph and it took a week to notice they were arguing with each other.
UNEXPECTED DISCOVERY
If a wire needs a warning label, the layout is already the safety system and the label is a confession. Spring contacts on a breadboard are a few milliohms each until they are a few hundred, and there is no warning between those two states.
WHAT CHANGED
The mechanical fixture and electrical packaging were revised. The breadboard was retired. This records a change; it does not certify the revised layout as safe.
Hand holding a solderless breadboard carrying the coil kickback protection circuit — diode, resistors and terminal blocks — surrounded by yellow and red silicone wire and lever-nut connectors, with a green label reading Coil Kickback Protection Circuit.
Green 3D-printed rail holding a potted red coil, a clear barrel tube and a capacitor, on a bench where several cables carry paper flags reading FIRING CURRENT.
v02 Three coils, three gates, and no number 2026-02-24 → 2026-04-04
HYPOTHESIS
Sequencing three stages off optical gates beats one large stage at the same stored energy.
WHAT ACTUALLY HAPPENED
Built the entire sequencer — arm coil 1, 2 or 3, a FIRE button, a gate chain, a run counter. By run 11 the GATE 1 TRANSIT VELOCITY field still read two dashes. Eleven firings, zero measurements.
UNEXPECTED DISCOVERY
An instrument reading -- is not broken; it is reporting. Gate spacing was shorter than the sensor's own response time, so the transit was over before the second edge existed. The measurement chain was the experiment and nobody had tested it.
WHAT CHANGED
The scope and custom log viewer became permanent residents of the bench. The energy-storage revision increased the stored energy substantially. The missing velocity measurement remained missing.
Full bench with an oscilloscope, a monitor running the Coil Gun Log Viewer, touchscreen controller, capacitor enclosure, and three-coil experimental rail under scope probes.
v03 Packaging, polarity, and a row of dead controllers 2026-07-08 → 2026-07-19
HYPOTHESIS
If the sequencer stops being a pile of loose modules, it will stop being intermittent.
WHAT ACTUALLY HAPPENED
Packaging fixed the intermittent faults and exposed a steadier one underneath. The shelf now holds an RPi Stepper PT Controller v1 and three v2s, which is two more v2s than the design called for. Two of them are on the shelf because they stopped working, not because they were spares.
UNEXPECTED DISCOVERY
The coil measures 4.3 Ω. Against a characteristic impedance of about 1.4 Ω, that makes the discharge overdamped — the coil is braking the current before the field gets interesting. This is a resistance problem that has been wearing a timing problem's coat since v01.
WHAT CHANGED
One enclosure, one connector standard, XT60 throughout. Coils marked NORTH, because polarity is not decorative. Gates moved out past the sensor response window.
Black 3D-printed enclosure labelled COIL GUN SEQUENCER holding a Raspberry Pi, a driver shield and a touchscreen, with an XT60 power lead, lit magenta.
v04 A supply of its own 2026-07-26
HYPOTHESIS
The last surviving false trigger is conducted through the shared sensor supply, not radiated into the gate line. If it were radiated, the enclosure would have killed it in v03.
UNEXPECTED DISCOVERY
Pending at the date of this entry. The separate sensor supply had been installed; its effect on false triggering had not been measured.
WHAT CHANGED
The gate sensors received a separate power supply. The change is documented; its effect on the remaining false trigger was still unmeasured at this entry.
Current rig close-up: white printed chassis, a touchscreen showing READY and ARM with per-gate run statistics, a COIL GUN SEQUENCER shield, two oscilloscope probes clipped in, and a copper coil labelled with its resistance.
03

Prototype Passport

// PROTOTYPE PASSPORT GAUSS · v04
Version
v04
Date
2026-07-26
Hypothesis
The surviving false trigger is conducted through the shared sensor supply rail, not radiated into the gate line.
Major Change
Separate power supply for the gate sensors. Effect on false triggering not yet verified.
Unexpected Discovery
Coil DC resistance of 4.3 Ω against a 1.4 Ω characteristic impedance puts the discharge well into overdamped. The coil brakes itself.
Next Experiment
Unanswered: does the remaining false trigger originate in the sensor supply, and how much of the velocity loss is attributable to coil resistance?
iterventions.com · project #001 · not finished, on purpose
04

