ApsidionAPSIDION
THE GAPTHE MISSING HALFWORKPROGRAMMES →
PROGRAMME 01DECOUPLING — PROVEN

Tether

Ophthalmic surgery, rehearsed against a computed eye.

It is the smallest decisions and movements that determine whether a sight is restored or lost. The principles can be taught; the rest is practice — and practice has always been the scarce half. Tether makes the rehearsal computable, starting with cataract.

Capsulorhexis mid-propagation against the red reflex, with the intended 5.2 mm rhexis circle and the tear-front vector annotated.

Three crafts moved indoors. Surgery did not.

Each of these was once physical, expensive and gatekept, and each became something a practitioner could iterate on at will. The exception is the one where the cost of iterating is borne by a patient.

01
Engineering design
CAD. Iterated and checked long before anything was cut.
02
Decision-making
The spreadsheet. Static ledgers became models anyone could challenge.
03
Creative craft
Photoshop. The darkroom became forgiving, and then free.
04
Surgery
No equivalent. Intuition is still earned on people, under borrowed supervision.
Every patient deserves a surgeon at their best. Every surgeon deserves the right to get there — to practise, to fail, to iterate — without a single patient bearing the cost of that journey.

Physics and appearance, decoupled by construction.

The architecture is the argument. A generative model asked to produce both a convincing image and a trustworthy force will give you neither, so it is never asked.

STATE — SOLVED
The mechanics
Instrument reaction forces, tissue stress and deformation are computed deterministically. Every physical quantity the system reports originates here.
APPEARANCE — DRESSED
The colourist
A LoRA-adapted video model receives geometry and returns appearance. It never calculates a metric, so it cannot hallucinate one.
THE BRIDGE — TWO ALIGNED CONTROL STREAMS PER CLIP
SegmentationA flat, hard-edged colour map over six ontology classes — cornea, pupil, lens, instrument and the rest — carrying what each region is.
EdgeSato ridge detection, carrying the creases and the capsulorhexis tear — where the structure breaks.
InjectionBoth streams enter through the renderer’s unused inactive latent channels, so structural boundaries are supplied without touching the model architecture.
AlignmentExactly registered to the solved geometry, frame for frame. Misalignment is the only path by which appearance could contaminate state, which is why it is the registration that is checked and never the picture.

What the rehearsal does.

Five capabilities, each of which exists because the mechanics are computed rather than scripted. None of them is available to a system that learned surgery from video.

01
Patient-specific eye
The anatomy under the instrument is built from that patient’s own imaging, not from an average.
02
Tissue mechanics
Cutting, tearing and force feedback come from a constitutive model — hyperelastic with a viscoelastic relaxation series — not from animation curves.
03
Adaptive difficulty
Complexity escalates against the surgeon’s current skill rather than against a fixed script.
04
Rare-case rehearsal
Anomalies and complications are generated on demand, so the tail is practised rather than waited for.
05
Shared theatre
A whole team rehearses one case together, from wherever each of them is.

What is instrumented.

The quantities the system records on every case. They are what a residency currently assesses by eye, and they are the raw material of the transitions this programme contributes to the corpus.

CCC circularityCapsulorhexis roundness against the intended circle
Tear-front vectorRadial deviation of the tear as it propagates
Tool forcesAxial and transverse load at the instrument tip, in millinewtons
Phaco energyPower fraction and dwell above threshold
Aspiration flowVolumetric rate at the stated vacuum
Peel completenessMembrane area removed, and traction while removing it
Chamber stateAnterior chamber depth and intraocular pressure through the case

The limits, first.

Stated at the same size as the capability. A system that names its boundary can be trusted inside it; one that does not cannot be trusted anywhere. These are the boundaries of the claim, not the state of the build.

Rehearsal changes where the early attempts happen

It does not replace supervised operating experience, and nothing here should be read as claiming it does. The same position this company takes on physical trials applies to surgical ones: fewer of them, earlier, at lower cost — not none.

Appearance is never evidence

The learned renderer is never asked for a physical quantity. Every number the system reports comes from the solved mechanics, not from what the image looks like. Where the two disagree, the mechanics are what is recorded.

Outside the anatomy it is built from, it states the boundary

The model carries the range of biometry it was constructed against. Beyond that range it reports that it is outside its envelope rather than returning a confident number — the same rule every other Apsidion output follows.

It rehearses the case. It does not assist during one

Tether is built for practice before theatre. It is not an intraoperative aid and makes no claim to be one.

Why this programme is the existence proof.

The renderer cannot hallucinate a force, because it was never asked for one.

Tether is where the decoupling thesis stops being an argument and becomes a working system. Physical state is produced by physics and is therefore deterministic and measurable; appearance is produced by a network and is therefore never trusted with a number. Every claim this company makes about world models rests on that separation holding in practice, and here it does.

NEXT PROGRAMME
Praxis
PRAXIS →