Pianist VS Bot

The Acoustic Turing League

[Draft] Core Constitution and Technical Auditing Specification of the 1st Acoustic Turing League

Organizer: pianist.bot (run by an individual organizer; a legal entity may be set up later)

Contact: [email protected]

Status: Trial phase. Warm-up rounds do not count toward any ranking.

Trial Notice and Purpose

This draft combines the League’s constitution with its trial rules. Rules may change before the official season, and registered participants will be told in advance. An oversight panel of technical and academic reviewers is being formed, and qualified people are welcome to join. Nothing here guarantees prizes, rankings or future placements unless it is confirmed in writing before the official season.

The League exists to collect honest audience feedback in an entertaining way, so that robotics teams can improve. The score is a simple summary, not a verdict on any robot’s overall ability. For this reason the League does not fix the human-to-robot ratio, the repertoire size or the performance length. Like formations in football, these are each team’s choice, and the uncertainty is part of the game. Teams that want precise engineering comparison can use metrics such as torque or degrees of freedom; this League measures what listeners perceive.

Definitions

Robot performance [proposed, organizer to confirm]. A robotic system physically presses the keys of an acoustic piano, controlled by software, with no live human teleoperation. The participant declares the type of system (robotic hand, arm or other key-pressing mechanism) before its window.

Not allowed in the trial [proposed, organizer to confirm]. Using recorded key data from a human performance (for example MIDI of a human take) as the control source, and any hidden human remote control. An exception must be declared and approved in advance, and disclosed to the audience at the reveal.

Human performance. A human pianist plays the same acoustic piano setup behind the same curtain.

Article I. Infrastructure and Autonomy

  1. Organizer. pianist.bot maintains the central website, the voting interface, the official schedule and the data services, and provides an iframe or equivalent integration for the active participant’s page.
  2. Participants. Each participant arranges and funds its own venue, piano, robot or human performers, local technical setup and broadcast. Streams may run on the participant’s own site or any lawful platform.
  3. No guarantees. The organizer does not guarantee the availability, reach or bandwidth of participant-controlled infrastructure.
  4. Verification is staged. The League describes openly what verification exists at each stage and may strengthen it as resources grow.

Article II. Scheduling and Exclusive Windows

  1. Weekly window. One participant holds the main page per seven-day window for its presentation and audience blind testing.
  2. First in, first out. Priority follows the timestamp of a complete application that meets the published requirements. The organizer keeps the official schedule.
  3. No consecutive windows. While others are waiting, the current participant cannot take the next slot. Returning in later cycles is not limited.
  4. Changes. A participant tells the organizer promptly about material schedule changes. The organizer may reschedule, offer the slot to the next applicant or record it as unused, and logs the change.

Article III. Broadcast Autonomy and Repertoire

  1. Participant-controlled broadcasts. Participants choose when and how often to perform within their window and publish enough scheduling information for audiences and reviewers to find the broadcast.
  2. Repertoire freedom. The League sets no universal limit on the number, length or difficulty of performances and no human-to-robot ratio. The Turing Score must be read with the actual set of performances behind it.
  3. Pairwise cross-assignment. Where compulsory repertoire is used, public-domain pieces are assigned by pairs:
    • Participants 1 and 2 choose each other’s pieces, participants 3 and 4 likewise, and so on.
    • An unpaired participant plays the same compulsory repertoire as the previous participant.
    • With only one participant, the organizer assigns the repertoire.
    • Both sides submit their choices sealed, and the organizer publishes them together, so neither side sees the other’s choice first [proposed].
    • Every qualifying robot performance earns equal credit whatever the piece (the goal-in-football principle).
  4. Assignments are recorded. The organizer records assignments and passes them on before the performance. A participant who cannot perform a piece says so and gives the reason. Substitutions and non-completions are recorded, not silently counted as completed.

Article IV. Blind Testing and Voting Integrity

  1. Anonymous performance. Official performances are behind a physical or virtual curtain until they end. Nothing shown to the audience may identify the performer as human or robot before the reveal.
  2. Voting window. Voting is open only during the active performance and closes when the piece ends. The final judgments are locked before the reveal. Later votes are not counted.
  3. Final judgment. Each eligible voter has one final binary judgment (Human or Bot) per performance. Earlier clicks may be kept as behavioral records but do not count as votes.
  4. Identifiers and abuse controls. The system documents how voters are identified. IP addresses may be used for deduplication and anomaly checks but do not map one-to-one to people. The organizer may apply reasonable rate limits and abuse controls, and records any material change.
  5. Sequence. Anonymous performance, collect judgments, close voting, lock and calculate, verification step (Article IV-A), reveal, record the result. If a technical failure breaks this sequence, the performance is marked as affected and handled under Article VIII.

