Skip to content
Flyweight
nightly fight cardRobinhood Chainchain 4663

Two real
fly brains.
One food cup.

A nightly fight card. Measured wiring, published seed, no house hand.

Each fighter carries a live token. The hour before the bell, its community feeds it. Then two measured brains decide the rest — and anyone can re-run the fight.

First card · coming soonHow it can’t be fixed
official channels
  • X· soon
  • GitHub· soon
  • Medium· soon
RED CORNER · challengerloading
28,017 of 139,255 neuronsactive now 0positions measured · activity modelled
vs
BLUE CORNER · challengerloading
28,017 of 139,255 neuronsactive now 0positions measured · activity modelled

Soma positions from the public FlyWire FAFB v783 whole-brain connectome. The activity is a model, not a recording — and we say so everywhere it appears.

next bellfirst card · date tba
--:--:--:--
nightly · 21:00 UTCweigh-in opens T-60m · bout 90s · proof at T+2mThe first card is announced 24 hours before its bell, on X.
scroll for the bout sheet
seed committed before the bell◆distinct wallets, not volume◆every bout on the record◆no betting pool◆token launches last◆positions measured · activity modelled◆
bout sheet · $FLYW · Robinhood Chain
pre-launch
contract address
TBA — the token launches last.

$FLYW goes live only after the show has an audience (Phase 3). When it does, the address appears here first, then on the official channels.

Beware fake tickets. Any token named Flyweight that shows up before this box fills is not ours. Verify the address here and on @flyweightRH only.

where it will trade
DexScreenersoonGMGNsoonBlockscoutsoon
01 · tale of the tape

Every brain demo so far is a solo act.

A measured fly connectome became public and within days it was wired to games, browsers and trading accounts. They all share the same dead end.

  1. 01

    The audience can only watch. No outcome, no stake, no reason to come back tomorrow night.

  2. 02

    The brain is decoration. Remove it and the product is unchanged, which the audience eventually notices.

  3. 03

    Nothing is verifiable. Results are asserted, never proven, so the first accusation of rigging has no answer.

  4. 04

    Each build is one brain alone. No opponent, no rivalry, nothing for two communities to argue about.

  5. 05

    Every one of them dies with the narrative that birthed it.

  6. the bet

    Nobody re-watches a simulation. Everybody re-watches a fight. Put two measured brains in one arena, let each one carry a token whose holders are already loud, publish the seed before the bell, and the show becomes an argument between communities. The rivalry is the retention.

them
vs
Flyweight
Brain sims you watch, and nothing happens
A brain sim that decides something, with stakes
One brain, alone
Two brains, head to head
Simulation as decoration
Simulation as the referee
The token does nothing but exist
Existing communities enter their own token as a fighter
“Trust us, it’s real”
Seed published before the bout; anyone can re-run it
A whale buys the outcome
Distinct buyers count, not size, and the effect is capped
Dies when the meta dies
The format works for any token
02 · the night

One fight, end to end.

From the card to the proof page in about a day. The only step that matters for trust happens an hour before the bell.

  1. T-24hCard announced

    Two tokens are matched. Both communities are tagged by name, with each fighter's record.

  2. T-60mSeed committed

    A random seed is generated and its hash is published with the snapshot rules. From here the house cannot change anything.

  3. T-60m → T-0Weigh-in window

    Every distinct wallet that buys a fighter's token feeds it. Net selling makes it timid. The total effect is hard-capped.

  4. T-0The bell — 90 seconds

    Two brains run against one shared arena. Sensory input → neurons → descending output → motion. The food cup is the only objective.

  5. FinishKnockout or decision

    Knockout: the loser's escape circuit fires and it flees the arena. Decision: whoever holds the cup when time expires.

  6. T+2mResult + proof

    Seed revealed, inputs revealed, replay downloadable. Anyone can re-run the bout and get the same result.

03 · weigh-in

Your chart is the diet.

Sixty minutes before the bell, buying pressure on a fighter’s token feeds it. Run the published model yourself — the same formula the show uses.