Failure Gallery

It Says Rail Gun
v00 · 2025-12-08
EXPECTED
A rail gun, as printed on the keycap.
ACTUALLY HAPPENED
A coil gun. There are no rails. There is no armature bridging any rails. There is a solenoid and a tube, which is a different machine operating on a different principle, confidently mislabelled in three places.
Lesson: if you can't name the mechanism, you can't predict it. Every wrong intuition for the first month came from reasoning about Lorentz force on a conductor when the actual physics was reluctance pulling iron toward a field maximum.
Repeat it? Already did. The keycaps still say it.
Firing Current, Meet Breadboard
v01 · 2026-01-24
EXPECTED
A temporary test fixture, used carefully, retired quickly.
ACTUALLY HAPPENED
It stayed for weeks, six inches from cables I had personally labelled FIRING CURRENT!!! Signal-level spring contacts remained in a discharge path. Its continued operation was not evidence that it was safe.
Lesson: the label you wrote is evidence about what you already knew. Writing FIRING CURRENT!!! on a wire and then routing it past a breadboard is not a safety measure, it's a signed confession.
Repeat it? No. Nothing solderless carries discharge current now.
Eleven Runs, No Velocity
v02 · 2026-02-24
EXPECTED
A velocity in metres per second, at gate 1, on the first shot.
ACTUALLY HAPPENED
Two dashes. Then two dashes, eleven times. The sequencer armed, fired, logged the run and reported nothing, because gate spacing was tighter than the sensor's own response time.
Touchscreen controller showing the coil gun interface: arm coil selector, a large FIRE button, run number 11, a gate chain G1 to G2 to G3, and a Gate 1 Transit Velocity field reading two dashes instead of a number.
Lesson: build the measurement chain first and test it with something whose speed you already know. Eleven shots of nothing is the same data as zero shots, at eleven times the wear.
Repeat it? No. Eleven blank readings were enough.
Three v2s and a v1
v03 · 2026-07-19
EXPECTED
One controller, revised once.
ACTUALLY HAPPENED
Four boards on a shelf. Two of the v2s are there because they died, and both died the same way — sensor side, not power side, which was the clue that took another two months to act on.
Lesson: a rev graveyard is a dataset. Two identical failures on the low-voltage side of a high-voltage machine was the answer to the false-trigger question, sitting in plain view, unread because it looked like clutter.
Repeat it? No. v2 is frozen and the dead ones got autopsied.
05

Data Notebook

GATE RINGING / CONCEPTUAL VIEWNOT TO SCALE
ILLUSTRATIVE RECONSTRUCTION / NOT A RAW CAPTURE
A conceptual picture of how ringing can create extra threshold crossings. This generated trace illustrates the mechanism; it does not verify what occurred in a particular run.
XY · VELOCITY vs CHARGE VILLUSTRATION
ILLUSTRATIVE RECONSTRUCTION / NOT A RAW CAPTURE
The notebook reports diminishing velocity improvement as stored energy increased. The normalized curve here is illustrative. Raw numerical samples are not supplied with this figure.
The original v00 build again, shown for scale: one coil, capacitors, and an acrylic barrel on a printed base.
gauss_v00_20251208_img_01.jpg
v00, kept on the shelf for scale. Six joules, one stage, and a name that was wrong.
Night bench under blue light: the sequencer enclosure and touchscreen beside the launcher rail, with a coil labelled 4.3 ohm and NORTH, gate flags, and a row of printed pedestals labelled RPi Stepper PT Controller v1 and v2.
gauss_v03_20260719_img_01.jpg
Stage 1 coil, marked with its measured DC resistance and its field polarity. Both marks exist because both were guessed wrong once.
Frosted sensor-supply enclosure with a handwritten version date and component labels.
gauss_v04_20260726_img_01.jpg
The v04 sensor supply. Dated in the version field because a rev number would have been a lie by the second build.
// BENCH NOTEBOOK · p.312026-07-19
Why the coil is its own brake
RECORDED: coil resistance 4.3 Ω
MODEL: damping ratio approximately 1.52
INTERPRETATION: overdamped response
OPEN QUESTION: what explains the velocity loss?
MARGIN NOTE

The resistance measurement undermined four versions of assumptions about timing. The damping estimate is a model, not a captured current waveform. It does not by itself establish when the field peaked relative to the projectile.

→ Better question: is resistive damping the dominant loss, or merely the next explanation that needs testing?

06

AI Lab Notebook

EXPERIMENT 01 Why gate 1 reported nothing for eleven runs
QUESTION
The sequencer logged 11 complete runs and every transit-velocity field came back empty. Is the sensor dead, the firmware wrong, or the geometry impossible?
DATASET
11 run logs with timestamps at each stage trigger, plus the gate sensor's own datasheet rise and fall times.
AI ANALYSIS
Gate separation divided by even a conservative slug speed gives a transit shorter than the sensor's specified output rise time. The firmware was waiting for a second edge that physically could not form. Nothing was broken; the geometry was asking for a measurement the parts cannot make.
DECISION
The measurement geometry was inconsistent with the sensor response time. The velocity claims could not be trusted until the measurement chain was independently validated.
EXPERIMENT 02 Where the velocity plateau actually comes from
QUESTION
Exit velocity stops improving above about 360 V. Is the limit the switch, the coil, or the slug?
DATASET
Measured coil DC resistance and inductance per stage, bank capacitance, and the exit velocities from the 11-shot series across charge voltages.
AI ANALYSIS
The estimated damping ratio is about 1.52, consistent with an overdamped response. The analysis proposed resistive damping as an explanation for the velocity plateau. The estimate alone does not establish current timing relative to projectile position or exclude other losses.
DECISION
Coil resistance became the leading explanation to investigate. A timing change alone would not establish the cause of the plateau.
NEXT EXPERIMENT Is the last false trigger conducted or radiated?
QUESTION
The enclosure removed two of three false-trigger modes. If the survivor were radiated, shielding should have caught it too — so is it riding in on the shared sensor supply?
DATASET
Pending. The v04 isolated supply went in on 2026-07-26; the 5 V rail has not been scoped under discharge yet.
AI ANALYSIS
Pending. Prior evidence in favour: both dead v2 controllers failed on the sensor side, not the power side.
DECISION
Pending.
07

