# ShillScore tracking policy v1.0

Effective 12 October 2026. Scope: public X spot-long token recommendations on Base, Solana and Robinhood Chain. The scoring formula remains experimental v0.1; these rules govern selection and coverage, not investment advice or proof of an influencer's honesty.

## What counts as a call

A qualifying call is an influencer-authored original or quote post that positively recommends buying, entering or accumulating an identifiable token. The original post, its author, exact UTC publication time, supported chain and contract/mint must be verifiable. A contract-sharing post qualifies only when its text or attached media clearly presents an entry recommendation; a contract or price screenshot by itself is insufficient. A curator must explain the recommendation and token identity in the evidence notes before review.

Explicit entries such as “buying TOKEN here” or “entering TOKEN at this level” qualify. Vague watchlists, questions, neutral token news, price commentary, jokes, hindsight profit claims, plain reposts, replies, and private-chat screenshots do not qualify for this pilot. Conditional signals such as “buy if X happens” require a later unambiguous public confirmation; use that confirmation's publication time. Short calls, derivatives, leverage, sell/exit instructions and wallet activity are outside this spot-long pilot.

Include paid, unpaid and unknown-disclosure calls under the same rules. Record the disclosure separately. Payment cannot change inclusion, entry time, return calculation or score.

## Fixed coverage period

Starting or enabling tracking declares one fixed 90-day pilot per influencer, beginning at the server-recorded current UTC time T and ending at T + 90 × 24 hours. Pausing or re-enabling does not reset or extend it. The window includes T and excludes its end. Cohort dates are set before future outcomes are observed; do not choose a more favourable start after seeing prices.

Only reviewed qualifying calls published inside that influencer's declared pilot contribute to their pilot score. Calls before or after it remain historical case studies in the receipts explorer, with no contribution to pilot scores. A missing or invalid pilot declaration also leaves a reviewed call unscored. A call just before the end still receives its full seven-day follow-up; measurement and review may finish after the 90-day call collection period.

Capture every eligible call during the pilot, not just successful examples. Retain losses, waiting windows and calls with missing price evidence. Coverage is explicitly **partial recorded coverage**: the current collector and manual review process cannot establish complete historical capture. Profiles show the declared interval and counts of recorded, reviewed, awaiting and pending calls. Pauses, unavailable posts, outages and missing pages are coverage gaps; the elapsed calendar window alone never implies completeness. Automatic timeline collection requires X credentials; declared tracking by itself does not mean collection is running.

## Duplicate and repeat-call rules

1. The same original post + chain + token identity is one record. Canonical X/Twitter links remove tracking parameters. EVM address casing is ignored; Solana mint casing is retained.
2. The same influencer + chain + token within seven days of an existing active call is one episode. Keep the earliest qualifying post. Later shills, quote posts, re-entry hype and profit updates within that interval do not add wins or reset entry prices.
3. Exactly seven days or more after the previous episode's start, a fresh, explicit recommendation may begin another episode. A hindsight profit post still does not qualify.
4. Different influencers can each have a call for the same token. Different chains or different contracts/mints are different token identities, even when tickers match.
5. If a backfilled post is earlier than a stored call in that episode, correct the stored record to the earlier qualifying source and recollect its prices. Do not add a second scored call. Corrections preserve history, clear old automated measurement provenance, and require review again.
6. Retracted records retain their receipts and exact-post identity. They do not block a different qualifying post in the repeat-call interval because they are no longer active calls. Identical retracted records must be corrected/restored, not recreated.

Exact and seven-day checks run on creation, imports, corrections, restoration and review. Conflicting imports reject the entire batch. Existing records are not silently rewritten by this release. Related update receipts currently belong in curator evidence notes; the platform has no separate structured update-post attachment flow.

## Evidence, unresolved outcomes and corrections

Measure the fixed seven-day USD reference return from the original call, using the same pool and the documented first-full-minute close method. Ignore screenshots of an earlier entry, later personal PnL and post-call all-time highs as entry-price substitutes. Claims about realised profits require separate evidence and do not change the reference outcome.

Missing candles, delisted tokens or unavailable original posts are unresolved, not automatic -100% outcomes. A zero exit requires affirmative supporting evidence and curator review. Unresolved calls remain in the recorded coverage counts and out of the measured-return denominator. Publish reviewed count alongside unresolved counts so missing evidence cannot masquerade as a high win rate. Fewer than eight reviewed pilot calls stays provisional; no reviewed eligible pilot calls means no score.

Keep the original source and an inspection capture where permitted. Later edits or deletions do not remove an already verified losing call. If its original meaning or identity cannot be substantiated, retract it with a reason. Corrections and retractions preserve the local audit history; they never erase a loss merely because it harms the score. Public outcomes reflect curator review, not independent certification of completeness.

## Operating checklist

Start the influencer in `/tracking`; confirm the fixed window. Inspect each candidate against the inclusion rules. Check for an existing token episode before drafting. Record the original timestamp, mint/contract, intent notes and disclosure. Use `/queue` and `/measure` for price evidence. Review only after the required window and evidence are complete. Check the public profile's coverage counts and keep historical case studies clearly separate. Do not publish “overall lifetime win rate” claims from this fixed, partial pilot.