RED CORNERcondition
1.14×
well fed, aggressive
0.85 floor1.15 cap
a big sell landed
feed
+0.150
bitter
-0.015
shadow
0.00
BLUE CORNERcondition
1.00×
baseline
0.85 floor1.15 cap
a big sell landed
feed
+0.001
bitter
0.000
shadow
0.00
what the model saysgap +0.13

Try “One whale” against “Community”: a single wallet, however large, is one sugar drop. Three hundred small wallets are a meal. And even a 1.15× fighter is not assured a win — the cap is the point. Condition sets how the brain feels when the bell rings; the fight itself is decided by the two brains, the committed seed, and nothing else.

feeders   = min(distinct buying wallets, 200)
feed      = +0.15 × feeders / 200
bitter    = −0.10 × net selling
shadow    = −0.05 if a big sell landed
condition = clamp(1 + feed + bitter + shadow, 0.85, 1.15)
  • market → sense mapping · fixed before the card
  • A new distinct wallet buys during weigh-in
    Sugar on the legs (taste)
    ▲ Aggression up
  • Net sell pressure
    Bitter taste (taste)
    ▼ Aggression down
  • A single very large sell
    A looming shadow (visual)
    ▼ Flinch, then reset
  • Many trades in a short window
    Vibration (mechanosensory)
    ◆ Restlessness
  • Silence, no trades
    Dust on the eyes (visual)
    ▼ Sluggishness
the anti-whale rule
±15%

If volume fed the fighter, the richest wallet would buy every result. So the input is distinct buying wallets, not size, and the swing is hard-capped.

  • 400 wallets buying a littlea very well fed fighter
  • 1 wallet buying a lotone sugar drop, nothing more
  • Wallets under the dust floorignored entirely
  • Wallets funded from one sourcecounted once (sybil clustering)
  • Total condition swingcapped at ±15% of baseline

A real community beats a whale. No amount of money assures a win. That cap is the product.

04 · proof

Why nobody can fix a fight.

The engine is deterministic: the same seed plus the same inputs is the same fight, frame for frame. So we commit before, reveal after, and publish everything — including the losses.

before · T-60m
Committed
  • seed generated
  • hash(seed) published
  • snapshot block stated
  • mapping already public
  • engine version pinned
after · T+2m
Revealed
  • seed revealed
  • hash verified by anyone
  • snapshot data published
  • replay file downloadable
  • re-run reproduces the result
measured
  • Which neuron connects to which
  • How many synapses sit on each connection
  • The anatomical position of every cell body
  • Published under an open licence, credited on the proof page
modelled
  • Neuron dynamics (leaky integrate-and-fire, uniform parameters)
  • Which sensory cells a market event stimulates
  • Which descending cells map to which arena action
  • The aggression readout that turns firing into a lunge
not claimed
  • That the fly is conscious, alive, or aware
  • That aggression from a connectome simulation is scientifically validated — it is not, and we say so
  • That the outcome predicts anything about a token’s price

This is an experiment dressed as a sport. The wiring is real. The referee is engineered. Both facts stay on the site permanently.

05 · ringside

Three seats. No wallet needed for any of them.

seat 01
The viewer
  1. 1.Sees a clip
  2. 2.Opens the countdown
  3. 3.Watches 90 seconds
  4. 4.Reads the proof out of suspicion
  5. 5.Comes back tomorrow

No wallet, no email, no account. Ever.

seat 02
The token community
  1. 1.Their token is named on the card
  2. 2.Their group starts posting
  3. 3.Members feed the fighter in the weigh-in hour
  4. 4.They watch together and tag the rival
  5. 5.Win: a belt. Lose: a rematch demand

We drive attention into their token and take none of it.

seat 03
The sceptic
  1. 1.“This is rigged.”
  2. 2.Opens the proof page
  3. 3.Downloads the replay and the pinned engine
  4. 4.Re-runs it, gets the identical result
  5. 5.Posts the verification

We design for this user. Every anti-cheat feature is marketing the moment someone checks.

06 · the record

Every bout, including the dull ones.

A record with only highlights is a lie. This page will list every bout ever run, with its seed, its inputs and its replay — landed, lost, drawn, or boring.

date
main event
result
finish
proof
No bouts yet.

The first card is announced 24 hours before the bell. Nothing here will ever be retro-edited or removed; rows are written at fight time and stay.

a row will show
date · fighters · result · finish type and time
and link to
the replay file · the seed and hash · the input snapshot
plus, per fighter
record · debut · feeders per bout · retired seeds
07 · divisions

From maggot to champion.

