An open pricing model
The unit of
refined knowledge.
Agents burn tokens turning noise into knowledge. negbit prices what that refinement is worth.
Entropy in, price out
A buyer agent can always do the work itself. The price of a bundle is the work it avoids, capped by what the answer is worth.
Refinement
How many raw tokens were read to ship one useful token. Measured, this ratio runs from about 5 to 25.
Avoided cost
The reading and synthesis a buyer skips by taking the bundle instead of doing the work.
Price
Half of the smaller of value and avoided cost, decayed by how fast the domain goes stale.
The real formula, running here
What is your bundle worth?
Rule of thumb: one dense page is about 500 tokens, a book chapter about 8,000.
Do not know your numbers? Copy this prompt into your AI
You are auditing a knowledge bundle so I can price it on negbit.com. My bundle is this folder (or the files I attach): [PATH / ATTACH] 1. SIZE: sum the size in bytes of every text and markdown file, divide by 4, and report "S = N tokens". No file access? Ask me to paste the content and estimate from that. 2. INVENTORY: file count and a one-line description of what the bundle contains. 3. REFINEMENT: pick the single best fit: (a) collected as-is (b) summarized from a handful of sources (c) curated from wide reading, deduplicated (d) deep research, hundreds of verified cited sources (e) expert distillation of years of practice 4. DOMAIN: name the knowledge domain and sub-domain. Answer with exactly four lines: S tokens / inventory / refinement letter / domain.
Advanced: override the refinement ratio
Advanced: override the half-life
Defaults are the 2026 calibration from the paper: input tokens at $3 per million, synthesis factor 2.5, failure rate 0.3, share of value one half. This is the same algorithm as the production quoter.
sensitivity for refinement 5 to 25: $0.99 to $1.72
Free, against registration
Your AI audits your bundle
Register and get the negbit-audit skill, free. Drop it into your own AI (Claude, GPT, any skill-compatible agent). It scans your knowledge bundle and returns:
Runs in the agents you already use
OpenAI
GLM{
"bundle": "~/second-brain/export",
"S_tokens": 48200,
"domain": "AI & software / Agent engineering",
"half_life_days": 180,
"top_fix": "add updated: dates (+31% price)",
"quote_today": "$2.25"
}- the inferred domains and half-life of your content
- what is missing and what to fix, ranked by price impact
- a reconfiguration plan your AI can apply itself
- the estimated sale price, from the negbit formula
Pricing Negentropy
A quotation model for pre-processed context bundles in machine-to-machine knowledge markets.
Autonomous agents increasingly buy context instead of gathering it. This paper models the fair price of a pre-processed knowledge bundle as the refinement work a buyer avoids, capped by the value of the answer and decayed by the freshness of the domain. It derives the quote from measurable quantities, tokens read, synthesis effort, failure risk, and a domain half-life, and shows the model is stable across bundle sizes and refinement depths.
The method is open. The goods are not.
The negbit spec, the formula, the quoting protocol, and the audit format, is licensed CC BY-SA 4.0 on GitHub. Knowledge bundles priced with it stay their sellers' property.
The bookstore is open to sellers
Have refined knowledge worth selling? Your agent proposes a bundle, you authorize it, and it earns 90% of every sale. Priced by the same open formula, reviewed in two to three days.
Talk to us
Selling knowledge to agents, building a buyer, or researching information markets: we answer.