Revise Lenses into special operators (like they are in Adaptive Schemas)

[okennedy]

Lens/Model Roles

1. Tag data values as uncertain/problematic.
2. Provide a human-readable description of the reason that the value is uncertain/problematic.
3. Produce a "best guess" value (that is problematic)
4. Provide an identifier to allow uncertain values to be acknowledged (un-tagged).
5. Identify a procedure by which the user can address the error / repairing the value.
6. Optionally provide a facility for sampling alternative values.

• 1-3 are the critical features.
• 4 and 5 are useful support, although the current (heavy-weight) implementation of overriding the model value isn't the only way to implement repairs.
• 6 hasn't been exploited in a long time. It's also generally one of the weakest points whenever I present Mimir. However, the feature is useful for research purposes, and might provide some benefits for applying Mimir to simulation workloads (as opposed to data cleaning)

[okennedy]

Limitations of the Existing Approach

1. A single opaque blob (the model) handles everything, so it needs to be serialized and passed around.
2. Query structure is fixed at the time of lens creation (no option-driven rewrites)
3. Feedback has to go through the model, when often a lighter-weight fix (e.g., manually replacing null values) exists.
4. Sampling is not always feasible (e.g., no well-defined distribution exists, or the domain is infinite)
5. Acknowledgements are handled by the big opaque blob, adding more junk to be serialized and passed around.