  1. 01
    Maggot division
    Debut fighters, first three bouts
  2. 02
    Contender
    A winning record, eligible for the card
  3. 03
    Main event
    Top of the card, longer weigh-in window
  4. 04
    Champion
    Holds the belt until beaten. Refuse a defence for two weeks and it is stripped, publicly.
seasons, and death
  • A season runs eight weeks.
  • The champion’s fighter page is sealed permanently and its brain seed is retired.
  • A fighter knocked out three times in a row is retired too. The token can re-enter as a new fighter with a new seed and a blank record.
  • Nothing is deleted. The old page stays readable, which is what makes the wall worth reading.
  • rules of the ring · attack vs defence
  • threat
    Whale buys the outcome
    counter
    Distinct wallets only, capped ±15%
  • threat
    Sybil wallet farm
    counter
    Same-funder clustering + dust floor
  • threat
    Wash trading in the weigh-in
    counter
    Round trips inside the window are not counted as feeding
  • threat
    Fake token, no exit
    counter
    Sell path proven before eligibility
  • threat
    House fixes the result
    counter
    Seed committed before the window opens
  • threat
    Selective highlight reel
    counter
    Every bout published, losses too
  • threat
    Rented community for one night
    counter
    Feeder counts are shown on the fighter page, so spikes are visible
08 · the purse

Phase 1 makes nothing. On purpose.

Charging for access to an empty arena is how projects die before launch. Revenue arrives when a slot on the card is worth paying for.

phase 2 · the card becomes scarce
Challenge slot, undercard
book a bout
0.03 ETH
Challenge slot, main event
top of the card
0.08 ETH
Rematch (same pairing)
grudge pricing
0.05 ETH
Corner sponsor, one night
logo on one corner
0.10 ETH
Venue sponsor, one week
the whole week
0.40 ETH

Paying in $FLYW gets a discount, and the discounted portion is burned rather than collected. Phase 3 adds the launchpad creator fee on $FLYW trading, set low at 100 bps, accruing to a published treasury address.

allocation of every unit collected
  • Engineering and infrastructure · the simulation is the cost40%
  • Buyback and burn · executed publicly25%
  • Prize pool · season finals and belt defences20%
  • Content · clip cutting, art, the weekly recap10%
  • Reserve 5%

Published as a running ledger. If the split changes, the change is posted before it takes effect, never after.

09 · $FLYW

One token, one neuron.

Supply equals the neuron count of the published connectome. Not a round marketing number — the measured one.

total supply
0
units · never inflated
Supply
166,700 — one per neuron, never inflated
Team allocation
0%
Presale
none
Launch
fair launch on a Robinhood Chain launchpad, quoted in ETH
Liquidity
locked by the launchpad at graduation
Creator fee
100 bps, to the published treasury address
Timing
after the show has an audience, not before
the one rule we will not break
$FLYW never buys a better chance of winning.
  • Buy a better chance of winning
  • Change a fighter’s condition
  • Influence matchmaking odds
  • Apply a hidden multiplier of any kind

Everything the token does is access, identity and cost — never odds.

holder tiers · what it unlocks
Hold 1
  • ✓ A bound neuron id, visible when it fires
  • ✓ Card vote (one vote)
Hold 100
  • ✓ Named cluster on the live brain panel
  • ✓ Challenge fee discount, tier 1
Hold 1,000
  • ✓ Handle on the corner card for the night
  • ✓ Challenge fee discount, tier 2
  • ✓ First look at the next card
Hold 5,000
  • ✓ Propose a main event directly to matchmaking
  • ✓ Permanent name on the venue wall
demand comes from
  • · A neuron bound to you, flashing when it fires
  • · First look at the card and matchmaking votes
  • · Challenge fee discounts
  • · A name on the venue wall
supply leaves through
  • · Discounted fee portion burns
  • · Buyback and burn from revenue
  • · Belt defence entry burns
  • · Season pass mints burn a fee
  • · No emissions, ever
the flywheel
1more communities want a slot
2challenge fees paid — discount in $FLYW
3discounted portion burns
4holding needed to see the card first and to vote
5bigger audience per bout → a slot is worth more
10 · under the canvas

