Measured live from node positions alone — the same blind reconstruction the study below uses.
There is no power bar. The rod is a loaded beam and the line is an inextensible filament; the loop you see is a wave travelling down the line, not an animation of one. Everything you do is applied to the rod as a torque at your hand, and the line does what the physics says.
| Space or click, or touch, and hold the water | drive the rod. Let go to stop it. The stop is what makes the loop: a short crisp stroke throws a tight loop, a long sweeping one throws a wide one. | |
| ← → or A D | aim left and right (between strokes). On a touch screen, hold the ◀ ▶ buttons. | |
| ↑ ↓ or W S | shoot or strip line. Legal only while the line is in the air, as rule VII of the event requires. On a touch screen, hold the shoot and strip buttons. | |
| R | new round | |
| H | this panel |
Event 1, Trout Accuracy. Four targets, each three rings of 60, 120 and 180 cm, placed randomly between 8 and 15 m in a fan of about 90 degrees. Sixteen casts, left to right, 1 2 3 4 four times. Centre 5 points, middle 3, outer 1; maximum 80. Five minutes. If the fly, the leader or the line strikes the water or the inside of a ring while you are false casting, that cast scores nothing — but touching the water behind the stand is expressly allowed.
Event 2, Trout Distance. A sector 8 m wide (the rules allow 5–10 m) and at least 50 m long. Longest cast wins, measured to the passed half metre. A fly outside the sector scores nothing.
This is an independent reimplementation of a published competitive format. It is not affiliated with, endorsed by or derived from the International Casting Sport Federation, the American Fly Fishing Trade Association, or any tackle manufacturer. See Method & sources for everything that is documented, everything that is measured, and everything that is a guess.
The rod and the line are one inextensible chain of point masses: twelve to sixteen segments of tapered beam for the rod, forty-eight to ninety-six for the fly line, its leader and the fly. Segment lengths are held exactly by constraint projection on positions and on velocities, so the line never stretches and the constraints put no stiffness into the timestep. Bending is a discrete elastic potential; air drag acts per node, split into a normal and a tangential component, and is integrated in closed form so a 40 mg fly does not set the step size for the whole cast. The caster is a fixed pivot at the hand with a single applied torque, capped at a human peak. That torque is the only thing the player controls.
The casting loop is not animated. Nothing in the code draws a loop or moves one along. The fold appears because the rod stops and the line carries on, and it travels because a flexible filament under tension carries transverse waves.
It does not. At the drag a real fly line actually sees, the loop slows by more than half over the second half of its run, and the taper makes that worse, not better.
Measured from the engine at the reference settings (WF5F, 11 m of fly line plus a 2.70 m leader, one delivery from a completed backcast, ninety-six line nodes, sixteen rod nodes):
| Fold travel from the rod tip | 4 m | 6 m | 8 m | 10 m | 12 m | 13 m |
|---|---|---|---|---|---|---|
| Loop speed, tapered WF5F (m/s) | 13.08 | 12.59 | 11.25 | 6.05 | 2.18 | 1.43 |
| Loop speed, level line of the same mass (m/s) | 12.61 | 11.22 | 9.88 | 7.56 | 3.01 | 2.26 |
| Tension at the fold, tapered (N) | 0.433 | 0.267 | 0.117 | 0.013 | 0.0011 | 0.0014 |
| Linear density at the fold (g/m) | 1.122 | 1.122 | 1.025 | 0.557 | 0.196 | 0.143 |
Speed ratio from 4 m to 10 m: 0.46 tapered, 0.60 level. From 4 m to 13 m: 0.11 tapered, 0.18 level.
The fold is a travelling transverse wave, so it moves at the local wave speed √(T/μ): in the loop's own frame the line runs through the turn at exactly that speed, and that is the condition for the shape to hold together. The blind reconstruction (which sees only emitted node positions, plus the published taper) rebuilds the tension by walking a force balance inward from the fly and finds the measured loop speed equal to √(T/μ) with a median ratio of 0.85 to 0.99 across an eleven-point drag sweep.
