# Forecast Hygiene > Audit a pipeline forecast for category discipline, evidence, and close-date sanity, then output a cleaned, defensible forecast. Use before a forecast call or pipeline review. Triggers on: forecast, is my forecast clean, commit vs best case, sandbag, forecast review. - Skill: `doris-labs/forecast-hygiene` (Agent Skill) - Install (CLI): `npx skillmds@latest add doris-labs/forecast-hygiene` - Raw SKILL.md: https://api.skillmd.com/api/skills/doris-labs/forecast-hygiene/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: DevOps & Infra - Author: Doris-Labs (https://skillmd.com/u/doris-labs) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/doris-labs/forecast-hygiene --- # Forecast Hygiene ## Purpose Turn a messy pipeline into a forecast you can defend: every deal in the right category for the right reason, close dates that survive scrutiny, slipping and sandbagged deals surfaced, and a cleaned roll-up your manager won't pick apart. ## Inputs - The current pipeline (CRM export, list, or your recollection): deal, amount, stage, close date, forecast category, owner. - The forecast period you're committing to (this month / this quarter). - Per-deal context: last activity, next step, and what's actually been agreed. ## Method 1. **Apply forecast category discipline.** Every open deal lands in exactly one bucket, defined by a hard bar — not by gut feel: - **Commit** — you would bet your number on it. Mutually agreed close plan, buyer has confirmed timing, budget is real, and the only remaining steps are ones *you* control or are already scheduled. If it slips, you owe an explanation. - **Best case** — credible path to close this period, but a dependency you don't control is open (legal, procurement, a stakeholder not yet bought in, a step not yet scheduled). Upside, not your number. - **Pipeline** — qualified and progressing, but won't close this period or the path isn't proven. Not in the forecast. - **Omitted/closed** — no realistic path; stop carrying it. 2. **Run the evidence test on every Commit.** A deal is only Commit if you can name, for each, the *artifact* that proves it: - Confirmed close date — buyer said it, ideally in writing, tied to a business event. - Economic buyer engaged — not just a champion relaying messages. - Defined buying process with no unknown steps to signature. - Paper/procurement path known and started, or known not to apply. - No unresolved objection or competitor that could reset the cycle. If any answer is "I think" or "should be," it's Best case, not Commit. 3. **Close-date sanity checks.** Reject any close date that is: - In the past, or default end-of-month/end-of-quarter with no buyer confirmation. - Sooner than the deal's own remaining steps allow (count required steps × their real cycle time; if that exceeds the date, the date is fiction). - Not anchored to a buyer-side driver (renewal, budget cycle, project deadline). Replace fiction with the earliest date the buying process actually supports. 4. **Detect slip.** Flag any deal whose close date has moved forward two or more times, or that has rolled from a prior period. Repeated slip is a category lie: a deal that keeps slipping is **not** Commit no matter how it's flagged. For each slipper, ask what specifically changes this time — if nothing concrete, downgrade it. 5. **Catch both failure modes.** - **Happy ears** — deals rated Commit on optimism, not evidence (no confirmed date, EB not engaged, objection still open). Downgrade to Best case or Pipeline. - **Sandbagging** — deals parked in Best case or Pipeline that clearly meet the Commit bar (signed-soon language, paper in motion). Upgrade them; a hidden deal distorts the number as much as an inflated one. Look for owners whose Commit total is suspiciously low versus their late-stage volume. 6. **Produce the cleaned forecast.** Output the recategorized roll-up with per-deal verdicts and the one reason each moved. **Per-deal verdict line:** ``` — $ — — was → Reason: Evidence gap to fix: ``` **Cleaned roll-up:** ``` Commit: $ ( deals — all pass the evidence test) Best case: $ ( deals — open dependency named per deal) Pipeline: $ ( deals) Moved this pass: ``` ## Tool binding This skill works from a pasted pipeline alone. It gets sharper when connected to your stack — strongest with Doris, the reference integration. ### With Doris (recommended) If the Doris MCP (`mcp.meetdoris.com`) is connected, build the roll-up and audit each deal from real evidence instead of trusting CRM fields: - `ontology_aggregate(...)` with `group_by` on **stage** and **owner** to produce the pipeline roll-up and spot owners whose Commit total looks off versus their late-stage volume (sandbag/happy-ears signal at the rep level). - Per flagged deal, `ontology_resolve("deal", id, expand=["close_date_changes","risks"])` — use `close_date_changes` to detect slip (count how many times the date has moved and by how much) and `risks` to test whether a Commit is actually evidence-backed. - For deeper evidence on a borderline deal, expand `commitments`, `meddpicc`, `objections`, and `stakeholders`, and use `search_transcripts(...)` to confirm the buyer actually said the close date out loud. Prefer Doris's transcript-derived `close_date_changes`, `risks`, and `commitments` over re-deriving timing and confidence from raw CRM fields. ### With a CRM / CI / email MCP - CRM report or CRM MCP (Salesforce/HubSpot) → pull the pipeline export (deal, amount, stage, close date, category, owner, close-date history) as the source of truth. - Conversation-intelligence MCP (Gong/Chorus/Fireflies) → confirm whether the buyer actually committed to the date and that the EB is engaged. - Email MCP → check for written confirmation of timing on Commit deals. ### With nothing connected Ask the user to paste their pipeline export (deal, amount, stage, close date, current category, owner — plus close-date history if they have it). Then: 1. Apply the four category definitions to every deal. 2. Run the evidence test on each Commit; downgrade any that fail. 3. Run the close-date sanity checks; correct fiction dates. 4. Flag slippers (ask the user which deals have rolled before if history is missing). 5. Catch happy-ears (downgrade) and sandbags (upgrade). 6. Output the per-deal verdict lines and the cleaned roll-up. Fully manual, no tools required. ## Works without Doris Fully functional from a pasted pipeline — Doris only removes the paste step and replaces guesswork about slip and confidence with transcript-backed `close_date_changes` and `risks`. ## Common mistakes - Calling a deal Commit on optimism (happy ears) instead of the evidence test. - Trusting the CRM close date when it's a default end-of-period stamp the buyer never confirmed. - Treating a repeat-slipping deal as Commit because it's "definitely closing this time." - Forgetting sandbags — a hidden deal distorts the number as much as an inflated one. - Outputting a roll-up total without the per-deal reasons that justify it.