Whitepaper

1. Mission

AI tools made everyone a writer. That created a new problem. AI-generated text carries patterns: em-dash overuse, formulaic transitions, hollow intensifiers, uniform paragraph rhythm. Readers notice. Search engines notice. When everything reads the same, nothing stands out.

The Avoid AI Writing project exists to fix that loop. The detection skill is open source. The browser extension runs locally. The paid web app pays for itself by burning AVOID, the project's utility token, every time someone uses it. The supply only goes one direction.

See /mission for the longer version.

2. What the project is

Avoid AI Writing scans text for patterns common in AI-generated writing. It runs in two places: a free Chrome extension that scans locally inside the browser, and a paid web app at avoidaiwriting.com that runs a deeper Claude-powered audit and returns a rewrite. The detection logic is open source under MIT license at github.com/conorbronsdon/avoid-ai-writing.

AVOID is a Solana SPL token used to pay for the paid audit. Each audit burns a fixed-USD-denominated amount of AVOID, removing those tokens from circulation forever. Audits can also be paid with a credit card; the project then uses those receipts to buy back AVOID and apply the same burn mechanic. Crypto-native and card-only users land on the same deflationary loop.

The token was created by community members on pump.fun the same day the open-source detector hit the GitHub front page. The project operator started with zero token holdings and has accumulated holdings only through the published buyback program.

3. The problem the project addresses

Existing AI-detection tools have known limitations. Some return high false-positive rates on human writing. Others rely on closed-source classifiers that publishers cannot audit. Several charge per query without offering the underlying patterns for inspection or improvement.

The project takes a different approach. Detection patterns ship as a public skill file: anyone can read and propose updates. The browser extension runs locally so source text never leaves the device. The paid web-app audit uses a documented prompt and returns both flagged passages and a rewrite the user can accept or reject. See /tech for the engine deep-dive.

4. Detection technology

The detector covers approximately 36 pattern categories across four severity tiers:

• Tier 0 (chatbot leakage) — literal artifacts of model responses. "I hope this helps," "Let me know if you need anything else."
• Tier 1 (LLM vocabulary) — high-frequency words and phrases that cluster in AI-generated text. "delve," "tapestry," "comprehensive," figurative "navigate," plus 40+ others.
• Tier 2 (structural patterns) — rhythm uniformity, paragraph-length uniformity, em-dash overuse, formulaic transitions, hollow intensifiers.
• Tier 3 (stylistic markers) — sycophantic tone, novelty inflation, false-concession constructions, vague attributions, cutoff disclaimers.

Two-pass rewrite

System architecture

Scoring runs locally in JavaScript inside the browser. The Chrome extension reads the focused text field, runs the patterns, and renders results in a shadow-DOM panel. No text is transmitted unless the user explicitly clicks the deep-audit button, which opens the web app. Source text is not stored after the audit response is returned. See /tech for full detail.

5. Token utility

AVOID has two on-chain functions: it gates the paid audit (where it's burned on use) and it backs the staking system. Anyone holding AVOID can also burn it directly to reduce supply, with two documented methods below.

Burn-on-use

Each web-app audit costs a fixed USD amount, currently set at $0.25. The token amount required for a single audit is recalculated every 30 seconds against the live AVOID/USD price from DexScreener and Jupiter, with automatic fallback. Whatever AVOID amount sits at $0.25 when the user submits gets burned via the SPL Burn instruction (which decreases total supply on-chain). The audit result returns once the burn confirms on-chain.

The full flow

Burn AVOID yourself

Anyone holding AVOID can burn it. Two methods:

1. SPL Burn instruction (preferred — reduces total supply on-chain). Open the token in Solscan, connect your wallet, find your AVOID account, and use the "Burn" action. From the CLI: spl-token burn <your-token-account> <amount>. Solscan and DexScreener register the burn against the mint's total supply.
2. Send to the Solana incinerator (transfer-to-burn — locks tokens forever). The incinerator address has no valid keypair, so tokens sent there are unrecoverable. This does not reduce on-chain total supply, but it does remove the tokens from circulation. Use this if your wallet doesn't expose a Burn UI.

The project's own buyback flow uses the SPL Burn instruction (method 1). When the operator buys and burns AVOID for credit-card audit revenue, those burns reduce total supply and are reflected in the live burn percentage on the supply bar above and on /tokenomics.

Staking

Users who lock AVOID for 7, 30, 90, or 180 days receive a share of platform revenue paid in additional AVOID. The locked tokens are removed from circulating supply for the duration of the lock. Operator stake is committed at launch (see section 6). Full design at /stake.

6. Tokenomics

Supply

• Genesis supply: 1,000,000,000 AVOID, minted on Solana 2026-04-05
• Decimals: 6
• Mint authority: disabled at genesis. No new tokens can be created.
• Freeze authority: disabled at genesis.
• Verify on-chain: Solscan token page.