Current Understanding

For four versions this was filed as a timing problem. The resistance measurement and estimated damping ratio instead point toward a coil-response problem. The working explanation is that the magnetic field persists at an unhelpful point in the transit. That timing has not been established by the damping estimate alone.

The stored-energy revision produced little improvement in exit velocity. More energy did not establish more useful motion. The loss appeared to be downstream of the bank; its precise mechanism remained a question.

The measurement chain has to be commissioned before it is trusted. Eleven runs were spent firing into an instrument that could not, in principle, produce a reading. That was not a sensor fault — it was never testing the sensor against something of known speed.

What is still unknown: whether the surviving false trigger is conducted through the shared sensor supply or radiated into the gate line. The v04 isolated supply is the experiment that addresses it. At the date of the v04 entry, its effect had not been measured.

08

Next Iteration

← ALL PROJECTS
PROJECT #002 MOTHBALLED V06 LAST

Residential Ordinance Platform

What began as a motorized pointing platform evolved through seven versions, four payloads, and increasingly gentle ammunition. The camera turned out to be a terrible independent witness, because it was attached to the thing it was supposed to judge. The useful controller defected to the coil-gun project. Its former host now enjoys retirement on a shelf.

ORIGINAL QUESTION
Can the platform point where the camera thinks it is looking?
UNNECESSARY ESCALATION
Seven versions, four payloads, and a control system with a self-referential witness.
RECORDED OUTCOME
Measured backlash. A field brownout. A reusable controller. Independent pointing accuracy remains unknown.
01

The Question

Can a dual-axis pan-tilt platform put a projectile where the camera says it is looking — repeatably — when every part of the aiming chain is 3D-printed, and the acceptable failure mode is defined not by a spec but by where the neighbour's fence is?
02

