Correlation Matrix - cov

This sheet defines correlation structure for multiplier draws, and optionally for organization-specific noise when using the vertical format.

Accepted Formats

The parser accepts two formats.

Matrix Format

Example:

Impact Multiplier	Hedginess	FundingAdditionality	ActivityAdditionality
Hedginess	1.0
FundingAdditionality	0.3	1.0
ActivityAdditionality	-0.1	0.0	1.0

Rules:

• The first column should be named Impact Multiplier.
• Row and column labels are normalized by removing spaces and dots.
• Lower-triangle-only input is acceptable; the parser symmetrizes the matrix.
• Missing pairs default to 0.
• Missing multipliers are allowed; unrecognized names are filtered out.

Vertical Format

Example:

Variable1	Variable2	Correlation	Type
Hedginess	FundingAdditionality	0.3	Multiplier
FundingAdditionality	Hedginess	0.3	Multiplier
OrgA	OrgB	0.5	Noise

Rules:

• Required columns are Variable1, Variable2, and Correlation.
• Type is optional. If omitted, the parser assumes Multiplier.
• Valid Type values are Multiplier and Noise.
• Noise rows are only used if organization names are available from the rest of the input.

See Noise And Model Coverage for how Type = Noise correlations are used after expected values are computed.

Parser Behavior

The current implementation:

• Detects the format automatically.
• Removes duplicate row or column names in matrix format, keeping the first occurrence and recording an error.
• Removes duplicate (Variable1, Variable2) pairs in vertical format, keeping the first occurrence and recording an error.
• Makes correlations symmetric by copying (A, B) to (B, A) when needed.
• Completes any missing pairs with 0.
• Returns an identity-like fallback for multiplier correlations if parsing fails badly enough to produce no matrix.

Safer Authoring Conventions

To keep docs and spreadsheets predictable:

• Prefer explicit 1 values on the diagonal.
• Prefer lower-triangle matrix input or explicitly symmetric vertical input.
• Use vertical format if you need Noise correlations between organizations.
• Keep names aligned with normalized multiplier and organization names used elsewhere in the workbook.

Notes For Agents

• The cov sheet is read after imp and before fos/orgmeta are fully combined, but multiplier correlation is finalized only once the engine knows the final multiplier set.
• Formula-generated multipliers can extend the final multiplier correlation matrix later; newly added formula multipliers get zero correlation with other multipliers by default.
• The Shiny preview can display either observed correlations from simulated draws or the input correlations from this sheet.