FieldCall

Paid calls for wildlife sightings. Anyone funds a call; naturalists post on iNaturalist as they always do; GenLayer validators check each sighting, look at the photos when the call asks for something specific, and pay the observer.

Conservation work runs on records: where an invasive hornet is nesting, whether hedgehogs still breed in a town, whether a rare orchid flowered this year. Millions of people already photograph wildlife and post it to iNaturalist, where a community identifies every record. But nobody pays for the one record that matters, and when someone tries (a bounty for nests, a reward for a first county record), every sighting has to be checked by hand and every payout trusted to whoever holds the money.

FieldCall turns that into a contract:

1. A sponsor opens a call: an iNaturalist taxon (a species, or any genus or family), an iNaturalist place, the dates sightings must fall between, a reward per sighting, and the rules: research grade only, at most N per observer, a minimum account age, a location accuracy. Optionally a photo condition, in plain words: "The photo shows a hornet nest." Validators read the taxon and place from iNaturalist and freeze them. Anyone can add GEN to a call's pot.
2. Anyone submits an iNaturalist observation to a call: their own, or someone else's.
3. Every validator reads the observation and the observer's public profile from iNaturalist and checks in code: the community identification is the call's taxon or below it, quality grade, not captive, the date, the places the observation falls inside, location accuracy and account age. If the call has a photo condition, each validator looks at the photos (up to three, one at a time) and answers MEETS, NOT_MEETS or UNCLEAR; they must agree. They never re-identify the species: the iNaturalist community already did.
4. A qualifying sighting reserves the reward for the wallet address in the observer's iNaturalist profile, which only the observer can edit. Anyone can submit; nobody can redirect the money. An observer who has not linked a wallet yet has the reward held until the call's collect date; they add a 0x… address to their profile and anyone presses Collect.
5. The side a ruling goes against has 10 minutes to ask once for a fresh reading (an identification can change). Every ruling stores a plain-English reason, a timeline and an evidence packet with sha256 hashes of what was read, including each photo looked at. When the call's collect date passes, anyone closes it, and anyone can then return each funder's share of what is left, in proportion to what they put in, one funder per transaction.

Live app: fieldcall-bars26.vercel.app. Paste any iNaturalist observation and see, without a wallet, which open calls it would earn and why. Writes need MetaMask on GenLayer Studio (chain 61999); Test GEN funds your wallet.

Contract: 0x58576b0a8C0091b8E7E276EDCa9b5D2d9bC826aE on GenLayer Studio.

Why GenLayer

• Looking at the photo. "The photo shows a nest", "a living hedgehog", "flowers open": the thing a sponsor actually wants is in the image, not in any field of the record. Each validator downloads the photo and looks at it; the answer set is fixed (MEETS, NOT_MEETS, UNCLEAR) so their agreement is strict. Verified on Studio: a nest photo and a photo of an adult hornet on a branch, from the same observer on the same day, get opposite answers.
• Reading the record. Every validator queries iNaturalist itself; the identification, the place and the observer's wallet are read by each of them, so no single party reports the facts money depends on.

Verified live, on real iNaturalist records

scripts/demo.mjs opened two demo calls with test wallets and submitted real observations of the yellow-legged (Asian) hornet, Vespa velutina, which the UK asks the public to report. Every transaction, reason, evidence packet and the balance reconciliation are in docs/REPRODUCTION.md.

Call	Observation	Ruling and stored reason
Asian hornet nests, UK, 0.5 GEN, photo condition "The photo shows a hornet nest"	397236824: a nest in a tree, Romsey, 2026-08-20	Qualifies. Validators looked at the photo and it shows a nest. A co-funder's fresh reading confirmed it. The observer has no wallet in their profile, so the reward is held for them
same	397236834: an adult hornet on a branch, same observer, same day	Does not qualify: validators looked at three photos, NOT_MEETS each
same	411158753: a European hornet, the native lookalike	Does not qualify: Vespa crabro is not Vespa velutina (decided in code; the scout's fresh reading agreed)
same	411204951: an Asian hornet in France	Does not qualify: not inside the United Kingdom
same	182971213: Hull, 2023	Does not qualify: outside the call's dates
Asian hornet sightings, UK, 0.25 GEN, 2 per observer	three adults by the same observer	Qualifies, qualifies, then refused: the observer reached the call's limit of 2

Also refused on-chain, with the GEN sent back: a taxon name instead of an id, a taxon and a place iNaturalist does not have, a call that has already ended, a reward below 0.1 GEN, a pot smaller than one reward, a start date of 2026-02-31, and a 0.05 GEN top-up (below the 0.1 GEN minimum). And refused by the contract: submitting an observation that is already open, already earned its reward, or was already submitted three times; a scout paying out early (only a funder can); collecting a held reward before the observer links a wallet; and closing a call before its collect date. The contract's balance was reconciled with totals() (open pots plus reserved rewards) at three points of the run, each time after every transfer had finalized.

From the live app with MetaMask: a tester opened a third call from the Live hedgehogs, UK preset (photo condition "The photo shows a living hedgehog") and submitted three real observations: a hedgehog walking on a road qualified (validators looked: MEETS, reward held for its observer), a dead hedgehog did not (NOT_MEETS), and a live hedgehog whose observer opened their account two days before the sighting did not (the call requires 30 days; decided in code). Transactions in docs/REPRODUCTION.md.