Over the cast the tension at the fold collapses by a factor of 318 (0.433 N to 0.0014 N) while the linear density falls by only 7.9. T/μ therefore collapses, and the wave slows. The taper helps μ fall, but it cannot make T fall more slowly — so a tapered line's loop dies sooner than a level line's, which is the opposite of the folk claim.
Two closures that are often assumed both fail. Fitting loop speed against the mass still moving gives an exponent of +0.56 (tapered) and +0.47 (level). Conserving the moving leg's energy predicts −0.5; conserving its momentum predicts −1. The measured exponent has the wrong sign for both. The loop's speed is not set by what the moving mass is doing; it is set by the local tension and the local weight of line.
One kinematic fact does hold exactly, and it matters for reading any claim about "the loop" against any claim about "the fly": with the rod leg at rest, the fly travels at twice the loop speed. A statement about one is a statement about the other, up to a factor of two.
Sweeping the line's normal drag coefficient, everything else fixed:
| CD,normal | 0.1 | 0.2 | 0.3 | 0.4 | 0.5 | 0.6 | 0.8 | 1.0 | 1.2 | 1.6 |
|---|---|---|---|---|---|---|---|---|---|---|
| Loop speed at 10 m ÷ at 4 m | 1.39 | 1.24 | 1.14 | 1.05 | 0.95 | 0.84 | 0.67 | 0.54 | 0.46 | 0.38 |
The loop accelerates only below CD ≈ 0.45. A smooth circular cylinder in cross-flow at the Reynolds numbers a fly line sees here (roughly 300 to 1500) sits near 1.0 to 1.4. So the acceleration is real in the drag-free limit — which is what the published literature reports — and is comprehensively lost by the time the drag is realistic.
It is tempting to answer that by casting harder. That does not work, and the reason is scaling rather than technique: the tension that holds the fold together and the aerodynamic drag that destroys it are both quadratic in the same speed, so their ratio contains no speed at all. Doubling the line speed doubles the drag's effect in step with everything else. Only gravity becomes relatively less important, and gravity is not what is stopping the loop.
These are the ICSF's published world records, read from its records page, and they are what the app compares a round against:
| Event | Men | Women | Set at |
|---|---|---|---|
| Trout Accuracy (of 80) | 72 points (men) | 47 points | World Championships, Ahus, 2024 |
| Trout Distance | 40.5 m (men) | 33.5 m | World Championships, Ahus, 2024 |
| Seatrout Distance | 51.0 m | 40.0 m | World Championships, Ahus, 2024 |
| Salmon Distance | 61.5 m | 52.0 m | World Championships, Ahus, 2024 |
Even the world record for Trout Accuracy is 72 out of 80 — eight points dropped by the best caster in the world. If your first round scores in single figures, that is the event, not the app.
Spolek (1986) reports, in the part of the paper that is public, that without air drag the fly can gain "an order of magnitude" in speed, and that air friction "dramatically decreases the effect" while leaving some acceleration. The first half reproduces here in direction but not in size: at CD = 0.1 this model gains 39%, not a factor of ten. The second half reproduces and then some — at a realistic drag coefficient this model finds no residual acceleration at all, only deceleration. Which of these is right cannot be settled from what this build could read. The full texts of Spolek (1986), Robson (1990), Lingard (1988), Anderson, Perkins & Richards (2006), Wang & Wereley (2003) and Ekander, Perkins & Richards (2024) are all behind paywalls; every one returned HTTP 403. Only the abstracts and the bibliographic records were opened, and nothing numeric was taken from any of them.
The AFFTA specification PDF is itself unreachable at the address the standard is published from — cdn.ymaws.com returns AccessDenied. The table used here was read from an Internet Archive capture of that exact file.
Three oracles, blind to different things, plus deliberate engine mutants. Each oracle is required to pass the unmutated engine before it is allowed to catch anything.