Locked and staked supply (10% combined)

A combined 10% of total supply is removed from active circulation by the project operator across two mechanics:

All bars scaled to total supply (1,000,000,000 AVOID at genesis). Burn percentage is fetched from the /api/supply endpoint when this page loads (endpoint cache: 60 seconds; refresh the page for an updated reading). Operator-controlled slices total 12.33%; of those, 10% is locked or staked at staking launch. See /tokenomics for the per-wallet breakdown. Burn activity is verifiable on Solscan; buyback and burn events are announced on @avoidaiwriting on X. Auto-tracking the operator wallets on-chain — and poll-based live updates — is planned for v2.

• 3% in long-term Streamflow vesting: 1.5% for 3 months (contract) and 1.5% for 6 months (contract).
• 7% in tier-aligned Streamflow cliff locks, mirroring the staking tier structure: 1% with a 30-day cliff (contract) + 2% with a 90-day cliff (contract) + 4% with a 180-day cliff (contract). Cliff-only — tokens unlock at the cliff date, not before. All five contracts are chain-enforced; no operator key can move the locked tokens before each contract's cliff or vesting completes. Per-wallet detail on /tokenomics.

The custom staking program at /stake is a separate mechanism (hybrid vault + off-chain lockup tracking) and is currently paused pending architecture review for migration to fully on-chain enforcement before community deposits open.

Fee allocation

Distribution origin

The token was created by community members on pump.fun on 2026-04-05. The project operator began with zero token holdings. All operator-side holdings have been accumulated through the published buyback program. As of mid-April 2026 the dev primary wallet plus locked contracts together held approximately 10.6% of supply (7.6% in the dev primary wallet from buybacks, 3% in Streamflow vesting locks). On 2026-04-26 the operator locked an additional 7% from the dev primary into three tier-aligned Streamflow cliff contracts (1%/30d + 2%/90d + 4%/180d), leaving 0.6% liquid in the dev primary for operational needs.

The remaining operator-controlled supply sits across two additional wallets disclosed at week 1 and on /tokenomics: the Fernwillow primary wallet (1.01%, day-zero accumulation that pre-dates project operations and has been actively traded), and the marketing wallet (0.72%, ~7.23M tokens, fee-funded marketing budget). Adding these to the 0.6% dev primary leftover gives the 2.33% liquid slice in the supply bar above; the full 12.33% of operator-controlled supply is the sum of 10% Streamflow-locked (3% long-term vesting + 7% tier-aligned cliff) + 2.33% liquid.

7. Staking architecture

Status: community staking is paused pending architecture review. The originally-proposed Phase 1 design is documented below for reference. The operator's 10% commitment ships via Streamflow contracts instead — see section 4 above for the five contract addresses, all chain-enforced.

Original Phase 1 design (paused): users lock AVOID for 7, 30, 90, or 180 days; the locked tokens earn a share of platform revenue paid in fresh AVOID. Reward distribution activates when card revenue starts routing into the reward pool (mobile apps shipping ~4-6 weeks post-staking-launch).

The custom staking program above is currently paused. The architecture as designed uses a hot vault key with off-chain (KV) lockup tracking — fast to ship, but it leaves an operator-key surface that can move user-staked tokens regardless of the off-chain lockup state. Pending architecture review, community staking is closed and the operator's 10% commitment is held via Streamflow contracts instead (chain-enforced, irreversible until cliff). Originally-proposed design + paused status: /staking-design; on-chain operator commitments: /tokenomics.

8. Versus alternatives

How AVOID + the Avoid AI Writing project compare to common AI-detection tools:

Comparison is based on publicly documented features as of 2026-04-25. We do not claim parity on classifier accuracy benchmarks. Detection is heuristic for every tool listed.

9. Trust signals

10. Operating entity and governance

Fernwillow Holdings LLC operates the project. The LLC is a Washington limited liability company approved 2026-04-25; it holds the software, the website, the Stripe account, and is the registered publisher of the Chrome extension.

Conor Bronsdon is a member and the project lead. Public bio: /about. Contact: support@fernwillowholdings.com.

The detection skill is open source under MIT and accepts community pattern proposals via pull request: github.com/conorbronsdon/avoid-ai-writing. Project terms: /terms. Extension privacy: /privacy-extension.

Disclaimer by Conor Bronsdon: All views, opinions, and statements expressed on this account and by this project are solely my own and are made in my personal capacity. They do not reflect, and should not be construed as reflecting, the views, positions, or policies of my employer. This account and the Avoid AI Writing project are not affiliated with, authorized by, or endorsed by my employer in any way.

11. Risk factors and disclaimers

This document is informational. It is not investment advice, legal advice, or tax advice. Anyone considering holding AVOID should do their own research and, if relevant, consult their own advisors.

12. References

• Project site: avoidaiwriting.com
• Mission: /mission
• Roadmap + tokenomics: /plan
• How it works: /tech
• Staking design: /stake
• Free Chrome extension: /extension
• Paid audit: /rewrite
• Open-source detector: github.com/conorbronsdon/avoid-ai-writing
• DexScreener: pair
• Jupiter: token page
• Solscan: token
• pump.fun: coin