Oracle Submission Analysis - Summary of Key Findings (December 2024)

1 Executive Summary

This document provides a straightforward summary of the key findings from an analysis of Autonity Oracle submissions data from December 2024. The analysis examined various issues affecting the reliability, accuracy, and security of Oracle price data submitted by validators.

1.1 Overview of Issues Analyzed

The investigation covered ten distinct issue areas:

  1. Missing or Null Submissions: Examining validators that failed to submit price data
  2. Irregular Submission Frequency: Analyzing abnormal timing patterns in submissions
  3. Out-of-Range Values: Detecting suspicious price values compared to benchmarks
  4. Stale/Lagging Data: Identifying validators that fail to update prices when markets move
  5. Confidence Value Anomalies: Examining issues with confidence metrics
  6. Cross-Rate Inconsistency: Assessing mathematical consistency across token prices
  7. Timing/Synchronization Issues: Analyzing timestamp disparities between validators
  8. Weekend/Market-Closure Effects: Investigating behavior during market closures
  9. Vendor Downtime: Detecting submission stoppages
  10. Security/Malicious Behavior: Looking for potential manipulation patterns

2 Key Findings

2.1 Missing or Null Submissions

  • Four validators had 100% missing-submission rates:
    • 0x100E38f7BCEc53937BDd79ADE46F34362470577B
    • 0xd625d50B0d087861c286d726eC51Cf4Bd9c54357
    • 0x26E2724dBD14Fbd52be430B97043AA4c83F05852
    • 0x3fe573552E14a0FC11Da25E43Fef11e16a785068
  • 40,317 distinct submission timestamps: on an average day 91.5% of those timestamps were fully covered (i.e., ≥ 90% of active validators supplied valid data for every pair)
  • 95.5% weekend vs 89.1% weekday coverages: indicating consistent behavior on weekdays and weekends
  • 8.5% lacked complete data from the validator set across all timestamps

2.2 Irregular Submission Frequency

  • Expected cadence: 1 submission every 30s (2,880 per day)
  • Validator 0x04d00379d1531e06F37782A65D30237A2F3885ac produced the worst timing regularity – 61.4% of its 2,368 inter-submission intervals fell outside a ±5% tolerance band (0.475–0.525 min)
  • ≥ 40 validators kept < 1% of their intervals outside tolerance
  • No validator exceeded the historical burst threshold of 15 submissions per minute
  • Intervals within a ±5% tolerance band around the 30-second target (0.475–0.525 min) were considered on-schedule for calculations

2.3 Out-of-Range Values

  • 1,454 total suspicious submissions were detected
  • 0 rows contained non-positive (≤ 0) prices
  • 1,063 rows deviated more than 20% from benchmark FX feeds
  • 391 rows failed the Autonity cross-rate sanity test
  • Four validators each accounted for ~25% of the suspicious rows:
    • 0xDCA5DFF3D42f2db3C18dBE823380A0A81db49A7E – 364 rows (0.9% of its submissions)
    • 0x1Be7f70BCf8393a7e4A5BcC66F6f15d6e35cfBBC – 364 rows (0.9%)
    • 0x22A76e194A49c9e5508Cd4A3E1cD555D088ECB08 – 363 rows (0.9%)
    • 0x9d28e40E9Ec4789f9A0D17e421F76D8D0868EA44 – 363 rows (0.9%)
  • 60+ validators produced zero suspicious rows

2.4 Stale/Lagging Data

  • 31,955 stale-data runs (≥ 30 identical prices) were detected
  • 18 validators reached the maximum observed stale-score of 6.0 (identical prices in at least six columns on every timestamp)
  • 4 validators achieved a stale-score of 0.0, indicating fully dynamic pricing throughout December
  • The median stale-run length was 38 submissions
  • 15,234 runs lasted longer than one hour, underscoring the persistence of some frozen feeds

2.5 Confidence Value Anomalies

  • 9 validators submitted fixed confidence values across every pair and timestamp
  • 192 validator-pair combinations exhibited near-zero correlation (< 0.1) between price changes and confidence, showing the metric carried little information
  • Roughly 75% of validators displayed meaningful variance in confidence values following the December software upgrade