A coarse mesh reports the opposite answer, confidently and smoothly.
| Line nodes | 32 | 48 | 64 | 96 | 128 | 192 |
|---|---|---|---|---|---|---|
| Loop speed at 10 m ÷ at 4 m | 0.93 | 1.09 | 0.79 | 0.46 | 0.56 | 0.57 |
| Energy residual | 0.4% | 0.6% | 1.0% | 1.9% | 3.1% | 5.7% |
| Rod nodes | 6 | 8 | 12 | 16 | 24 |
|---|---|---|---|---|---|
| Loop speed at 10 m ÷ at 4 m | 0.97 | 1.08 | 0.58 | 0.46 | 0.47 |
At 48 or 64 line nodes, or at 6 or 8 rod nodes, this model says the loop holds its speed or speeds up. It settles on the opposite answer only past 96 line nodes and 12 rod nodes, where refining further moves it by under a tenth. Two separate bugs were found by insisting on that: the rod's bending stiffness was being sampled at segment midpoints, which made the tip softer every time the mesh was refined, and the reel mass and the hand's lever arm were pinned to the first rod node, whose position moved with the mesh. Both are fixed; both had been invisible in every other check.
Note that the energy residual gets worse as the mesh is refined at a fixed step tolerance, while the answer gets better. Refining the tolerance instead (0.16 to 0.0025) drives the residual from 12% to 1.1% and moves the loop-speed ratio only between 0.42 and 0.55. The two knobs are not the same knob, and the one that changes the physics is the mesh.
Between casts — and only between casts, never once a stroke has begun — the line is re-laid straight along the aim you are pointing at. A real competitor turns to the next target with a change-of-direction cast and a retrieve. Without this, one collapsed delivery leaves a pile of line in front of the stand that no stroke in this model can lift, and the remaining fifteen casts of the round are simply gone. That is arguably realistic and definitely not a game, so the app grants the reposition. Everything inside a cast — the pick-up, the backcast, the wait, the delivery, and every fault the rulebook defines — is played out in full.
The interactive round runs the same engine at a coarser step tolerance and forty-eight line nodes, so a stroke costs about four milliseconds a frame instead of a second and a half a cast. It is a game, not the measurement. Every number quoted on this page comes from the reference settings above.
The line-by-line source list this build worked from, with a URL for every fact and an explicit mark on everything that could not be opened, ships with the app as facts.md.
DOCUMENTED 23 · qualified 6 · MEASURED 9 · DERIVED 6 · CALIBRATED 2 · RECONSTRUCTED 9 · total 55
A qualified entry is documented, but of something adjacent — a generic catalogue figure rather than this line, a material class rather than this leader, a standard atmosphere rather than this venue. Folding those into DOCUMENTED would flatter the count, so they are separated. The page harness recounts this table from the shipped files and fails the build if it disagrees with the printed tally.