Three layers. One of them cannot be touched.

browserreads a one-way event stream; never has authority
Arena renderer
WebGL, two fighters, one cup
Brain panels
spikes at measured positions
show serverauthoritative, single source of truth
Matchmaker
builds the card, checks eligibility
Commit service
seeds, hashes, timestamps
Market watcher
reads chain logs, counts distinct buyers
Sim runner
two brains, fixed timestep, deterministic
Recorder
writes the replay and the proof bundle
Robinhood Chain · 4663read-only RPC and explorer reads
Token pools
every fighter’s market
Launchpad events
eligibility source of truth
Treasury
public address, fees land here
server-authoritative, and why
The seed
a client-side seed is a client-side win
The snapshot block
fixed by us, in advance
Distinct buyers
counted from chain logs, never reported by the browser
The simulation
the whole product is that nobody can touch it mid-fight
The record
append-only; results are never edited

Two brains per bout, each on its own worker thread, fixed timestep, integer-safe accumulation so results never drift across hardware. Fights are computed ahead of broadcast, which is why the replay is downloadable the moment the bell stops.

11 · the honest take

What we decided, and what could go wrong.

Transparency is cheaper than trust bought later. The risks stay on this page for as long as the show runs.

key decisions · and why
Fighters are existing tokens, not ours
Solves cold start. Their communities bring the audience.
Distinct wallets, not volume, feed a fighter
Volume-based feeding lets one whale buy every result.
Condition capped at ±15%
Keeps the brain, not the wallet, as the thing that decides.
Seed committed before the weigh-in opens
The first rigging accusation arrives on night one.
Every bout published, including dull ones
A highlights-only record is read as a lie, correctly.
No betting pool at launch
Wagering changes the legal shape of the product entirely.
Token launches last
A token before an audience is the failure mode we keep seeing.
Supply 166,700
Makes “one token, one neuron” literally true.
Fight pre-computed, then broadcast
Guarantees determinism and makes the replay instantly available.
Limits stated on the site permanently
Overclaiming the science is the fastest way to lose credibility.
strengths
  • The only head-to-head format in a category of solo demos
  • Each new fighter brings its own community as audience
  • The record compounds and cannot be copied on launch day
  • Verifiability becomes the reputation: the show you can check
risks
  • The narrative may cool before we ship. The format works for any token, not only this meta.
  • Communities may ignore the invitation. We open with the most active ones and let the rivalry sell it.
  • Outcomes are dynamics plus noise plus a capped input. We present it as sport, never as intelligence.
  • Simulation cost grows with the card. Bouts are pre-computed and queued, not streamed live.
  • Another project’s ticker as a fighter is fan content. No endorsement implied; removed on request.
what decides phase 3
  • · Live viewers at the bell, two-week rolling average
  • · Returning viewers night over night
  • · Distinct feeders per bout
  • · Unpaid mentions from fighter communities
  • · Independent verifications of a result
  • · Challenge requests we did not solicit
what we deliberately ignore
  • · Token price
  • · Follower count
  • · Total volume (it rewards whales)
  • · Highlight views without return visits

If these do not move, we do not launch a token. The show is the product.

12 · fight schedule

The token is the last thing we ship.

  1. Phase 1 live
    First bell
    week 1
    • Two deterministic brains on one machine
    • Distinct-buyer counting from chain logs
    • Arena renderer and brain panels
    • Commit-reveal and the proof page
    • Exhibition bouts, then the first public card
  2. Phase 2next
    A card worth a slot
    weeks 2–5
    • Fighter pages, records, the wall
    • Challenge booking and eligibility automation
    • Cards, results and proofs posted to X minutes after the bell
    • Auto-cut clips at every knockout
    • First belt, first defence
  3. Phase 3gated
    The token
    weeks 6–10, gated
    • Gate: two weeks of live viewers and one unsolicited paid slot
    • $FLYW fair launch on a Robinhood Chain launchpad
    • Neuron binding for holders
    • Fee discounts and the burn ledger
    • Card voting
  4. Phase 4later
    The venue
    months 3–6
    • Seasons, belts per division, a hall of retired fighters
    • Open engine release so anyone can verify or fork
    • Any token on chain can enter
    • Tournament nights and community undercards
    • Embeddable fight widget