Issue 6

6. Cross-Rate Inconsistency

This notebook documents the analysis for Issue #6: Cross-Rate Inconsistency in the Autonity Oracle data. It covers:

• What is this issue about?
  
• Why conduct this issue analysis?
  
• How to conduct this issue analysis?
  
• What are the results?

---

6.1 What Is This Issue About?

In the Issues Table, Issue #6 describes Cross-Rate Inconsistency:
> “NTN-USD * ATN-USD ≠ NTN-ATN (when scaling properly).”

This implies that if NTN-ATN = X and ATN-USD = Y, expect NTN-USD ≈ X * Y. Significant mismatches indicate potential errors or data inconsistencies.

---

6.2 Why Conduct This Issue Analysis?

• Data Reliability: Ensuring consistent data across pairs is critical.
• Trust: Maintaining accuracy and confidence in the Oracle.
• Debugging: Identifying systemic or validator-specific issues.

---

6.3 How to Conduct the Analysis?

Use Python with the Polars library (v1.24.0) to:

1. Load and preprocess Oracle submission CSV files.
2. Convert Wei-based prices to decimal.
3. Calculate NTN-USD_estimated = NTN-ATN * ATN-USD.
4. Flag rows exceeding a threshold (10%).
5. Summarize by date and validator.

Below is the Python code for the analysis:

import polars as pl
import glob
import math
from typing import Optional
import warnings

warnings.filterwarnings("ignore")

def load_and_preprocess_submissions(submission_glob: str) -> pl.DataFrame:
    """
    Loads Oracle Submission CSVs and returns a Polars DataFrame.
    """
    files = sorted(glob.glob(submission_glob))
    if not files:
        raise ValueError(f"No CSV files found matching pattern {submission_glob}")

    lf_list = []
    for f in files:
        lf_temp = pl.scan_csv(
            f,
            dtypes={"Timestamp": pl.Utf8},
            null_values=[""],
            ignore_errors=True,
        )
        lf_list.append(lf_temp)

    lf = pl.concat(lf_list)

    lf = lf.with_columns(
        pl.col("Timestamp")
        .str.strptime(pl.Datetime, strict=False)
        .alias("Timestamp_dt")
    )

    lf = lf.with_columns(
        [
            pl.col("Timestamp_dt").cast(pl.Date).alias("date_only"),
            pl.col("Timestamp_dt")
            .dt.weekday()
            .alias("weekday_num"),
        ]
    )

    df = lf.collect()
    return df


def convert_wei_to_decimal(price_wei: Optional[float]) -> Optional[float]:
    """
    Converts a Wei-based price to a normal float decimal.
    """
    if price_wei is None or math.isnan(price_wei):
        return None
    return price_wei / 1e18


def check_cross_rate_inconsistency(
    df: pl.DataFrame,
    atn_usd_col: str = "ATN-USD Price",
    ntn_usd_col: str = "NTN-USD Price",
    ntn_atn_col: str = "NTN-ATN Price",
    threshold: float = 0.01,
) -> pl.DataFrame:
    """
    Computes cross-rate mismatch and measures relative differences.
    """
    df_local = df.clone()
    decimal_cols = []
    for col in [atn_usd_col, ntn_usd_col, ntn_atn_col]:
        col_decimal = col + " Decimal"
        df_local = df_local.with_columns(
            (pl.col(col).cast(pl.Float64) / 1e18).alias(col_decimal)
        )
        decimal_cols.append(col_decimal)

    atn_usd_dec = atn_usd_col + " Decimal"
    ntn_usd_dec = ntn_usd_col + " Decimal"
    ntn_atn_dec = ntn_atn_col + " Decimal"

    df_local = df_local.with_columns(
        [(pl.col(ntn_atn_dec) * pl.col(atn_usd_dec)).alias("ntn_usd_estimated")]
    )

    epsilon = 1e-18
    df_local = df_local.with_columns(
        [
            (
                (pl.col("ntn_usd_estimated") - pl.col(ntn_usd_dec)).abs()
                / (pl.col(ntn_usd_dec).abs() + epsilon)
            ).alias("rel_diff_cross")
        ]
    )

    df_local = df_local.with_columns(
        [
            pl.when(pl.col("rel_diff_cross") > threshold)
            .then(pl.lit(f"Cross-rate mismatch > {int(threshold*100)}%"))
            .otherwise(pl.lit(""))
            .alias("suspect_reason")
        ]
    )

    df_flagged = df_local.filter(pl.col("suspect_reason") != "")

    keep_cols = [
        "Timestamp_dt",
        "Validator Address",
        atn_usd_dec,
        ntn_usd_dec,
        ntn_atn_dec,
        "ntn_usd_estimated",
        "rel_diff_cross",
        "suspect_reason",
    ]

