AlphaSense Launched SuperAnalyst. Let's Build a Tiny Assumption Watcher in TypeScript.
Most AI answers are finished the moment they are written. That is also the moment they start going stale. The memo says "renew the contract." Two weeks later, the price went up and half the seats sit empty. The memo still says "renew the contract." Confidently. On October 6, 2026 , AlphaSense announced SuperAnalyst , an always-on agent for financial and strategic work. It plans the work,…
<b>The Problem</b>
AlphaSense unveiled SuperAnalyst, an always-on agent designed for financial and strategic tasks. This agent plans work, reconciles evidence, runs code, and generates memos, models, and dashboards. Unlike static AI outputs, SuperAnalyst continuously monitors assumptions, identifies material changes in the market, and updates the work product with new information without requiring a new prompt. This distinction is crucial, as an answer that cannot indicate its assumptions cannot inform when it is wrong.
<b>Step 1: Write Down the Assumptions</b>
A report records the inputs it was built on and includes a tolerance for each input. The assumption type includes a key, value, and tolerance percentage. The tolerance percentage indicates how far a value can deviate before the conclusion becomes questionable. For example, price changes may have a tighter tolerance compared to seat count changes. The build function represents the analysis, taking inputs and producing the same output consistently.
<b>Step 2: Measure the Move</b>
A move percentage function calculates the percent change between the original value and the current value. This function does not compare the current value to the previous day's value but rather to the original value used in the report's construction. This approach highlights significant changes over time, as even small daily fluctuations can accumulate into a substantial change.
<b>Step 3: Classify the Change</b>
A check function classifies the change into four categories: fresh, noise, stale, or missing. If any assumptions are missing, the function returns a 'missing' check. For other changes, it identifies material moves exceeding the specified tolerance percentage, categorizing them as 'material' or 'noise' changes. If all moves are within the tolerance, the function classifies the change as 'fresh'. If no data is available for any assumptions, it marks the change as 'missing'.
<b>Step 4: Rebuild Only When Stale</b>
This step rebuilds the report only when a material change is detected, keeping the report current without unnecessary computational overhead. Stale reports remain unchanged, optimizing resource usage while ensuring accuracy when updates are essential.
<b>The Bigger Idea</b>
1. The problem lies in balancing between rebuilding reports daily, which incurs unnecessary computational costs, and never rebuilding them, leading to outdated information. The ideal solution lies in rebuilding reports only when significant changes occur.
2. Step 1: Writing down assumptions involves defining the inputs and their tolerances, allowing humans to set judgment calls on the acceptable deviation levels for each input.
3. Step 2: Measuring the move calculates the percentage change between the original value and the current value, focusing on significant alterations in the data.
4. Step 3: Classifying the change involves categorizing the observed changes into fresh, noise, stale, or missing, helping to determine the appropriate action.
5. Step 4: Rebuilding only when stale ensures that the report remains current without unnecessary computations, optimizing resource usage and maintaining accuracy.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
