aelife3v10 AN EXPERIMENT IN EVOLUTION
LIMITED SILK. UNLIMITED POSSIBILITIES.

Every thread counts.

Build a web. Scatter flies over it. Keep the winner.
Catch more flies. Spend less silk. Build quicker.

Round · five scatters001Generation 1 champion
Champion’s benchmark0.000 benchmark scoreFixed comparison · fresh fly trials
Selected web’s silk0 / 120
Connectors + catching silk
Champion defences0Web #1 · kept unchanged
Same budget.
Five scatters. Same designs.

All competing webs 8 WEBS

Champion in slot 1 · click any web to inspect
READY TO SCATTER
2 scatters/sSilk stays visible · hide flies to inspect the web
0 / 80 blobs countedAll 8 webs compete together
↳
Beat the champion. Earn its place.Keep the same champion and contenders for five different scatters. Their average decides the winner.

The building recipe

WEB #1
    0 strands · 0 m of thread

    Champion benchmark progress

    WAITING FOR A RESULT
    WinnerParentROUND 1

    Every champion is measured on the same 24 varied fly patterns. Higher scores mean more catches for less silk and build time. Promotion also requires winning the five fresh fly trials. The benchmark only increases while settings stay the same. Changing budget, fly count or distribution starts a new comparison around the same champion.

    Keep a good idea.

    Your champion is saved in this browser. Reload to continue, paused.

    How the web, flies and evolution work

    01 / Evolve the builder.A small construction program chooses a hub and anchor spokes, then adds loops, spirals or cross-links between those spokes. Its parameters and instruction list mutate. Slot 1 keeps the exact champion; every other slot is a mutation of it, including in the first brood. A challenger must beat the champion on five fresh scatters and improve its score on the fixed 24-pattern benchmark. Losing and tied challengers leave its identity, generation and recipe unchanged. This is a bounded recipe language, not a neural network or arbitrary executable code.

    02 / Pay for anchors, silk and time.Each anchor point costs 1 silk unit. Each support line and catching strand additionally costs length × thickness × 10, all deducted from the same budget. An anchor and its line must both fit before either is built. The external square frame is not silk. Construction stops when the next whole step cannot be afforded. The estimated build time is 2 seconds per anchor, 3 seconds per metre of thread, plus 0.15 seconds per catching-strand join. These are simulation costs, not biological measurements. Slowing playback does not change them.

    03 / Catch efficiently and build quickly.Small circular flies, 6–13 mm across on the one-metre field, are scattered over the web. A blob touching or crossing any sticky strand counts as one catch, even if it touches several. Anchor points and support lines cannot catch flies. The score is 10 × mean catches ÷ silk spent ÷ (1 + build seconds / 60). Every anchor and strand uses silk; catches do not alter or damage the web. With equal catches and silk, the quicker web wins. The live trial score averages five fresh scatters. The benchmark averages a separate fixed set of 24 varied patterns, shared by all candidates. A new champion must improve both, so easier or luckier fresh flies cannot erase its earlier standard.

    04 / Keep the fight fair.Choose 2–100 webs and press Apply. Slot 1 is the exact champion; all other slots are mutations of it. Keep this whole line-up for five different scatters. Every contestant gets identical blob positions and sizes within each test. The web geometry is reused unchanged across all five scatters. No promotion happens after tests 1–4. After test 5, benchmark every contender on the same 24 reference patterns. A contender must strictly beat the parent’s fresh-trial score and its benchmark. Choose the highest benchmark among those that qualify; ties retain the parent. Slot 1 and the inspector show the champion with the current static scatter. Different blobs never change its strands. Each contender names its parent. Current-test catches and misses remain in the card statistics. New contenders are generated only after that complete comparison. If the champion defends its place, the next brood uses 35% less mutation, down to 0.0001%. If a contender wins, keep the current variance and generate the next brood from that new champion. The champion itself is never mutated. A wider brood at the minimum limit still uses this same parent; it does not choose a random ancestor. Exactly 20 complete rounds at that minimum trigger a wide brood around the current champion. Below 1%, only numeric parameters change. Pausing and reload retain the line-up, completed test totals and mutation progress. Changing any experiment setting restarts the five-test comparison around the same champion; a population change keeps the mutation cycle, while other condition or wide-amount changes restart it.

    A deliberately simple world.This is a two-dimensional circle-to-line overlap test. There is no fly animation, impact energy, tearing or structural collapse. Each scatter is counted immediately; speed controls only the pause for inspection. Turbo runs one scatter per display frame. Machine max removes that limit: it continuously batches tests at the available CPU speed and refreshes the display about five times a second. It briefly yields between batches so Pause and other controls remain usable. Speed never changes scores or selection rules. Estimated construction time affects fitness without making you wait for the web to be drawn. Silk is drawn bolder for clarity; scoring still uses the recipe’s actual thickness. Show flies affects visibility only. Spokes provide fixed support; the frame and fly behaviour do not evolve.

    Save, share, continue.The champion, mutation cycle and settings are stored only in this browser, separately from aelife2. Reload resumes paused with the same champion and mutation-cycle progress. An unfinished scatter restarts with the same line-up and previously completed test totals. A completed scatter continues at the next test. After the fifth test, reload begins the next brood. Version 1–3 web files remain importable with their exact recipes. Anchor costs can leave less silk for outer strands, and all scores are recalculated for the smaller flies under the same silk/time rules. An older timed mutation cycle restarts wide under the defence-only rule; an existing defence-only cycle keeps its variance. Older capture totals are cleared. Subsequent reloads preserve the new mutation progress and active comparison. Start fresh creates a new test and ancestor. Export a JSON file for a permanent copy. Import explicitly replaces the champion and pauses. Start fresh asks before replacing it. Evolution runs while this page is active; background tabs and sleeping devices may pause it.

    A NEW ANCESTOR

    Start a fresh experiment?

    This replaces the saved champion with the starting web. Export your champion first if you want to keep it.