| Tag | Entry | Where it comes from |
|---|---|---|
| DOCUMENTED | Rod for Event 1: any brand or AFTM class, maximum length 277 cm (9 ft = 274.3 cm). | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Line for Events 1 and 2: Scientific Anglers Mastery Expert Distance competition WF5F, minimum length 35.50 m. | ICSF Fly casting rules 01-2026 |
| DOCUMENTED | Leader: monofilament, minimum 2.50 m; tip at least 40 cm long, maximum diameter 0.30 mm, highly visible. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Fly: hookless, floating, hackle 16-20 mm, total length 15-25 mm, wire body at least 13 mm, must float 3 s. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Four targets, each three concentric rings of 60, 120 and 180 cm diameter. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Targets placed about 90 degrees on the stand, two each side of the centreline, randomly between 8 and 15 m. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Stand 1.2 m x 1.2 m, maximum 50 cm above the water; court 20 m in front. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Casting order left to right 1, 2, 3, 4 four times - sixteen casts. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Centre ring 5 points, middle 3, outer 1; maximum score 80. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Time limit five minutes; equal scores are separated by elapsed time. | ICSF Fly casting rules 01-2026, Event 1 |
| DOCUMENTED | Fly, leader or line striking the surface or the inside of a ring during a false cast makes the cast invalid. | ICSF Fly casting rules 01-2026, Event 1 and VI.3.f |
| DOCUMENTED | Hitting the water, or laying the line down, BEHIND the stand is allowed. | ICSF Fly casting rules 01-2026, Event 1 General |
| DOCUMENTED | Trout Distance: parallel sector 5-10 m wide, minimum 50 m long. | ICSF Fly casting rules 01-2026, Event 2 |
| DOCUMENTED | Trout Distance: six minutes, divided into two rounds of three minutes. | ICSF Fly casting rules 01-2026, Event 2 |
| DOCUMENTED | Trout Distance: measured from the stand to the fly, to the passed half metre. | ICSF Fly casting rules 01-2026, Event 2 |
| DOCUMENTED | A fly landing outside the sector in a distance event scores nothing. | ICSF Fly casting rules 01-2026, VI.3.g |
| DOCUMENTED | AFFTA line 5: 134 / 140 / 146 grains (8.70 / 9.10 / 9.50 g). | AFFTA Approved Fly Line Weight Specifications |
| DOCUMENTED | The AFFTA weight is that of the first 30 feet of line minus the level tip. | AFFTA Approved Fly Line Weight Specifications |
| DOCUMENTED | ICSF world records, Trout Accuracy: 72 points (men, Steinar Rostad, 2024) and 47 (women, Anita Strand, 2024). | ICSF world records table |
| DOCUMENTED | ICSF world records, Trout Distance: 40.5 m (men, Bernt Johansson, 2024) and 33.5 m (women, Tone Brathen, 2024). | ICSF world records table |
| DOCUMENTED | Spolek (1986) reports that the fly accelerates as the line rolls out; without air drag by up to an order of magnitude, and that air friction dramatically decreases the effect. | Am. J. Phys. 54, 832 - ABSTRACT ONLY, full text not opened |
| DOCUMENTED | Robson (1990) proposes a computer model of the forward motion of the flyline. | Am. J. Phys. 58, 234 - ABSTRACT ONLY, full text not opened |
| DOCUMENTED | Wang and Wereley (2003) model a fly cast as a nonlinear finite-element rod plus a lumped-parameter line and reproduce loop generation, propagation and turn-over. | ASME IMECE2003-43645 - ABSTRACT ONLY, full text not opened |
| qualified | Fly line tip diameter more than 0.030 inch - a catalogue generality for fly lines, not a figure for this line. | Wikipedia, Fly fishing tackle |
| qualified | Leaders taper from about 0.020 inch to 0.007-0.010 inch - generic leaders, not the ICSF competition leader. | Wikipedia, Fly fishing tackle |
| qualified | A typical fly line is 90 ft (27 m) long - a generality; the ICSF line is at least 35.50 m. | Wikipedia, Fly line |
| qualified | Leader density 1140 kg/m3 - the bulk density of nylon 6,6, not a measurement of this leader. | material handbook value |
| qualified | Air 1.225 kg/m3 and 1.48e-5 m2/s - ISA sea level, not the venue. | ISA standard atmosphere |
| qualified | Gravity 9.80665 m/s2 - the defined standard value, not a local one. | CGPM standard gravity |
