[Exchange] Prediction market quality gate — audit 120-in-24h burst and enforce signal standards

← All Specs

Context

As of 2026-04-29, SciDEX has 139 prediction markets. In the last 24h:

  • 120 new markets were created (from the experiment-to-prediction-market bridge)
  • 23 markets were resolved

The experiment-to-prediction-market bridge (task 7a2a51d6, merged 2026-04-28)
works — but creating 120 markets in 24h without quality gates produces noise.
Markets should represent genuine scientific uncertainty with tradeable signal.
At that creation rate, most markets likely have:
  • Boilerplate yes/no probabilities (0.5/0.5)
  • No distinguishing resolution criteria
  • Low liquidity (no meaningful trading)
  • Duplicate or near-identical propositions

The platform's prediction-market credibility depends on market quality, not
volume. A market that resolves to 50/50 with no trades provides zero signal.

Goal

  • Audit all 120 newly-created markets for quality (resolution criteria,
  • probability calibration, uniqueness)
  • Identify and archive the lowest-quality markets
  • Define a quality gate standard for future market creation
  • Implement the quality gate in the experiment bridge to enforce it
  • Ensure the 23 resolved markets resolved correctly with evidence
  • What success looks like

    ☐ All 120 new markets audited with quality scores
    ☐ Bottom 30% (≤ 30-40 markets) archived with rationale
    ☐ Quality gate implemented: market creation requires resolution_criteria,
    probability justification, and link to hypothesis or experiment
    POST /api/markets enforces quality gate before insert
    ☐ 23 resolved markets verified: resolution evidence documented

    What NOT to do

    • Do NOT delete markets — archive them (status='archived' or similar)
    • Do NOT lower all probabilities toward 50/50 — some genuine asymmetry
    is valid (e.g., "APOE4 increases AD risk" should be 90/10 already)
    • Do NOT break the existing experiment bridge — patch it, don't rewrite it

    Agent guidance

  • Read prediction_markets schema: hypothesis_id, yes_prob, no_prob,
  • liquidity, resolution_date, resolved_at, resolution, experiment_id
  • Quality criteria: (1) resolution_date set, (2) liquidity > 0,
  • (3) resolution criteria in title or linked hypothesis description,
    (4) yes_prob not exactly 0.5 (uninformative priors)
  • Use LLM to assess market proposition quality from title + hypothesis link
  • Gate implementation: add validation in the market creation route in api.py
  • Spec notes

    • Created by quest task generator Cycle 4 (2026-04-29T03:05Z)
    • Priority 91: protects Exchange value proposition (criterion #2) and
    implements feedback loop (criterion #4) for market signal quality
    • One-shot: quality gate + audit is bounded work; the gate prevents future
    burst creation

    Work Log

    Created 2026-04-29

    Spec created by ambitious quest task generator (Cycle 4). Discovery: 120
    markets created in 24h from experiment bridge — rapid creation without quality
    gates is noise amplification. Platform credibility requires market signal
    integrity. This task enforces it.

    File: quest_exchange_market_quality_gate.md
    Modified: 2026-05-01 20:13
    Size: 3.2 KB