Depeg Duration Resolver Changelog: v5 archive
Historical release evidence. Return to current versions and the complete history index.
What this controls
Depeg Duration Resolver Changelog: v5 archive version history
Use this changelog to trace how Pharos changed the related scoring, data-source, threshold, or interpretation rules over time. Entries record the contract active at each release; archived rules are retained so historical charts and citations can be read against the rules active at the time.
Primary references
Latest version
v5.0Oct 25, 2026
Evidence-class verdicts, competing-risk duration odds and reviewer v5
DDR v5 condemns a coin only on terminal evidence, publishes duration as unconditional recovery odds that keep non-recovery visible, and grades its calls against naive baselines.
Impact snapshot
- Verdicts read as evidence classes: No break signals, Stress signals and Terminal evidence. Wire tier codes are unchanged; rows locked before v5 keep their original Recovery Likely / At Risk / Recovery Unlikely labels.
- Static exit thinness no longer condemns a coin. V9 exit capacity is re-scaled from the Safety Score stress request (a $1M floor) to the dislocation's value gap, and it, no-viable-route and thin redemption capacity cap at elevated. K5 is severe only when DEX TVL collapses at least 40% from the event's onset baseline on a thin market.
- Two or more elevated kill signals without a strong structural anchor are terminal evidence only on a severe below-peg break (deeper than 1,000 bps, using the worse of the stored peak and live deviation). A deep break with no kill signal reads as Stress signals.
- Duration odds estimate the share of comparable incidents that recover, with terminal incidents as a competing outcome and still-open or non-recovery closures censored at their last observation (coin-weighted Aalen–Johansen). Quantiles the curve never reaches stay empty and the share not recovered within 30 days is published.
v5.0Oct 25, 2026Evidence-class verdicts, competing-risk duration odds and reviewer v5
DDR v5 condemns a coin only on terminal evidence, publishes duration as unconditional recovery odds that keep non-recovery visible, and grades its calls against naive baselines.
- Verdicts read as evidence classes: No break signals, Stress signals and Terminal evidence. Wire tier codes are unchanged; rows locked before v5 keep their original Recovery Likely / At Risk / Recovery Unlikely labels.
- Static exit thinness no longer condemns a coin. V9 exit capacity is re-scaled from the Safety Score stress request (a $1M floor) to the dislocation's value gap, and it, no-viable-route and thin redemption capacity cap at elevated. K5 is severe only when DEX TVL collapses at least 40% from the event's onset baseline on a thin market.
- Two or more elevated kill signals without a strong structural anchor are terminal evidence only on a severe below-peg break (deeper than 1,000 bps, using the worse of the stored peak and live deviation). A deep break with no kill signal reads as Stress signals.
- Duration odds estimate the share of comparable incidents that recover, with terminal incidents as a competing outcome and still-open or non-recovery closures censored at their last observation (coin-weighted Aalen–Johansen). Quantiles the curve never reaches stay empty and the share not recovered within 30 days is published.
DetailsHide details
v5.0Oct 25, 2026
Evidence-class verdicts, competing-risk duration odds and reviewer v5
DDR v5 condemns a coin only on terminal evidence, publishes duration as unconditional recovery odds that keep non-recovery visible, and grades its calls against naive baselines.
- Verdicts read as evidence classes: No break signals, Stress signals and Terminal evidence. Wire tier codes are unchanged; rows locked before v5 keep their original Recovery Likely / At Risk / Recovery Unlikely labels.
- Static exit thinness no longer condemns a coin. V9 exit capacity is re-scaled from the Safety Score stress request (a $1M floor) to the dislocation's value gap, and it, no-viable-route and thin redemption capacity cap at elevated. K5 is severe only when DEX TVL collapses at least 40% from the event's onset baseline on a thin market.
- Two or more elevated kill signals without a strong structural anchor are terminal evidence only on a severe below-peg break (deeper than 1,000 bps, using the worse of the stored peak and live deviation). A deep break with no kill signal reads as Stress signals.
- Duration odds estimate the share of comparable incidents that recover, with terminal incidents as a competing outcome and still-open or non-recovery closures censored at their last observation (coin-weighted Aalen–Johansen). Quantiles the curve never reaches stay empty and the share not recovered within 30 days is published.
Impact notes
- Verdicts read as evidence classes: No break signals, Stress signals and Terminal evidence. Wire tier codes are unchanged; rows locked before v5 keep their original Recovery Likely / At Risk / Recovery Unlikely labels.
- Static exit thinness no longer condemns a coin. V9 exit capacity is re-scaled from the Safety Score stress request (a $1M floor) to the dislocation's value gap, and it, no-viable-route and thin redemption capacity cap at elevated. K5 is severe only when DEX TVL collapses at least 40% from the event's onset baseline on a thin market.
- Two or more elevated kill signals without a strong structural anchor are terminal evidence only on a severe below-peg break (deeper than 1,000 bps, using the worse of the stored peak and live deviation). A deep break with no kill signal reads as Stress signals.
- Duration odds estimate the share of comparable incidents that recover, with terminal incidents as a competing outcome and still-open or non-recovery closures censored at their last observation (coin-weighted Aalen–Johansen). Quantiles the curve never reaches stay empty and the share not recovered within 30 days is published.
- Typical-range and median readouts come from the incident-level curve instead of percentiles of per-coin medians. Horizon odds that fail the replay skill gate are published as not informative.
- New predictions persist a raw decision-input snapshot so later methodology changes can be replayed over sealed history. Frozen v2–v4 predictions keep their contents, hashes and methodology; they are not regraded.
Commit provenance not recorded