    optional_cols = []
    for c in ["date_only", "weekday_num"]:
        if c in df_flagged.columns:
            optional_cols.append(c)

    df_flagged = df_flagged.select(keep_cols + optional_cols)
    return df_flagged


def summarize_cross_rate_inconsistency(df_flagged: pl.DataFrame) -> pl.DataFrame:
    """
    Summarizes the number of cross-rate mismatches.
    """
    if df_flagged.is_empty():
        return pl.DataFrame(
            {
                "date_only": [],
                "Validator Address": [],
                "mismatch_count": [],
                "avg_rel_diff": [],
                "max_rel_diff": [],
            }
        )

    lf = df_flagged.lazy()
    summary_lf = (
        lf.group_by(["date_only", "Validator Address"])
        .agg(
            [
                pl.count("rel_diff_cross").alias("mismatch_count"),
                pl.mean("rel_diff_cross").alias("avg_rel_diff"),
                pl.max("rel_diff_cross").alias("max_rel_diff"),
            ]
        )
        .sort(["date_only", "Validator Address"])
    )
    return summary_lf.collect()


def analyze_cross_rate_inconsistency(
    submission_glob: str,
    atn_usd_col: str = "ATN-USD Price",
    ntn_usd_col: str = "NTN-USD Price",
    ntn_atn_col: str = "NTN-ATN Price",
    threshold: float = 0.01,
):
    """
    Main analysis function.
    """
    df_all = load_and_preprocess_submissions(submission_glob)

    df_flagged = check_cross_rate_inconsistency(
        df_all,
        atn_usd_col=atn_usd_col,
        ntn_usd_col=ntn_usd_col,
        ntn_atn_col=ntn_atn_col,
        threshold=threshold,
    )

    df_summary = summarize_cross_rate_inconsistency(df_flagged)

    return {
        "df_flagged": df_flagged,
        "df_summary": df_summary,
    }

results = analyze_cross_rate_inconsistency(
    submission_glob="../submission-data/Oracle_Submission_*.csv",
    atn_usd_col="ATN-USD Price",
    ntn_usd_col="NTN-USD Price",
    ntn_atn_col="NTN-ATN Price",
    threshold=0.01,  # 1% mismatch threshold
)

---

6.4 What are the resuults?

Below presents the results from the analysis:

Flagged Submissions (Cross-rate mismatch > 10%)

df_flagged = results["df_flagged"]
if df_flagged.is_empty():
    print("No cross-rate inconsistencies found beyond 10% threshold.")
else:
    print(f"Total flagged cross-rate mismatches: {df_flagged.height}")
    df_flagged

No cross-rate inconsistencies found beyond 10% threshold.

• Timestamp_dt and Validator Address indicate problematic submissions.
• Prices (ATN-USD, NTN-USD, NTN-ATN) and estimated NTN-USD.
• rel_diff_cross shows percentage mismatch.

Daily Validator Summary

df_summary = results["df_summary"]
if df_summary.is_empty():
    print("No daily cross-rate mismatches to summarize.")
else:
    print(f"Number of daily mismatch records: {df_summary.height}")
    df_summary

No daily cross-rate mismatches to summarize.

• mismatch_count: Number of daily flagged submissions per validator.
• avg_rel_diff and max_rel_diff: Mean and maximum relative mismatch percentages.

---

Combine both findings

total_flagged = df_flagged.height
if total_flagged == 0:
    print("Data consistency is good. No critical cross-rate mismatches identified.")
else:
    num_validators = df_summary.get_column("Validator Address").n_unique()
    print(f"{total_flagged} total mismatches across {num_validators} validators found. Investigate further:")
    high_impact = df_summary.filter(pl.col("max_rel_diff") > 0.5)
    if high_impact.is_empty():
        print("No extreme mismatches (>50%) identified. Likely minor synchronization issues or rounding errors.")
    else:
        print(f"Validators with extreme mismatches (>50%): {high_impact.height}")
        high_impact

Data consistency is good. No critical cross-rate mismatches identified.

• If no mismatches, system appears healthy.
• Small mismatches (<50%) often reflect minor timing or rounding discrepancies.
• Extreme mismatches (>50%) require thorough validator-specific investigation:
  • Verify decimal conversions.
  • Confirm synchronized data updates.
  • Review validator configurations or potential malicious behavior.