2.6 Cross-Rate Inconsistency

  • Only 9 daily mismatch records exceeded the 5% cross-rate tolerance
  • No validator accounted for more than 3 of those mismatches
  • The most frequent inconsistency was NTN-ATN × ATN-USD ≠ NTN-USD (3 days)

2.7 Timing/Synchronization Issues

  • Mean submission offsets (per validator, relative to the group median) ranged from -3.0s to +0.4s
  • The maximum individual offset observed was 15s; no validator breached the ±30s alert threshold
  • The median absolute offset across all validators was 7.46s
  • The most precise validator (0x7b06f608aB874E21f8FFC35D04B32bc03D8dCE1f) averaged 3.05s
  • The least precise validator (0xbfDcAF35f52F9ef423ac8F2621F9eef8be6dEd17) averaged 7.70s

2.8 Weekend/Market-Closure Effects

  • Full-coverage was 95.5% on weekends versus 89.1% on weekdays – a 6.4-percentage-point improvement when major FX markets were closed
  • Price variance, stale-run frequency, and timing offsets were all marginally lower on weekends

2.9 Vendor Downtime Issues

  • 64 stoppage events were recorded (one final submission record per validator plus several early stops)
  • No intra-month outage gap > 15 min was observed under the current detector settings
  • Zero “all-zero price” placeholders were found
  • No evidence of multi-validator concurrency beyond the natural month-end stoppage

2.10 Security/Malicious-Behavior Indicators

  • 2,176 validator-pairs (≥ 75% identical prices) were flagged for potential collusion
  • 8,476 timestamps featured simultaneous extreme outliers (> 2σ from the median) posted by at least three validators, concentrated around 18-19 December
  • The four validators most heavily implicated are:
    • 0x9d28e40E9Ec4789f9A0D17e421F76D8D0868EA44 – collusion_score 1,921,511; extreme_event_count 8,472
    • 0xDCA5DFF3D42f2db3C18dBE823380A0A81db49A7E – 2,077,207; 8,470
    • 0x1Be7f70BCf8393a7e4A5BcC66F6f15d6e35cfBBC – 2,075,152; 8,470
    • 0x22A76e194A49c9e5508Cd4A3E1cD555D088ECB08 – 1,130,038; 8,470
  • The amount of collusion/outlier evidence elevates the Security Concern Level to 🟡

3 Notable Validators

3.1 Highest Performing Validators

No validator satisfied every reliability, accuracy, and consistency requirement once collusion and stale-data metrics were incorporated.

3.2 Most Problematic Validators

Issue Validator Notes
100% missing submissions 0x100E38f7BCEc53937BDd79ADE46F34362470577B Appears in validator set but never submitted data
100% missing submissions 0xd625d50B0d087861c286d726eC51Cf4Bd9c54357 ”
100% missing submissions 0x26E2724dBD14Fbd52be430B97043AA4c83F05852 ”
100% missing submissions 0x3fe573552E14a0FC11Da25E43Fef11e16a785068 ”
61.4% irregular intervals 0x04d00379d1531e06F37782A65D30237A2F3885ac Worst cadence violation
Stale-score 6.0 0xA284470fa70D8A2A8402054e40A36077fEAdCF51 Identical prices in ≥ 6 columns per slot
High collusion & extreme outliers 0x9d28e40E9Ec4789f9A0D17e421F76D8D0868EA44 See security section
High collusion & extreme outliers 0xDCA5DFF3D42f2db3C18dBE823380A0A81db49A7E ”
High collusion & extreme outliers 0x1Be7f70BCf8393a7e4A5BcC66F6f15d6e35cfBBC ”
High collusion & extreme outliers 0x22A76e194A49c9e5508Cd4A3E1cD555D088ECB08 ”

3.3 Validators with Coordinated Behavior

The majority of collusion signals involve the four addresses listed above; no additional stable clusters large enough to warrant a separate grouping were detected under December’s thresholds.

