Contributor Guide

This page is the original program write-up. $ARCTIVAR on Solana is a separate token. A finished task does not send tokens, and points are not a claim on the token. Supply and unlocks are not published here.

Points

How your submissions turn into Points — the five factors behind them, how a submission is scored, and when they settle.

Points are Arctivar's record of what you contribute. A submission counts once you sign it on-chain, and Points settle on a two-week window called an Epoch.


What Drives Your Points

Five factors move your point total:

FactorWhat it measures
VolumeHow many valid tasks you complete
DifficultyHow hard the tasks you take on are, rated 1 to 5 stars
ScoreThe quality of each signed submission — detailed below
DiversityHow widely your tasks spread across scenarios and skills
ReferralsWhat the contributors you invite actually produce

The first four are not a checklist. No single one carries the total, and neglecting one drags it down: a hundred trivial submissions in one scenario are worth less than a small set of hard, varied, well-executed ones.

On referrals: only direct invitees count, and only by what they produce. Accounts that never complete a task add nothing; headcount earns no credit.

How a Submission Is Scored

Four of the five factors are tallied across the Epoch. Score is per submission: assigned once the submission has been verified, then fed into the Epoch total as one input of five.

Each completed task produces one submission, evaluated on its own:

Complete task → Trajectory uploaded → Verified by replay → Score assigned

Scoring is algorithmic end to end; no other contributor reviews your submissions. Each trajectory passes three layers:

  1. Sanity & Replay Checks

    Schema validation and trajectory integrity: no corrupted or truncated data, and the scene config has to reload (asset IDs valid, config parseable). Every submission is then replayed in simulation and checked for physical consistency; invalid ones never reach training.

  2. Rule-Based Success Checking

    Did the final object pose fall inside the goal region? Did the robot avoid catastrophic collisions? Were task-specific constraints respected?

  3. VLM-Based Assessment

    A vision-language model evaluates key frames against the language goal, checking that the motion is purposeful rather than random.

The three layers resolve to one quality score, Q(traj): it gates entry to the training pool, sets the trajectory's weight there, and through the Score factor sets how much the submission adds to your Points.

What Produces a Higher Score

Pre-training submissions are scored on the demonstration: success, efficiency, motion smoothness.

Post-training submissions are governed by one hard rule and scored on two quantities. The rule: the task has to be completed, and the 30-second takeover budget is what enforces it — burn through it and the run fails outright, with no score and no credit. Within a completed run, the score then falls with trajectory length and rises with the share of the run the policy drove. Minimal, precise input where the policy is weakest is what the correction dataset needs. See Post-training Tasks.

What Isn't Scored

  • Failed attempts (not uploaded, no penalty)
  • Spammy or random behavior (caught by automated evaluation)
  • Duplicate or near-duplicate trajectories (detected during ingestion)
  • Submissions past the per-task caps — see Don't Farm a Single Task

What Moves Points Fastest

Points track data produced, not time spent. Three things move them fastest, because they are what the pipeline is short of.

Take harder tasks. Hard demonstrations are where policies still fail, which is why they are worth several easy ones.

Do more tasks. Volume is the base the other factors act on; every signed submission in the Epoch adds to it.

Spread out. Diversity is measured on two axes: the scenarios your tasks cover and the skills they exercise. Many scenarios with one skill, or many skills in one scenario, leaves diversity near the floor. Policies generalize from spread, not repetition.

Don't Farm a Single Task

The Hub now caps each task at 5 completions per account, and what counts for that task is the average score across your completions of it — not your best one. Repeating a task you already did well can therefore lower what it is worth to you: a few sloppy repeats pull the average down. Repeat a task to do it better, not to do it more.

That cap landed after the platform had been running, so contribution from before it — when a task could be repeated without limit — is credited under one rule: per task, only the first 10 submissions from an address that pass evaluation count. Failed submissions don't consume the allowance. Anything past the tenth valid submission is excluded from scoring, from Points, and from every reward statistic, and addresses that ran far past the cap carry a penalty factor.

Task slots are finite. An address filling them with repeats crowds out contributors who would have brought a different grasp or a different failure, and the task comes back narrow — short on realism and diversity, which is not data worth training on. If you want more from a task family, take the next task in it.


How Points Are Released

Points are issued in two phases, split by the retroactive snapshot taken when the Point System launched, at 16:00 UTC on 29 July 2026. Everything signed before that moment was settled once, in a single backward-looking grant; everything after it is settled Epoch by Epoch.

PhaseCoversFrequency
1 · RetroactiveEvery signed transaction recorded before the snapshotOne time, already issued
2 · Epoch EmissionEverything signed after the snapshotEvery two weeks, ongoing

Phase 1 — Retroactive

Your history counts. Every transaction you had signed before the snapshot was tallied once and credited in a single retroactive grant, and nothing needs to be redone. Work you completed but never signed was not recorded and did not enter that grant. This phase is closed.

Phase 2 — Epoch Emission

From the snapshot forward, Points are issued per Epoch — a two-week window. Epochs run back to back: no start date to wait for, no gap between them.

  1. Cut-off — Friday at 00:00 UTC

    Each Epoch closes Friday at 00:00 UTC. That timestamp bounds the calculation, not the payout: everything signed inside the window is what gets counted.

  2. Distribution — the following Monday, 12:00 UTC

    Calculation and settlement run over the weekend. Points land in your account on the following Monday at 12:00 UTC.

Only Signed Submissions Count

Both phases run on the same gate: a submission counts only once it is signed on-chain, with a transaction recorded on Base. Completing a task is not the last step; signing it is.

Finish a task in one Epoch, leave it unsigned, and it earns nothing for that Epoch. Sign it later and it counts in the Epoch you signed it in — signing late costs you nothing but time. It takes one click from your Portfolio, Base gas runs a tiny fraction of a cent, and gasless campaigns cover it outright while they are running. See Sign & Verify.


Planned Community Distribution

The Contributor Rewards allocation, 10% of $ARCTIVAR total supply, is distributed through the Community Contributor Airdrop in two seasons.

SeasonFocusAllocation (% of supply)
Season 1 · Pre-TGESimulation40,000,000 (4.00%)
Season 2 · Post-TGESim-to-real60,000,000 (6.00%)

Accepted contributions earn Arctivar Points. Points may be considered as one input for eligibility and reward allocation under separately published program terms. Holding Points guarantees no eligibility, allocation, or distribution, and the Point System establishes no token balance, conversion ratio, or claim right.

Season 1 covers contributions made before TGE and the Points they earn. Its unlock schedule is set out in Tokenomics. Season 2 terms will be announced when the season opens.

Eligibility, allocation method, timing, geographic availability, and claim terms will be published separately before any claim event. Independently of that, your signed submissions remain an on-chain record of what you contributed: see Sign & Verify and Data Provenance.


Going Deeper

Next: FAQ →
Points