IV-A. Verification steps. The organizer may require participants to complete verification steps during or around a performance. The steps are communicated to registered participants in the participation documents and may be updated between seasons. A performance that does not complete a required verification step is marked unverified under Article VIII.

Article V. On-Site Technical Integrity and Independent Review

  1. Purpose. Review supports confidence that a robot performance was produced by the declared system, not by hidden remote operation, prerecorded audio or another undeclared method. Its scope is limited to that.
  2. Reviewers. During a participant’s window, the next contender in the queue may send up to two delegates, or the organizer may name an independent technical unit. Access, scope and reviewer identity are agreed beforehand and recorded.
  3. Boundaries. Reviewers may inspect equipment, control enclosures, wired controllers and media paths as far as needed to verify the declared setup. Source code, proprietary algorithms and unrelated confidential material are out of scope. Reviewers sign a confidentiality agreement and respect venue safety rules.
  4. Findings. Reports separate an observed fact, a suspected anomaly and a verified violation.
  5. Venue security. The host may restrict phones, wireless devices, photography and network access in audit areas, applied equally to everyone under the same conditions and announced in advance. The host may allow exceptions.
  6. Status record. The organizer records whether a performance was reviewed and any limitation found. A review is not a certification of a robot’s overall capability.
  7. Proportionate audit. The absence of an advanced audit tool does not stop the League. A performance whose integrity cannot be established under the current rules may be marked unverified, excluded from the affected result or referred under Article VIII, with the reason recorded.

V-A. Verification levels. Each official performance is published with a verification level, and the organizer states the level next to every result. Level 0: self-declared by the participant. Level 1: passed the organizer’s standard verification steps. Level 2: additionally reviewed on site under this Article. The methods behind each level are described to registered participants and may be strengthened over time. A verification result describes only what was checked; it is not a guarantee that no misconduct occurred.

Article VI. Official Turing Score

  1. Purpose. The score shows how often robot performances receive a majority audience judgment of Human, relative to all qualifying performances in the scoring period. It is not a measure of robotic intelligence, artistic quality, safety or general technical maturity.
  2. Reading the score. Because human performances sit in the denominator, the score reflects both robot outcomes and the participant’s overall mix. Compare participants alongside their published records, not as a pure robot imitation rate.
  3. Records. For each period the organizer keeps the counts of human performances, robot performances and points earned, plus any exclusions or incidents, so the published score can be reproduced.
  4. Corrections. A participant may ask for review of a calculation or a recorded incident within the published period. The organizer documents any correction and explains it publicly if it materially changes a result. Correcting a recorded fact is not a change to the formula.

Article VII. Technical Data and Perception Research

  1. Objective. Beyond scores, the League may align authorized technical records (for example joint position, velocity, torque, contact, timing) with audience judgments to study what drives perceived mechanical or human-like playing. Fields differ across platforms and are not treated as interchangeable without a technical basis.
  2. Quality and provenance. Submitted data should identify the performance, the declared system, the time base, units and sampling rate, the source and submission time, missing fields and known limits. The organizer may record file hashes and timestamps; these help detect later changes but do not prove authenticity.
  3. Status. Any model or correlation analysis is described according to its real stage and is not presented as validated or universal before the work is done. Participation guarantees no particular result.
  4. Permissions. Before submitting data, participants receive a clear description of collection, retention, analysis, sharing and publication, and give the needed authorization. Data ownership and exchange work as follows.
    • (a) Each participant keeps ownership of its raw technical logs.
    • (b) The participant authorizes the organizer to use those logs for integrity audits, such as checking for hidden remote operation or replay, and for the perception research described above, as stated on the participation form.
    • (c) The organizer owns the audience voting data and provides it to each participant for that participant’s own performances, for the participant’s research and development; it is not given to other participants, and it is provided in aggregated or pseudonymous form without raw network identifiers.
    • (d) These permissions are an exchange: the participant shares logs for audit, and the organizer shares voting data. Public results, research-approved data and confidential records are kept apart. Model training, commercial analysis, release or transfer needs explicit coverage in the data terms.
  5. Reports. Because platforms log different fields in different formats, the organizer does not promise a combined analysis. Where validated analysis exists, participants who shared complete logs may receive a summary before any public release; otherwise participants can run their own analysis on the voting data they receive. The organizer does not promise a report by a given date.