| MEASURED | The loop SLOWS as it nears the tip. The fold falls from 13.08 m/s at 4 m of travel to 6.05 m/s at 10 m and 1.43 m/s at 13 m - a factor 0.46 and then 0.11. | this engine, reference settings |
| MEASURED | A level line of the same mass slows LESS, not more: 0.60 over the same span against the tapered line's 0.46. | this engine, mass-matched control |
| MEASURED | The loop's speed equals the local transverse wave speed. Median ratio of measured loop speed to sqrt(T/mu) is 0.85-0.99 over an eleven-point drag sweep. | blind reconstruction from node positions |
| MEASURED | Neither closure fits. Fitting loop speed to moving mass gives an exponent of +0.56 (tapered) and +0.47 (level), where energy conservation predicts -0.5 and momentum conservation -1. | this engine |
| MEASURED | Tension at the fold collapses from 0.433 N to 0.0014 N (310x) while the local linear density falls only 7.9x. | blind reconstruction from node positions |
| MEASURED | The sign of the answer is decided by air drag: the loop accelerates only below a normal drag coefficient of about 0.45. | drag sweep, 0.1 to 1.6 |
| MEASURED | Energy-budget residual 1.9% of the caster's work at the reference step tolerance, 1.1% at the tightest. | engine-side energy oracle |
| MEASURED | Angular-momentum residual about the hand pivot: 0.28% of the applied angular impulse. | engine-side momentum oracle |
| MEASURED | An under-resolved mesh reports the OPPOSITE answer. The loop-speed gain flips from above 1 to below 1 between 64 and 96 line nodes, and between 8 and 12 rod nodes. | mesh refinement |
| DERIVED | For a fold whose rod leg is at rest, the fly travels at exactly twice the loop speed; the moving leg shortens at the loop's own rate. | inextensibility |
| DERIVED | A steady fold must run at the local transverse wave speed sqrt(T/mu), because material flows through the turn at that speed in the loop's own frame. | this model |
| DERIVED | Effective fly-line density 886 kg/m3 - below water, as a floating line must be. It follows from the AFFTA mass and the reconstructed diameters and was not fitted to it. | consistency check |
| DERIVED | Casting harder does not out-run air drag: fold tension and aerodynamic drag are both quadratic in the same speed, so their ratio contains no speed at all. | scaling |
| DERIVED | The two halves of the published AFFTA table do not agree. Converting the grain column gives a different window from the gram column, by up to 0.367 g at line 15 - about 1.1%, wider than several of the tolerances the table sets. | arithmetic on the primary source |
| DERIVED | Wikipedia's copy of the AFFTA table mis-transcribes line 13 high as 30.30 g; the AFFTA page says 30.20 g. | comparison of the two |
| CALIBRATED | Effective fly-line density set so that the first 30 ft minus the level tip weighs exactly the AFFTA target of 140.0 grains. | one free parameter, one documented constraint |
| CALIBRATED | Interactive hand torque set so a well-timed cast lands inside the documented 8-15 m target band. | playability against the rulebook |
| RECONSTRUCTED | WF5F section lengths and diameters: 0.25 m level tip, 3.05 m front taper, 7.60 m belly, 3.05 m rear taper, running line to 35.50 m; 0.76 to 1.27 to 0.84 mm. | no published taper could be opened |
| RECONSTRUCTED | Rod bending stiffness: 22 N m2 at the butt falling to 0.05 N m2 at the tip, blank mass 48 g. | no published rod curve could be opened |
| RECONSTRUCTED | Normal drag coefficient 1.2 for the line. Swept from 0.1 to 1.6; the headline is reported with the sweep attached. | no fly-line drag measurement could be opened |
| RECONSTRUCTED | Tangential (skin-friction) drag coefficient 0.02. | slender-body convention |
| RECONSTRUCTED | Fly mass 40 mg, drag frontal area 18 mm x 20 mm from the documented hackle size. | no mass is published |
| RECONSTRUCTED | Water contact: a penalty surface with damping, tuned to stop the line at the surface rather than to model a splash. | modelling choice |
| RECONSTRUCTED | The caster: a fixed pivot at the hand with one applied torque. No hand translation, no haul, no double haul. | modelling choice, stated because it limits distance |
| RECONSTRUCTED | Individual target angles, drawn inside the documented 90 degree fan; the rulebook fixes the fan, not the angles. | rulebook gap |
| RECONSTRUCTED | Young's moduli: 25 MPa for the line coating, 2.5 GPa for the nylon leader. | material class values, not measurements |