Evolution Timeline

v00 Valve first — everything else later 2025-04-25 → 2025-05-05
HYPOTHESIS
The first pneumatic experiment asked whether the valve was the limiting part. The more urgent question turned out to be how unexpectedly the stored pressure could be released.
WHAT ACTUALLY HAPPENED
It can — the valve was never the limiting part. The limiting part was that a single switch between 150 psi and open air is not a design. It is an accident with a schedule.
UNEXPECTED DISCOVERY
The photograph has handwritten Safety and Trigger annotations. They record an evolving idea about control; they are not a complete interlock design or a safety assessment.
WHAT CHANGED
The release controls and mechanical support were revised after unintended activations. A revision is not proof that all hazardous failure modes have been addressed.
Overhead photo of a 24 volt solenoid valve wired into brass quick-connect air fittings, hand-annotated with the words Safety, Trigger and 150psi pointing at the relay, the switch line and the red regulator fitting.
Black pressure chamber and steel barrel assembled on a low stand on a workbench, next to a bench power supply.
v01 Pick the projectile before you pick the barrel 2025-05-13
HYPOTHESIS
Bore diameter is downstream of the projectile, so choose the projectile and let the rest follow.
UNEXPECTED DISCOVERY
A foam toy rocket has an appalling ballistic coefficient and excellent neighbourly relations. Only one of those two properties appears in a textbook, and on this project the other one is the harder constraint.
LESSONS LEARNED
I cut two candidate barrels before choosing the ammunition. One of them is still on the shelf, correct for a projectile that was never adopted.
WHAT CHANGED
Bore sized to the rocket rather than to available tube stock. Cradle printed to hold it at a known angle.
Workbench with a foam-tipped toy rocket beside two candidate barrels, a red printed cradle, and the solenoid valve assembly plumbed to a polished air chamber.
v02 Dual-axis, and an opinion about where to look 2025-05-20 → 2025-05-26
HYPOTHESIS
A pan-tilt stage — one stepper per axis — plus a webcam turns a launcher into a platform.
WHAT ACTUALLY HAPPENED
The pan-tilt moved. It moved past the commanded position and settled somewhere near it. Printed gear teeth and a long cantilevered barrel produce backlash, and the controller experiences backlash as an encoder that lies to it in one direction only.
UNEXPECTED DISCOVERY
The camera is bolted to the tilt axis, so pointing error is invisible to the camera — the instrument moves with the fault it is supposed to measure. The sensor and the error share a mounting bracket.
WHAT CHANGED
Bench integration: pneumatic column, stepper driver stack and gimbal onto one frame instead of three piles.
Red 3D-printed pan and tilt yoke on a large printed turntable, carrying a webcam and a steel barrel, mounted on a cork board with an orange coiled air hose behind it.
Bench integration: the red turret beside a vertical pneumatic column with brass manifold and blue tubing, wired to a heatsinked stepper driver stack.
v03 Off the bench, into the residential 2025-05-31
HYPOTHESIS
If it works on the bench it will work on a patio table.
WHAT ACTUALLY HAPPENED
Mostly. Everything that was a cable on the bench became a trip hazard on a lawn, and the first genuinely field-caused fault was a brownout that bench power had been quietly papering over for a month.
UNEXPECTED DISCOVERY
Untethering it — power-tool battery through an inverter, a portable air tank, the controller riding along on the same board — did not just change the packaging. It changed which failures are reachable. A bench supply with amps to spare hides every current-related fault you have.
WHAT CHANGED
Portable air tank, battery and inverter, and everything mounted on a single board that can be carried outside with two hands and no toolbox.
The turret rig set up on a mesh patio table in a fenced back garden, plumbed with blue tubing and a red air hose, aimed across the lawn.
The same field setup annotated in text: Turret Platform, RPi plus Stepper Drivers, Firing Capacitor Tank, EGO battery with inverter, and a portable air tank with pressure gauge.
v04 Ammunition, reconsidered 2025-06-01 → 2025-06-14
HYPOTHESIS
A commercial gel blaster is a better payload than a bespoke pneumatic launcher.
WHAT ACTUALLY HAPPENED
It is, comprehensively, and admitting that took a fortnight. It feeds from a drum, it fires repeatably, and it had already solved the sealing problem I was three weekends from solving badly. Two months of pneumatics went onto a shelf.
UNEXPECTED DISCOVERY
Swapping the payload for an off-the-shelf one moved the entire project's difficulty into the gimbal — which is where it had belonged since v02. Changing the payload did not resolve the pointing error or make an uncontrolled projectile safe.
WHAT CHANGED
Gel blaster mounted to the tilt yoke. Pan and tilt each get their own labelled driver, because chasing a wiring fault across two identical unlabelled cables happened exactly once.
An orange and white camouflaged gel blaster on cardboard beside its black drum magazine and a tub of blue hydrated gel beads.
The red turret yoke carrying a graffiti-patterned gel blaster with drum magazine and a clear bead feeder, wired to a blue terminal breakout board.
Field configuration outdoors: the turret with gel blaster beside an aluminium extrusion frame whose cables are labelled GIMBAL PAN and GIMBAL TILT.
v05 The JK-47 2025-08-17 → 2025-09-21
HYPOTHESIS
The platform should be agnostic to payload. Prove it by fitting the least serious payload available.
UNEXPECTED DISCOVERY
Every version has made the projectile softer and the pointing more accurate. Foam rocket, then gel beads, then pool-noodle segments from colour-coded magazines. That ordering was not planned and it is exactly the right ordering.
LESSONS LEARNED
An eleven-dollar commercial air-tool component outperformed the custom-specified part. Considerable design effort had been spent rediscovering an existing product.
WHAT CHANGED
The payload interface was revised again. The useful question remained pointing error, which none of the payload changes had answered.
Pneumatic air tools hanging on a red pegboard above the project's aluminium extrusion frame and labelled gimbal wiring.
Brightly coloured 3D-printed magazine frames stamped JK-47, each loaded with foam pool-noodle segments, flanking a printed launcher with a blue pistol grip.
v06 The controller outlived the payload 2026-07-26
HYPOTHESIS
Pan-tilt control is general. The same controller should drive this platform and the stage sequencer on the coil gun next door.
UNEXPECTED DISCOVERY
It does, and that is the actual result of this project. The payload has changed four times; the controller has changed once. The thing treated as scaffolding turned out to be the deliverable, and the launchers turned out to be interchangeable accessories.
WHAT CHANGED
Pan-tilt controller v2 frozen after three builds. Pan and tilt drivers split onto separate boards so a failure takes out one axis instead of the machine. A v2 now runs stage timing next door on the coil gun.
RGB-lit shelf of Raspberry Pi stepper controller assemblies standing on printed pedestals labelled RPi Stepper PT Controller v1 and three labelled v2, with one board marked PAN DRIVER.
03

Prototype Passport

// PROTOTYPE PASSPORT ORDINANCE · v06
Version
v06
Date
2026-07-26
Hypothesis
Pan/tilt control is general enough to be shared across projects; the payload is an accessory, not the platform.
Major Change
RPi Stepper PT Controller v2 frozen. Pan and tilt drivers split onto separate boards.
Unexpected Discovery
The controller is the product. Four payloads have come and gone; the controller has revised once and is now running a second project.
Next Experiment
Unanswered: how much pointing error remains when measured against an independent reference, and how much is attributable to backlash?
iterventions.com · project #002 · not finished, on purpose
04