List of all Validators and their Consistency Indices

df_all = load_and_preprocess_submissions("../submission-data/Oracle_Submission_*.csv")

total_stats = (
    df_all.lazy()
    .group_by("Validator Address")
    .agg(pl.count().alias("total_submissions"))
    .collect()
)

flagged_stats = (
    df_flagged.lazy()
    .group_by("Validator Address")
    .agg(
        pl.count().alias("flagged_count"),
        pl.mean("rel_diff_cross").alias("mean_rel_diff"),
        pl.max("rel_diff_cross").alias("max_rel_diff"),
    )
    .collect()
)

validator_stats = (
    total_stats
    .join(flagged_stats, on="Validator Address", how="left")
    .fill_null(0)
)

validator_stats = (
    validator_stats
    .with_columns(
        (
            (pl.col("flagged_count") / pl.col("total_submissions")) *
            (pl.col("mean_rel_diff") + pl.col("max_rel_diff"))
        ).alias("mismatch_penalty_rate")
    )
    .with_columns(
        (1 / (1 + pl.col("mismatch_penalty_rate")))
        .alias("consistency_index")
    )
    .sort("consistency_index", descending=False)
)

print("Validator Consistency Ranking (higher index = better)\n")
for row in validator_stats.to_dicts():
    print(
        f"Validator {row['Validator Address']}: "
        f"subs={row['total_submissions']:5d}, "
        f"flags={row['flagged_count']:3d}, "
        f"index={row['consistency_index']:.4f}"
    )

Validator Consistency Ranking (higher index = better)