VII-A. Logs and official scores. Providing the required technical logs is a condition of an official score. The participation form lists the two purposes separately (integrity audit and perception research) together with the data terms, and the participant ticks each knowingly. A performance without the required logs may still be shown, but it is marked unverified and does not count toward an official result. How and when logs are delivered and checked is set out in the participation documents.

Article VIII. Incidents and Disputes

  1. Reportable incidents. Stream interruption that prevents reliable evaluation, voting failure, identity reveal before voting closes, mismatch between the declared and actual setup, a missing or corrupted record that affects the score, and material rule breaches.
  2. Classification. Operational incident (a problem without evidence of misconduct), unverified performance (not enough evidence to confirm compliance), verified rule violation (evidence supports a breach). These are not interchangeable.
  3. Remedies. Restart, reschedule, exclude the affected result, mark as unverified, or another remedy the published rules allow. The decision is proportionate and recorded with its reason and effect.
  4. No automatic inference. A failure, gap or unexplained anomaly does not by itself establish misconduct.
  5. Review requests. Participants may request review through the official channel by the published deadline. Where practical, someone with a direct conflict of interest is not the sole reviewer.

Article IX. Recognition, Violinist.ai Pages and Audience Incentives

  1. Season results. The organizer determines standings with the published formula. Season schedule, scoring period, treatment of invalid results and tie handling are published before standings are final.
  2. Page content. The team supplies the materials for its page, and nothing is published until the team has approved it. A page is a showcase, not a certification of technical maturity, safety or superiority, and the organizer will not describe it as one. Display location, duration, branding and removal terms are published in writing before the official season. All of this is a plan, not a promise.
  3. Audience incentives. The organizer promises no money to the audience. A participant may offer incentives under published rules on eligibility, allocation and delivery. Email addresses or other contact details are requested only from people who choose to claim an incentive and are handled under the privacy notice. Incentives must not change the score formula or let the host change results.
  4. Sponsors. Funding, sponsorship or promotion gives no right to change scores, hide valid results or direct a technical review. Any material conflict of interest is disclosed and handled through review.

Article X. Public Records, Privacy and Participation Terms

  1. Public records. The organizer may publish schedules, results, score calculations, declared system information, verification status and research summaries, and labels each statement as a recorded fact, a participant declaration, an audit finding or a research interpretation.
  2. Logs. [To confirm before launch, keep only what is true:] The system records each voter’s choice and timestamp. It does not ask for names. Network identifiers are used only for rate limiting and duplicate prevention, stored hashed, and raw values are deleted after 30 days (or when an open review request closes, if later). A fuller privacy notice is published before the official season, and anonymity is not promised beyond what the system supports.
  3. Participant acceptance. Before an official slot, the participant receives the current rules and the participation and data terms, and states its intent to take part by submitting the form at the bottom of the rules page; the organizer records the version, time and submission. Connecting technically is evidence of interaction only, not consent to every data use. The form states that schedule or performance failures by the participant are its own responsibility, so each team can judge its own risk. Submitting the form shows intent to take part and does not by itself guarantee a slot.
  4. Pre-match check. At least 48 hours before the first live session, the participant completes a technical test of the stream and the voting channel. If it is not completed, the organizer notifies the participant and may withhold or reassign the slot, recording the remedy.
  5. Audience notice. Before voting, audiences see a short notice that the broadcast is run by the participant, that network conditions may affect availability, that votes count only during the active window, and what data is collected. The voting form under the embedded page carries a checkbox labeled CONFIRMED, by which the voter acknowledges this notice and the rules, including that a participant may cancel or fail to appear and that the organizer does not control the participant’s broadcast. The checkbox does not waive any right that cannot lawfully be waived.

Article XI. Operational Disclaimer

  1. Service. The website, voting interface and data services are provided as available during the trial. The organizer makes reasonable efforts to keep official records and report material interruptions but does not guarantee uninterrupted service.
  2. Participant infrastructure. Participants are responsible for their own venue, equipment, content, local safety and third-party streaming services. How an affected performance is treated stays subject to Article VIII.
  3. Legal rights. Nothing here limits any right or liability that cannot lawfully be limited. Additional terms are stated in the participation or service terms.

Article XII. Amendments

  1. The current version, effective date and version number are published. Material changes to scoring, eligibility, voting, data use, awards or review are announced before they apply to a new scoring period and are not applied retroactively to completed performances, except for urgent security or legal reasons.
  2. A version history of material changes is kept.

Closing Principle

The competition produces observations. The records preserve evidence. The accumulated evidence may help explain what the score alone cannot. The League does not need every advanced capability before it begins, but it states the rules it applies, keeps the records that explain its results, and separates verified findings from future research goals.

➡️ Frequently Asked Questions

Name