Failure Gallery

One Switch Between 150 PSI and Open Air
v00 · 2025-04-25
EXPECTED
A test fixture. Press the button, the valve opens, air goes somewhere.
ACTUALLY HAPPENED
Exactly that — including twice when the button was bumped rather than pressed. A charged chamber and a single momentary switch means the only interlock in the system is attention, and attention is not a component you can buy a datasheet for.
Lesson: the chamber released stored pressure twice when the button was bumped. An attention-dependent control was inadequate. The later changes do not amount to a complete safety assessment.
Repeat it? No. Nothing gets charged without an arming step now.
The Camera Rides the Error
v02 · 2025-05-20
EXPECTED
A camera on the tilt axis gives the controller ground truth about where the barrel is pointing.
ACTUALLY HAPPENED
It gives the controller ground truth about where the camera is pointing, which is the same lie the steppers are already telling. Backlash moves the barrel and the camera together, so the loop closes perfectly around an error it cannot see.
The red pan-tilt yoke with the webcam bolted to the same tilt axis as the barrel, so both move together.
Lesson: this camera could not independently verify absolute pointing while sharing the motion of the error. It was telling the truth about its own view. I had asked it a different question.
Repeat it? Still mounted that way. The fix is a fixed reference and it isn't built.
Bench Power Was Hiding a Brownout
v03 · 2025-05-31
EXPECTED
Moving outside changes the scenery, not the electronics.
ACTUALLY HAPPENED
On battery and inverter, a simultaneous pan and tilt move dropped the rail far enough to reset the controller mid-motion. The bench supply had been absorbing that inrush without comment since v02, so the fault had existed for a month with no symptom.
Lesson: a generous bench supply is a debugging tool that removes evidence. Test on the power you will actually ship on, as early as it is embarrassing to do so.
Repeat it? No. Field power first now, bench supply only for diagnosis.
05

Data Notebook

XY · PAN · COMMANDED vs MEASURED1.8° backlash
ILLUSTRATIVE RECONSTRUCTION / NOT A RAW CAPTURE
Illustration of directional backlash, based on the notebook's reported 1.8° finding. The two generated traces explain the gap; they are not the original angle samples.

Wheels on the bench. Printed parts. The assembled JK-47. Then the shelf. The debris is in here too. Select a photograph to inspect the full frame.

Annotated field rig showing the turret platform, the RPi and stepper drivers, the firing tank, the battery with inverter, and the portable air tank.
ordinance_v03_20250531_img_02.jpg
The whole untethered system, annotated in the field. Every label here is a subsystem that failed differently once mains power was gone.
Workshop bench with a red pan-tilt assembly, while a white air tank, pressure gauge, tubing, and another wired mechanism occupy the floor in front.
ordinance_v04_20250720_img_01.jpg
Bench hardware and the pneumatic apparatus share the frame.
Black rubber wheels with blue hubs, brass couplings, and a small display-and-knob control module laid out on a bench.
ordinance_v05_20250816_img_01.jpg
Wheels, couplings, and a control module. The loose parts precede the assembled JK-47 photographs.
White printed half-shell lying open on a bench, with rough layered surfaces around a central form.
ordinance_v05_20250817_img_02.jpg
White printed parts. The rough internal surfaces stay in the record.
Two black wheels with blue hubs seated in a white printed frame on a metal patio table.
ordinance_v05_20250824_img_01.jpg
A white frame and paired wheels. At this point the photograph is more wheel than launcher.
Yellow cylindrical rollers with blue hubs mounted in a black printed frame on the bench.
ordinance_v05_20250830_img_01.jpg
Yellow rollers in a black frame.
Bright green printed assembly laid on a parts case, with a long threaded form and matching printed housings.
ordinance_v05_20250831_img_01.jpg
The printed parts acquired their own color scheme. Assembly photographs follow.
Side view of the assembled JK-47, with a green marked magazine, pink and blue printed structure, yellow rollers, and exposed wiring.
ordinance_v05_20250913_img_01.jpg
The assembled JK-47, with a green magazine and exposed wiring.
Container full of multicolored discarded printing material, including stringy filament and irregular printed fragments.
ordinance_v05_20250914_img_01.jpg
The color scheme also appears in the scrap.
Side view of the JK-47 with a blue printed frame and grip, yellow rollers, a pink upper section, exposed wires, and a detached blue printed part on the bench.
ordinance_v05_20250917_img_01.jpg
Blue structure, yellow rollers, and a printed grip. The wiring is still out in the open.
Colour-coded printed magazines stamped JK-47 loaded with foam pool-noodle rounds beside the printed launcher.
ordinance_v05_20250921_img_01.jpg
The JK-47 and its magazines. Payload number four, and the softest thing this platform has ever pointed at anything.
JK-47 and magazine on a lower wall shelf, beneath a shelf holding the red pan-tilt platform and a separate wired controller assembly.
ordinance_v05_20250927_img_01.jpg
The JK-47 below the pan-tilt hardware.
The controller shelf lit in colour: one v1 and three v2 Raspberry Pi stepper controllers on labelled pedestals.
ordinance_v06_20260726_img_01.jpg
One v1, three v2s. The same board revision is now sequencing coil stages on PROJECT #001.
// BENCH NOTEBOOK · p.072025-05-26
What 1.8° of backlash actually costs
backlash θb = 1.8° (measured, pan)
miss @ 3 m = 3000 · tan(1.8°) ≈ 94 mm
miss @ 8 m = 8000 · tan(1.8°) ≈ 251 mm
stepper step = 1.8° / 16 µsteps = 0.1125°
→ backlash = 16 microsteps of pure fiction
MARGIN NOTE

