Architecture & Systems
Replay of Valuation Runs
Replay means re-running a past valuation with the same trades, snapshot, models, and code version to recover the original numbers.
Mental model
A valuation is replayable only when its trades, inputs, models, configuration, and code identity are recoverable.
Replay starts from a run ID, then loads the referenced snapshot, model parameters, and trade set as they existed at the time.
The same library version and configuration are used to recompute PVs and Greeks, which should match stored results within tolerances.
Explore the mechanism
Can the original result be regenerated from preserved inputs and then compared with a controlled change?
- A replay is: same trades + snapshot + models + library + config.
- Tight tolerances suit deterministic pricers. Monte Carlo needs tolerance tied to noise/SE.
- Fixed seeds make replay a stable check instead of a lottery.
What explains a difference between the original and replay?
What to notice
For Monte Carlo methods, fixed seeds or deterministic random generators ensure that replays remain consistent across time.
Replay is essential for debugging unexplained P&L, investigating model changes, and satisfying audit and governance demands.
Where it appears
- investigating valuation disputes
- reproducing regulatory reports
- measuring the effect of code or model upgrades
Where the intuition breaks
What this view leaves out
- Replay fails when external services, randomness, time-dependent behavior, or mutable artifacts were not captured.
Where this concept sits
Follow the strongest prerequisites, applications, extensions, and contrasts.
Connections for Replay of Valuation Runs
Replay requires the exact versions of both market inputs and model implementation.
UsesMarket Snapshots as BundlesReplaying a valuation requires the same market snapshot or an explicitly versioned equivalent.
UsesRun Configuration and DeterminismReplay is only meaningful when the original configuration, seeds and other deterministic settings are recoverable.
UsesSnapshot Versioning and CorrectionsReplay must resolve the exact historical snapshot version rather than silently using a corrected current one.
Supported byData Lineage and Audit TrailLineage identifies the exact inputs and transformations that a replay must reconstruct rather than silently substituting current data.
Supported byRun ID and ProvenanceA run ID provides the handle used to recover the inputs, configuration and versions needed for replay.
Replay requires the exact versions of both market inputs and model implementation.
Explore concept →