> For the complete documentation index, see [llms.txt](https://thesium.gitbook.io/thesium-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://thesium.gitbook.io/thesium-docs/feature-deep-dives/trenches-pulse-and-conviction-score.md).
# Trenches Pulse and Conviction Score
The conviction score is the most opinion-loaded number in the product, and it deserves its own section explaining what it is, what it is not, and how to read it.
## What the score represents
A conviction score of N out of 100 means: across the signals the model evaluated for this token, the agreement that the near-term direction is `sentiment` (bullish or bearish) has weight N. A score of 50 with sentiment `mixed` means the signals contradict each other. A score of 90 with sentiment `bullish` means the signals broadly agree on a bullish read.
The score is a confidence statement about the signal stack, not a recommendation, not a price prediction, and not a probability of any specific outcome. A 95 conviction score does not mean "95% chance the token goes up." It means "the signals we evaluated agree this strongly that the direction is up." Markets do what markets do regardless.
## What signals feed the score
The signal categories the model evaluates include — non-exhaustively:
* **Volume trajectory** — direction and slope of recent volume, buy/sell imbalance, transaction count vs volume ratio.
* **Holder concentration** — top-N hold rate, holder count trajectory, fresh-wallet ratio.
* **Insider structure** — sniper hold rate, bundler trader rate, dev-team hold rate.
* **KOL flow direction** — net inflow or outflow of tracked wallets in the recent window, with weight by tier.
* **Social velocity** — recent rate of feed entries above the position-size threshold, sentiment of those entries.
* **Price action shape** — recent price profile relative to ATH, time since launch, drawdown depth.
The model is shown each signal with its current value and its short-term trajectory, and asked to produce a structured judgment plus three highlights — the three signals most responsible for the current read.
## Why the score is reproducible
For a fixed input snapshot — same token, same signals at the same values — the model returns the same output. Reproducibility is enforced two ways:
* Temperature is set to zero for summary generation. Standard mechanism to make LLM outputs deterministic given a fixed prompt.
* Inputs are timestamped and cached. The model never re-judges the same exact input twice; the cache returns the prior output. New evaluations happen only when at least one input has changed.
This matters because traders take screenshots. A conviction score that shifts on refresh with no underlying change would erode trust within minutes. The score moves when the market moves, not when the model has a different mood.
## How to read the score
* **80–100, bullish or bearish.** Strong consensus in the signal stack. Worth taking seriously, but ask why — the highlights explain which signals are driving the score.
* **50–79, bullish or bearish.** Lean in one direction with some dissent. Useful as a tilt, not a thesis.
* **Below 50 with mixed sentiment.** The model is telling you the signals disagree. Trade your own conviction, not the panel's.
* **A score that updates dramatically across polls.** Something material happened in the last polling window — a large KOL entry, a structural metric flipped, a social burst. The highlights tell you what.
## What the score does not do
* Consider macro conditions outside the token's own footprint.
* Consider broader narratives except inasmuch as social velocity picks them up.
* Predict price targets or recommend position size.
* Adjust for the user's existing position, risk tolerance, or time horizon.
It is a signal-stack confidence read. Nothing more, and nothing pretending to be more.
---
# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.
## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:
```
GET https://thesium.gitbook.io/thesium-docs/feature-deep-dives/trenches-pulse-and-conviction-score.md?ask=&goal=
```
`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.