The resolution is sixteen times finer than the error. On a direction reversal, sixteen commanded microsteps can disappear into gear play before the output moves. Raising the gain does not remove that deadband. The resulting oscillation looked like a tuning problem for three weeks.

→ Open question: what error would an independent reference reveal?

06

AI Lab Notebook

EXPERIMENT 01 Is the hunting a tuning problem or a mechanical one?
QUESTION
The gimbal oscillates around the target on a moving track but holds lock on a stationary one. Is the loop badly tuned, or is it compensating for something mechanical?
DATASET
Commanded versus measured pan angle over sweeps in both directions, plus the oscillation frequency at three different proportional gains.
AI ANALYSIS
The out-sweep and back-sweep do not overlay; they are offset by a near-constant angle regardless of speed, which is the signature of backlash rather than lag. Oscillation frequency tracks gain in the way a limit cycle around a deadband does, not the way an underdamped second-order system does. No gain setting removes a deadband — it only changes how fast you cross it.
DECISION
Backlash became the mechanical explanation to investigate. Further tuning would not establish whether the measured deadband had changed.
EXPERIMENT 02 Which subsystem browned out first in the field
QUESTION
On battery power the controller resets during simultaneous pan and tilt moves. Is it inverter capacity, inrush, or a wiring drop?
DATASET
Rail voltage during single-axis and dual-axis moves, on bench supply and on battery-plus-inverter, with the reset events timestamped against the motion profile.
AI ANALYSIS
Resets cluster on simultaneous acceleration, not on sustained motion, and the same profile is clean on the bench supply. That points at inrush against source impedance rather than average power — the inverter is adequately sized and inadequately stiff.
DECISION
The source's transient response became the leading suspect. A larger inverter's average-power rating would not by itself explain or resolve the reset.
NEXT EXPERIMENT Pointing accuracy against a reference the turret cannot move
QUESTION
With the camera bolted to the tilt axis, every pointing measurement so far has been self-referential. What does the error look like against a fixed external reference?
DATASET
Pending. Needs a second camera off the platform, or a target board with known geometry at a measured distance.
AI ANALYSIS
Pending. Prediction on record: the true error is larger than the on-board camera has ever reported, because the on-board camera structurally cannot report it.
DECISION
Pending.
07

Current Understanding

This was never a controls project. It is a gear-train project that has been wearing a controls project's clothes since v02. Backlash of 1.8° on the pan axis against a microstep of 0.1125° means the pan-tilt's position resolution is sixteen times finer than its position error, and no amount of loop tuning closes a deadband.

Mounting the camera to the tilt axis was the single most expensive decision in the project. It makes the pointing error structurally invisible to the only instrument on board, which is why the error survived four versions without ever being measured properly.

The payload does not matter and never did. Pneumatic launcher, gel blaster, foam noodles — each swap took a weekend and none of them changed the accuracy. The platform is the project; the ordnance is a fixture.

The genuinely useful output of this project is the controller. It has revised once in fifteen months, it now drives a completely different machine on PROJECT #001, and it was built as scaffolding. That is why this platform is mothballed rather than finished: it answered its question, the answer was "you were building the wrong thing," and the right thing walked off to another bench. What is still unknown is how big the pointing error actually is, and that cannot be answered until there is a reference the turret is not bolted to. The rig is on a shelf, intact, waiting for someone to build that reference.

08

Next Iteration

← ALL PROJECTS
PROJECT #003PHOTOGRAPHIC RECORDJAN – FEB 2025

Evil Lab Secret Door

There are two ways to make a secret door. The first is to conceal it. The second is to install an actuator, a control cabinet, a wall display, theatrical blue lighting, and a sign reading EVIL LABORATORY. Guess which approach won the design review.

THE PROPRIETOR'S DESCRIPTION
The overengineered “Evil Lab” secret door.
VISIBLE ESCALATION
Bench wiring became a populated enclosure beside a door-mounted actuator.
RECORDED OUTCOME
Ten photographs. The installation is visible; operating reliability and current project status remain unknown.
Open white doorway with blue lighting around the frame, the door swung inward, and part of the actuator visible near the floor.
labdoor_v00_20250222_img_03.jpg
Blue lighting around the open inner doorway. The actuator is visible near the floor.
Blue-lit wall beside the door with a display, a small digital module, a black round device, and an open wired control enclosure.
labdoor_v00_20250222_img_04.jpg
Display, controls, and the open cabinet under blue light.
01

The Question

The original design brief is not recorded here. The visible hardware leaves a practical question open:

Can the door remain manually operable when its powered controls are unavailable?
02

Evolution Timeline

Dates follow the photographs. Hardware revision numbers, original hypotheses, and the reasons for each change are not recorded.

