estimate accuracy calibrator
Converts raw engineering estimates into calibrated P10/P50/P90 probability distributions using historical evidence.
$5
Works with the AI tools you already use
Claude Code
Cursor
Codex CLI
GitHub Copilot
Gemini CLI+20 moreClaude CodeCursorCodex CLIGitHub CopilotGemini CLI
OpenClaw
Manus
Grok+17 agentsestimate accuracy calibrator
Example session with this skill installed
Calibrate this estimate: 10 days for a new payment gateway integration. Historicals show a 1.4x median overrun on integrations (N=5). Drivers: API docs are messy, team is new to this gateway.
- Read your context and instructions
- Compiled the estimate accuracy calibrator
I have processed the payment gateway integration estimate using the calibration framework. I applied a 1.4x historical multiplier and adjusted for the "New Domain" class quality and documented uncertainty drivers.
Calibrated Forecast: Payment Gateway Integration
1. Original Estimate (Preserved)
10.0 days (Inside-view anchor)
2. Historical Reference Data
| Metric | Value | Notes |
|---|---|---|
| Sample Size (N) | 5 | Low confidence due to small N |
| Median Overrun | 140% (1.4x) | Baseline for comparable integrations |
| P90 Overrun | 195% (1.95x) | Worst-case historical performance |
| Data Quality | 0.60 | Moderate; relevant but limited sample |
3. Calibration Adjustments
| Factor | Adjustment | Impact |
|---|---|---|
| Class: New Domain | +50% | High novelty architecture risk |
| Driver: Messy Docs | +20% | Search/debugging overhead |
| Driver: New Team | +15% | Ramp-up time |
| Combined Multiplier | 2.25x | Compounded uncertainty |
4. Calibrated Forecast Distribution
| Percentile | Forecast | Probability |
|---|---|---|
| P10 (Best Case) | 14.5 days | 10% chance of finishing earlier |
| P50 (Median) | 22.5 days | Most likely outcome |
| P90 (High Conf) | 31.0 days | 90% chance of finishing within this |
5. Commitment Threshold Test
- Target: 10 days (Original Estimate)
- Probability of Success: < 5%
- Risk Level: EXTREME. The target falls well below the P10.
6. Controlling Uncertainty
Class Quality (New Domain) is the primary driver, contributing ~45% of the total range width. Reducing this through a 1-day spike would significantly tighten the forecast.
7. Forecast vs. Promise Separation
Evidence-Based Forecast
22.5 days (P50) to 31.0 days (P90). The data suggests the current 10-day target is mathematically improbable.
Recommended Promise
28 days. This provides a 75% confidence level while allowing for the documented documentation and team experience risks.
Next steps
- Conduct a 24-hour technical spike to reduce the "New Domain" uncertainty.
- Review the P90 with stakeholders to negotiate a realistic buffer.
- Update the forecast once the API documentation has been fully audited.
Connects securely to your tools. The creator never sees your data.
What you get
About this skill
The problem
Engineering estimates are often based on the "inside view," leading to systemic underestimation and missed deadlines. Developers struggle to account for hidden complexity, historical overruns, and external dependencies when pressured to provide a single number.
What it does
- Preserves the original anchor estimate while applying evidence-based calibration.
- Calculates P10, P50, and P90 probability distributions based on historical overrun factors and class quality.
- Decomposes uncertainty into specific drivers like requirement volatility and dependency risk.
- Tests commitment thresholds to determine the statistical probability of meeting a hard deadline.
- Separates the evidence-based forecast from the external promise to prevent conflating data with commitments.
Why this beats prompting it yourself
General prompts usually result in vague "add a 20% buffer" advice. This skill uses a structured mathematical framework to decompose uncertainty and apply historical reference class forecasting, providing a defensible audit trail for why a project might take longer than originally guessed.
Use cases
- Calibrating a sprint plan before committing to a hard release date.
- Evaluating the risk level of a high-novelty R&D project with no clear precedent.
- Presenting a data-backed range to stakeholders instead of a single, fragile date.
- Identifying which specific technical unknown is creating the most scheduling risk.
Known limitations
Requires honest inputs regarding historical data quality. Forecast accuracy scales with the quantity of comparable historical data points provided.
How to install
Works the same in every agent - Claude, Cursor, Codex, Copilot and 20+ more.
- 1
Download the ZIP
Free skills download straight away. Paid skills unlock right after purchase.
- 2
Unzip into your skills folder
Every agent reads skills from one folder on your machine. Drop the unzipped folder in there.
- 3
Ask your agent to use it
Restart the agent if it was already running. It picks the skill up automatically - no config needed.
Skills folder by agent
Click the path to copy it. Create the folder if it does not exist yet.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill passes our trust checks below.
Security scanned
Passed our 8-point scan before listing
Fresh listing
Recently published to Agensi
30-day refund
Not a fit? Get your money back
Trust & safety
Security scanned
Verified clean 17 days ago
- Passed all security checks, Safe to install