Who can do what (and the attacks this blocks)

Rule	Why
The reward goes only to the 0x address in the observer's own iNaturalist profile	Submitting someone else's observation cannot redirect their reward; anyone can submit, nobody can steal
Taxonomy is the iNaturalist community's: the observation's identification must be the call's taxon or below it	Validators judge the photo condition only; a lookalike species fails in code, not by opinion
Research grade, not captive, inside the place, inside the dates, located precisely enough, account old enough: all in code	Every validator gets the same answer from the same record; no model judgement where a field decides
A photo condition is answered per photo with a fixed answer set; the first photo that clearly shows it decides	Strict consensus on an image judgement; a vague photo is UNCLEAR and reserves nothing
One payout per observation per call, at most N per observer, and a funder's wallet is never paid by its own call	One prolific account, or the sponsor, cannot drain the pot
Only the side a ruling goes against asks for a fresh reading, once each, inside 10 minutes	An identification that changes can still be read; nobody spends the other side's answer
Early settlement only against your own interest (a funder pays now, the submitter accepts a refusal)	The submitter cannot rush a payout; funders cannot rush a refusal
iNaturalist outages, unreadable profiles and photos are UNCLEAR: nothing reserved, submit again	An outage never decides a sighting
Held rewards wait until the collect date (call end + 30 days), then return to the pot; the pot returns to funders pro rata	Money never strands; observers who join late still get paid
Closing does a fixed amount of work; each funder's share is then returned by its own refund transaction, which anyone can send (always to that funder)	No call can grow so large that closing it runs out of gas and strands the funders' GEN
Funders, observer counts and submission counts live in their own maps, never in a list on the call; top-ups are at least 0.1 GEN; one observation can be submitted to a call at most 3 times	Spam cannot make any write more expensive than the first one
Payable calls never revert; they refund and return REFUNDED: <reason>, including for dates that do not exist (2026-02-31)

The app

• Would my sighting qualify? Paste any iNaturalist observation link: the photo, the identification, and for every open call the same checks as the contract, ticked or crossed, with the reward it would earn. One click submits it.
• Every call shows its taxon and place (linked to iNaturalist), the reward, how much of the pot is paid, reserved and free, its terms, a submit box and a fund box.
• Every sighting shows the observation's photo with its author's attribution, the ruling and its reason, the window countdown with the actions your wallet can take, and what the validators read: the facts, the photos they looked at with their answer under each, the evidence packet hashes and a link to the GBIF copy of the record.
• Open a call with presets (Asian hornet nests, Asian hornet sightings, live hedgehogs) or search any iNaturalist taxon and place.
• Studio's 30 requests per minute: page loads read the registry server-side with one get_calls and one get_sightings call, CDN-cached; iNaturalist is read straight from the browser.

Run it

pip install -r requirements.txt
genvm-lint check contracts/field_call.py
pytest tests/direct/ -v                     # 55 tests: iNaturalist, profiles, photos and the LLM mocked

npm install
node scripts/deploy.mjs
node scripts/verify-code.mjs <address>
node scripts/demo.mjs <address>             # the live run above (about 30 minutes)
node scripts/reproduction.mjs               # regenerate docs/REPRODUCTION.md

cd frontend && cp .env.example .env.local && npm run dev

Changes in v1.1 (after steward review)

• Closing cannot be blocked by size. v1.0's close() looped over every sighting and every funder in one transaction, and both lists could be padded cheaply. Now close() only moves the pot and the expired held rewards into returnable; every funder is then repaid by refund(call_id, funder), one bounded transaction each, sent by anyone, in any order, with the last one taking the rounding remainder. Funders, per-observer counts and submission counts moved out of the call record into their own maps, top-ups are at least 0.1 GEN, and an observation can be submitted to a call at most three times. test_stress_hundreds_of_funders_and_refused_sightings_still_close_and_repay_everyone pads a call with 301 funders and 120 refused submissions, checks that close() reads no sighting and no funder, then repays all 301 and checks every wei left.
• Impossible dates are refunded. create_call checks that starts and ends are real calendar dates before any date arithmetic, so 2026-02-31 returns REFUNDED: ... with the GEN, instead of raising and keeping it; covered by tests that record every transfer and by the live run.

Limits, stated plainly

• The identification is the iNaturalist community's. A wrong community ID that reaches research grade would pass until someone corrects it; that is why funders can ask for a fresh reading, and why calls can require research grade and an account age.
• Validators look at up to three photos. A condition that only a later photo shows may be missed; observers can reorder photos. Photos are judged as images: a condition about something not visible (a sound, a smell) cannot be met.
• A wallet in a profile proves control of the iNaturalist account, not who the person is. Per-observer caps and account age limit, but do not remove, sock-puppet accounts; research grade still needs other people's identifications.
• The demo calls are funded by test wallets; the observers in the demo have not linked wallets, so their rewards are held and will return to the funders.

Observations and photos are © their authors on iNaturalist and are shown with their attribution and licence.

MIT licensed.