01The enclosure fills up.2025-01-10 → 2025-01-18
HYPOTHESIS
Not recorded.
WHAT CHANGED
The January 10 view shows a mounted terminal strip. By January 18, the enclosure contains labelled wiring, a board, and a module in the lid. The amount of hardware is visible. Its operating history is not.
Open grey enclosure on a tool case, with an orange terminal strip mounted inside and loose wiring beside it.
02The electronics move to the wall.2025-01-19 → 2025-01-26
HYPOTHESIS
Not recorded.
WHAT CHANGED
An actuator appears low on the white door. The later January views show an open cabinet on the adjacent wall and a display beside it. The mounting arrangement is documented; automatic operation is not established by these still images.
Open wall-mounted control cabinet beside the doorway, exposing the circuit board, cable labels, terminals, and a black module inside the lid.
03The room beyond the paperwork.2025-02-22
HYPOTHESIS
Not recorded.
WHAT CHANGED
The February photographs include the stencilled wooden door, the illuminated inner doorway in an open position, and a view toward a 3D printer. They add context. They do not establish another hardware revision or a successful test sequence.
View through the open wooden door toward a blue-lit enclosed 3D printer at the end of a narrow passage.
03

Prototype Passport

// PROTOTYPE PASSPORTLAB DOOR · PHOTO RECORD
Version
Hardware revision not recorded.
Date
Photographs: 2025-01-10 → 2025-02-22.
Hypothesis
Original hypothesis not recorded. Manual operation with powered controls unavailable remains an open question.
Major Change
The dated views progress from a bench enclosure to an installed cabinet, door-mounted actuator, and adjacent display.
Unexpected Discovery
Not recorded.
Next Experiment
Not recorded. Unanswered: what happens to manual opening when power or control is unavailable?
iterventions.com · project #003 · photographs do not establish safe operation
04

Failure Gallery

RECORD INCOMPLETE

No failure event is documented here. Reliability remains unknown. That is different from a clean test history.

05

Data Notebook

The enclosure, the installation, and the room it was all apparently necessary to enter. Full frames, in capture-date order. Select a photograph to inspect it.

Open grey enclosure on a tool case, with an orange terminal strip mounted inside and loose wiring beside it.
labdoor_v00_20250110_img_01.jpg
The enclosure on the bench. Even the door's electronics needed somewhere to live.
Populated control enclosure with labelled cables, terminal blocks, a circuit board, and a black module attached inside the lid.
labdoor_v00_20250118_img_01.jpg
Eight days later, the enclosure contains a board, terminals, and labelled cables. The labels include Main Power and Retract.
Compact metal-bodied actuator mounted low on the back of a white door beside a blue-lit frame.
labdoor_v00_20250119_img_01.jpg
A compact actuator on the door. The mounting is visible beside the blue-lit frame.
Wooden door marked EVIL LABORATORY beside a white doorway edged with blue light, with black electronics and cables attached to the wooden door.
labdoor_v00_20250119_img_02.jpg
The sign was explicit. The word secret was doing less work than the electronics.
Open wall-mounted control cabinet beside the doorway, exposing the circuit board, cable labels, terminals, and a black module inside the lid.
labdoor_v00_20250126_img_01.jpg
The cabinet is now on the wall. The labelled wiring remains visible when the lid is open.
Back of the white door and surrounding wall, showing the low-mounted actuator, an open control cabinet, and a display beside the doorway.
labdoor_v00_20250126_img_02.jpg
The door, cabinet, and display in one frame. Opening a room had acquired supporting infrastructure.
Wooden door bearing the EVIL LABORATORY stencil, several black electronic modules, hanging cables, and conventional door hardware.
labdoor_v00_20250222_img_01.jpg
A door with a name, door hardware, and several additional reasons to keep track of cables.
View through the open wooden door toward a blue-lit enclosed 3D printer at the end of a narrow passage.
labdoor_v00_20250222_img_02.jpg
Beyond the door: a blue-lit printer.
Open white doorway with blue lighting around the frame, the door swung inward, and part of the actuator visible near the floor.
labdoor_v00_20250222_img_03.jpg
Blue lighting around the open inner doorway. The actuator is visible near the floor.
Blue-lit wall beside the door with a display, a small digital module, a black round device, and an open wired control enclosure.
labdoor_v00_20250222_img_04.jpg
Display, controls, and the open cabinet under blue light.
06

AI Lab Notebook

VISUAL REVIEWThe photograph is not the test result.
QUESTION
What do these photographs actually establish?
DATASET
Ten selected photographs dated January 10 through February 22, 2025. This review uses still images.
AI ANALYSIS
The enclosure, actuator, display, labels, and door positions are visible. Still images do not establish the trigger logic, operating reliability, failure behavior, or safe exit.
DECISION
Keep the visible construction and the unresolved questions together. Do not turn an open-door photograph into a successful-cycle claim.
07

Current Understanding

The secret door has an unusually public support system: an openable control cabinet, cable labels, a display, and lighting around the frame. The sign says EVIL LABORATORY. The wiring has a more detailed vocabulary.

