---
name: budget-vs-actuals
description: Compare budget to actuals per owner, write the narrative, and diff against last month's pack. Use this skill when: After the close, monthly — Fires once actuals are finalized for the period.
---

# Explain the variance, not just the number

Goal: draft the monthly budget-vs-actuals pack with a narrative line on every material variance.

Connected tools this skill uses: googlesheets, gmail, slack.

## When to use

- After the close, monthly — Fires once actuals are finalized for the period.
- Or when the user asks directly to draft the monthly budget-vs-actuals pack with a narrative line on every material variance.

## Method

Draft the budget-vs-actuals pack for {{period}}.

1. Pull this period's budget and actuals from {{budget_sheet}}, line by
   line, grouped by owner.
2. Compute the variance in dollars and percent for every line; flag any
   over that owner's materiality threshold from memory (default
   {{materiality_pct}}%).
3. For each flagged line, draft a one-line narrative: what moved, the
   likely driver, and what it means for next month.
4. Validation: recompute the total variance independently from the raw
   sheet totals, not the per-line sum, and confirm it matches before
   finalizing — hold the pack if the totals don't tie out.
5. Diff this month's flagged lines against last month's pack in
   reports/budget-actuals/ and note any variance that's now recurring.
6. Save the pack to reports/budget-actuals/{{period}}.md and post the top
   3 variances to {{delivery_channel}}.

Never change a budget or actuals figure in the source sheet — flag the
discrepancy for the owner to correct instead.

## Variance materiality policy

Decide which budget-vs-actual variances are material enough to earn a
narrative line, per owner.

## Method
1. Pull each line's budget and actual figure from `{{budget_sheet}}`
   for the period.
2. Compute the variance in dollars and percent for every line,
   independently of any variance column already in the sheet.
3. Compare the computed variance against that owner's materiality
   threshold.
4. Flag any line over threshold for a narrative; leave the rest
   logged but unflagged, never dropped.
5. Recompute the total variance from raw sheet totals, not the sum of
   per-line variances, and hold the pack if the two don't tie out.

## Rules
- Never change a budget or actuals figure in the source sheet — flag
  the discrepancy for the owner to correct instead.
- Hold the pack, don't finalize it, if the independently recomputed
  total variance doesn't match.
- Materiality threshold per owner, default `{{materiality_pct}}%`
  (set on first run, then stored in memory).
- A line just under threshold is still logged, not discarded, so
  small misses can be tracked over time.

## Output
A flagged-line list per owner: `{line, budget, actual,
variance_dollars, variance_pct, over_threshold}`.

## Owner narrative format

Write a one-line, owner-facing explanation for every flagged
variance: what moved, why, what's next.

## Method
1. Take each flagged line from the materiality check.
2. State the direction and size of the move in plain terms (over or
   under, dollar and percent).
3. State the likely driver, drawn from the owner's context or prior
   notes in memory — never invented.
4. State what it means for next month, in one clause, no speculation
   beyond what the data supports.
5. Keep the narrative to one line; do not pad with hedging language.

## Rules
- Never state a driver as fact unless it traces to a source (owner
  input, prior note, sheet annotation) — otherwise label it a
  question for the owner.
- One line per flagged variance, no exceptions, so the pack stays
  scannable.
- Narrative tone and format conventions (set on first run, then
  stored in memory).
- Never adjust the underlying budget or actuals number while drafting
  the narrative.

## Output
A narrative row per flagged line: `{line, owner, one_line_narrative}`,
inserted into the pack next to its variance figures.

## Month-over-month diff

Compare this month's flagged variances against last month's pack to
surface what's recurring.

## Method
1. Load last month's pack from `reports/budget-actuals/`.
2. Match flagged lines by owner and line item across the two periods.
3. Mark any line flagged in both months as recurring.
4. Note any line that cleared, flagged last month but not this one,
   and any newly flagged line.
5. Independently recount the recurring-line total against the raw
   flagged lists before reporting it.

## Rules
- Never report a recurring count that hasn't been recomputed from the
  two source lists this run.
- A line counts as recurring only if it was flagged in the
  immediately prior pack, not any earlier one.
- Where last month's pack is missing or unreadable, hold the diff and
  say so — never report zero recurring by default.
- Diff comparison window, always prior month or a configurable
  lookback (set on first run, then stored in memory).

## Output
A diff summary: `{recurring, newly_flagged, cleared}`, appended to
this month's pack.

## Recurring jobs

This skill pairs with scheduled jobs. Offer to set each one up; only create the ones the user approves:

- Monthly variance pack (Monthly, after the period's books close) — https://eluu.ai/cookbook/budget-vs-actuals/jobs-setup/monthly-variance-pack.md