Validator 0x383A3c437d3F12f60E5fC990119468D3561EfBfc: subs= 2880, flags=  0, index=1.0000
Validator 0xDCA5DFF3D42f2db3C18dBE823380A0A81db49A7E: subs= 2880, flags=  0, index=1.0000
Validator 0xcf716b3930d7cf6f2ADAD90A27c39fDc9D643BBd: subs= 2880, flags=  0, index=1.0000
Validator 0xbfDcAF35f52F9ef423ac8F2621F9eef8be6dEd17: subs= 2833, flags=  0, index=1.0000
Validator 0x1Be7f70BCf8393a7e4A5BcC66F6f15d6e35cfBBC: subs= 2880, flags=  0, index=1.0000
Validator 0x64F83c2538A646A550Ad9bEEb63427a377359DEE: subs= 2880, flags=  0, index=1.0000
Validator 0xB5d8be2AB4b6d7E6be7Ea28E91b370223a06289f: subs= 2880, flags=  0, index=1.0000
Validator 0x5603caFE3313D0cf56Fd4bE4A2f606dD6E43F8Eb: subs= 2880, flags=  0, index=1.0000
Validator 0xdF239e0D5b4E6e820B0cFEF6972A90893c2073AB: subs= 2880, flags=  0, index=1.0000
Validator 0xBE287C82A786218E008FF97320b08244BE4A282c: subs= 2879, flags=  0, index=1.0000
Validator 0x5E17e837DcBa2728C94f95c38fA8a47CB9C8818F: subs= 2876, flags=  0, index=1.0000
Validator 0x551f3300FCFE0e392178b3542c009948008B2a9F: subs= 2880, flags=  0, index=1.0000
Validator 0xE4686A4C6E63A8ab51B458c52EB779AEcf0B74f7: subs= 2880, flags=  0, index=1.0000
Validator 0x100E38f7BCEc53937BDd79ADE46F34362470577B: subs= 2876, flags=  0, index=1.0000
Validator 0xF9B38D02959379d43C764064dE201324d5e12931: subs= 2880, flags=  0, index=1.0000
Validator 0x3fe573552E14a0FC11Da25E43Fef11e16a785068: subs= 2880, flags=  0, index=1.0000
Validator 0x718361fc3637199F24a2437331677D6B89a40519: subs= 2880, flags=  0, index=1.0000
Validator 0x358488a4EdCA493FCD87610dcd50c62c8A3Dd658: subs= 2880, flags=  0, index=1.0000
Validator 0x527192F3D2408C84087607b7feE1d0f907821E17: subs= 2880, flags=  0, index=1.0000
Validator 0x9C7dAABb5101623340C925CFD6fF74088ff5672e: subs= 2880, flags=  0, index=1.0000
Validator 0x6a395dE946c0493157404E2b1947493c633f569E: subs= 2874, flags=  0, index=1.0000
Validator 0x2928FE5b911BCAf837cAd93eB9626E86a189f1dd: subs= 2829, flags=  0, index=1.0000
Validator 0x9d28e40E9Ec4789f9A0D17e421F76D8D0868EA44: subs= 2880, flags=  0, index=1.0000
Validator 0x01F788E4371a70D579C178Ea7F48E04e8B2CD743: subs= 2837, flags=  0, index=1.0000
Validator 0xDF2D0052ea56A860443039619f6DAe4434bc0Ac4: subs= 2879, flags=  0, index=1.0000
Validator 0x8f91e0ADF8065C3fFF92297267E02DF32C2978FF: subs= 2879, flags=  0, index=1.0000
Validator 0x3597d2D42f8Fbbc82E8b1046048773aD6DDB717E: subs= 2880, flags=  0, index=1.0000
Validator 0x7232e75a8bFd8c9ab002BB3A00eAa885BC72A6dd: subs= 2877, flags=  0, index=1.0000
Validator 0x8584A78A9b94f332A34BBf24D2AF83367Da31894: subs= 2879, flags=  0, index=1.0000
Validator 0xc5B9d978715F081E226cb28bADB7Ba4cde5f9775: subs= 2879, flags=  0, index=1.0000
Validator 0xd625d50B0d087861c286d726eC51Cf4Bd9c54357: subs= 2856, flags=  0, index=1.0000
Validator 0x00a96aaED75015Bb44cED878D927dcb15ec1FF54: subs= 2866, flags=  0, index=1.0000
Validator 0x197B2c44b887c4aC01243BDE7E4bBa8bd95BC3a8: subs= 2880, flags=  0, index=1.0000
Validator 0x24915749B793375a8C93090AF19928aFF1CAEcb6: subs= 2880, flags=  0, index=1.0000
Validator 0xcdEed21b471b0Dc54faF74480A0E700fCc42a7b6: subs= 2880, flags=  0, index=1.0000
Validator 0x19E356ebC20283fc74AF0BA4C179502A1F62fA7B: subs= 2833, flags=  0, index=1.0000
Validator 0x831B837C3DA1B6c2AB68a690206bDfF368877E19: subs= 2863, flags=  0, index=1.0000
Validator 0x36142A4f36974e2935192A1111C39330aA296D3C: subs= 2880, flags=  0, index=1.0000
Validator 0xfD97FB8835d25740A2Da27c69762D74F6A931858: subs= 2880, flags=  0, index=1.0000
Validator 0x94d28f08Ff81A80f4716C0a8EfC6CAC2Ec74d09E: subs= 2880, flags=  0, index=1.0000
Validator 0xD9fDab408dF7Ae751691BeC2efE3b713ba3f9C36: subs= 2880, flags=  0, index=1.0000
Validator 0xE9FFF86CAdC3136b3D94948B8Fd23631EDaa2dE3: subs= 2880, flags=  0, index=1.0000
Validator 0xBBf36374eb23968F25aecAEbb97BF3118f3c2fEC: subs= 2880, flags=  0, index=1.0000
Validator 0xf34CD6c09a59d7D3d1a6C3dC231a7834E5615D6A: subs= 2834, flags=  0, index=1.0000
Validator 0x3AaF7817618728ffEF81898E11A3171C33faAE41: subs= 2874, flags=  0, index=1.0000
Validator 0x94470A842Ea4f44e668EB9C2AB81367b6Ce01772: subs= 2880, flags=  0, index=1.0000
Validator 0x4cD134001EEF0843B9c69Ba9569d11fDcF4bd495: subs= 2823, flags=  0, index=1.0000
Validator 0x984A46Ec685Bb41A7BBb2bc39f80C78410ff4057: subs= 2877, flags=  0, index=1.0000
Validator 0x22A76e194A49c9e5508Cd4A3E1cD555D088ECB08: subs= 2880, flags=  0, index=1.0000
Validator 0x26E2724dBD14Fbd52be430B97043AA4c83F05852: subs= 2880, flags=  0, index=1.0000
Validator 0x791A7F840ac11841cCB0FaA968B2e3a0Db930fCe: subs= 2880, flags=  0, index=1.0000
Validator 0x6747c02DE7eb2099265e55715Ba2E03e8563D051: subs= 2840, flags=  0, index=1.0000
Validator 0xEf0Ba5e345C2C3937df5667A870Aae5105CAa3a5: subs= 2879, flags=  0, index=1.0000
Validator 0xC1F9acAF1824F6C906b35A0D2584D6E25077C7f5: subs= 2880, flags=  0, index=1.0000
Validator 0x1476A65D7B5739dE1805d5130441A94022Ee49fe: subs= 2462, flags=  0, index=1.0000
Validator 0xf10f56Bf0A28E0737c7e6bB0aF92f3DDad34aE6a: subs= 2880, flags=  0, index=1.0000
Validator 0x99E2B4B27BDe92b42D04B6CF302cF564D2C13b74: subs= 2880, flags=  0, index=1.0000
Validator 0xd61a48b0e11B0Dc6b7Bd713B1012563c52591BAA: subs= 2873, flags=  0, index=1.0000
Validator 0x59031767f20EA8F4a3d90d33aB0DAA2ca469Fd9a: subs= 2880, flags=  0, index=1.0000
Validator 0x23b4Be9536F93b8D550214912fD0e38417Ff7209: subs= 2880, flags=  0, index=1.0000

Please note, total_submissions represents the total number of submissions for this validator. flagged_count is how many of those were caught by the 1 % cross-rate check. consistency_index is a 0-to-1 reliability score (1 = no mismatches; lower values indicate more frequent or larger inconsistencies).