The photographs document construction and installation. They do not provide cycle counts, opening-time measurements, or observations during power loss. Current project status is not recorded.

The remaining engineering question is manual access: what does the door do when the powered controls are unavailable?

08

Next Iteration

// contemporaneous evidence

Bench Log

Sixteen dated entries. The prototypes changed. The failures accumulated. This is the record that prevents the next confident theory from quietly deleting the previous one.

v04
GAUSS →

Gate sensors get a supply of their own

A separate sensor supply, built and labelled on the date of this entry. Its effect had not yet been verified. Both dead v2 controllers failed on the sensor side rather than the power side, which is the whole argument for suspecting a conducted path rather than a radiated one.

v06
ORDINANCE →

Mothballed, and the controller emigrated

Four payloads, one controller revision, fifteen months. The v2 board is now sequencing coil stages on PROJECT #001, which means the scaffolding outlived every single thing it was holding up. Platform goes on the shelf intact.

v03
GAUSS →

The coil is its own brake

Stage 1 measures 4.3 Ω against a characteristic impedance of about 1.4 Ω. The model gives a damping ratio near 1.52, consistent with an overdamped response. Four versions of timing assumptions now had a resistance measurement to answer to. The actual current-to-projectile timing remained unverified.

v03
GAUSS →

It stopped being a pile

Everything into one enclosure, one connector standard, XT60 throughout. Two of the three intermittent faults disappeared on the way in and I cannot tell you which connection fixed them, which is the price of changing six things at once.

v02
GAUSS →

Runs stop being anecdotes

The scope and custom log viewer became permanent bench equipment. Substantially more stored energy had produced little improvement in exit velocity. The bank had become more hazardous without becoming more persuasive.

v02
GAUSS →

Eleven runs, no velocity

The sequencer armed, fired and logged eleven complete runs. Gate 1 transit velocity read two dashes for every one of them. The gates were closer together than the sensor's own rise time, so the measurement was impossible before the first shot.

v01
GAUSS →

Firing current, meet breadboard

The kickback clamp works. It works on a solderless breadboard six inches from cables I had personally labelled FIRING CURRENT with three exclamation marks. Both facts are in the same photograph and it took a week to notice they were arguing.

v00
GAUSS →

The keycap says RAIL GUN

It is not a rail gun. No rails, no armature bridging them — it is a coil and a tube, which is a coil gun operating on reluctance rather than Lorentz force. A month of wrong intuitions came from reasoning about the wrong machine. About 6 J stored, and the slug moved roughly as hard as a strong exhale.

v05
ORDINANCE →

The JK-47

Pool-noodle segments from colour-coded printed magazines, quick-change interface, stamped JK-47. Payload number four and the softest thing this platform has ever pointed at anything. Every version has made the projectile softer and the pointing better, in that order, entirely by accident.

v05
ORDINANCE →

An eleven dollar valve beat the designed one

The trigger valve off a pneumatic nailer on the shop wall is faster than the solenoid I specified, sourced, wired and mounted. Some parts of this hobby are solved and the skill is noticing which ones.

v04
ORDINANCE →

Field day with the new payload

Gel blaster on the yoke, pan and tilt on separately labelled drivers, everything on one carryable board. The labels exist because chasing a wiring fault across two identical unlabelled cables happened exactly once and lasted an afternoon.

v04
ORDINANCE →

Buying the thing I was about to build badly

A commercial gel blaster feeds from a drum, fires repeatably, and had already solved the sealing problem I was three weekends from solving worse. Two months of pneumatics went on a shelf. A softer payload did not settle the accuracy or containment questions.

v03
ORDINANCE →

Bench power was hiding a brownout

First run on battery and inverter. A simultaneous pan and tilt move dropped the rail far enough to reset the controller mid-motion — a fault that had existed for a month with no symptom, because the bench supply had amps to spare and absorbed every inrush without comment.

v02
ORDINANCE →

1.8 degrees of nothing

Measured backlash on the pan axis. The microstep is 0.1125°, so a direction reversal can consume sixteen commanded microsteps before the output takes up the gear play. The controller had more resolution than the mechanism had cooperation.

v01
ORDINANCE →

Pick the ammunition first

Two barrels cut before choosing a projectile. One of them is still on the shelf, perfectly correct for a foam rocket that was never adopted. Bore diameter is downstream of the payload and I did it in the other order.

v00
ORDINANCE

One switch between 150 psi and open air

The valve was never the limiting part. The limiting part was that a charged chamber behind a single momentary switch means the only interlock in the system is attention. It fired twice on a bump rather than a press before that sank in.

↻ end of log · next entry when something breaks