4 Implications and Recommendations

4.1 Data Quality Concerns

  • Missing-slot and stale-data metrics improved relative to November, yet the four fully inactive validators drag aggregate coverage below the 95% target
  • Confidence metrics remain highly problematic for a minority of validators, undermining consumer trust

4.2 Validator Performance

  • Top quartile of active validators exhibited < 1% irregular intervals and 0 suspicious submissions
  • Bottom decile exceeded 25% problematic submissions, emphasizing the need for stronger incentives

4.3 Recommendations

  1. Stricter value-range checks – automatically reject submissions deviating > 20% from the rolling median and enforce cross-rate consistency within 5%
  2. Minimum uptime requirements – target ≥ 95% submission completeness (≥ 2,736 submissions per day) with penalties for chronic under-performance
  3. Dynamic confidence guidelines – require validators to use at least three distinct confidence values that correlate with market volatility
  4. Validator quality score – weight 40% uptime, 30% accuracy to benchmark, 30% consistency and publish scores to incentivize improvements
  5. Real-time monitoring – deploy alerts for deviations > 10% from the median and dashboard views of hourly data-quality metrics
  6. Focused reviews – prioritize investigation of the four validators exhibiting strong collusion/outlier patterns

5 Conclusion

While December 2024 showed good measurable statistics in coverage and irregular-frequency metrics, the persistence of fully inactive validators and evidence of coordinated outlier submissions highlight ongoing risks. Implementing the recommended validation checks, incentive mechanisms, and monitoring tools will substantially strengthen Oracle data quality and, by extension, the robustness of the Autonity ecosystem.

6 Monthly Comparison Table

Issue Area Rating Scale Description
Missing/Null Submissions 🟡 ⚫ Critical (> 60%) 🔴 Poor (30–60%) 🟠 Fair (15–30%) 🟡 Good (5–15%) 🟢 Excellent (< 5%)
Irregular Submission Frequency 🟢 ⚫ Critical (> 25% irregular) 🔴 Poor (15–25%) 🟠 Fair (8–15%) 🟡 Good (2–8%) 🟢 Excellent (< 2%)
Out-of-Range Values 🟢 ⚫ Critical (> 8%) 🔴 Poor (3–8%) 🟠 Fair (1–3%) 🟡 Good (0.3–1%) 🟢 Excellent (< 0.3%)
Stale/Lagging Data 🟠 ⚫ Critical (> 15% runs) 🔴 Poor (7–15%) 🟠 Fair (3–7%) 🟡 Good (0.5–3%) 🟢 Excellent (< 0.5%)
Confidence Value Anomalies 🔴 ⚫ Critical (> 85% fixed) 🔴 Poor (60–85%) 🟠 Fair (35–60%) 🟡 Good (15–35%) 🟢 Excellent (< 15%)
Cross-Rate Inconsistency 🟢 ⚫ Critical (> 12%) 🔴 Poor (6–12%) 🟠 Fair (3–6%) 🟡 Good (1–3%) 🟢 Excellent (< 1%)
Timing/Synchronization 🟢 ⚫ Critical (> 60s) 🔴 Poor (30–60s) 🟠 Fair (10–30s) 🟡 Good (3–10s) 🟢 Excellent (< 3s)
Weekend/Market-Closure Effects 🟢 ⚫ Critical (> 30%) 🔴 Poor (15–30%) 🟠 Fair (7–15%) 🟡 Good (2–7%) 🟢 Excellent (< 2%)
Vendor Downtime Impact 🟢 ⚫ Critical (> 10% time) 🔴 Poor (4–10%) 🟠 Fair (2–4%) 🟡 Good (0.5–2%) 🟢 Excellent (< 0.5%)
Security Concern Level 🟡 ⚫ Critical (confirmed) 🔴 Poor (strong evidence) 🟠 Fair (some evidence) 🟡 Good (minimal) 🟢 Excellent (none)
Overall Rating 🟢 ⚫ Critical 🔴 Poor 🟠 Fair 🟡 Good 🟢 Excellent