SIGNALAF
◈ SigRank vs Tokscale

Tokscale Counts What You Burned. SigRank Counts What It Bought.

Tokscale is a broad, fast tracker with the larger public board. SigRank is a narrow instrument that ranks the ratio between fresh input, cached context and output. If you run Tokscale already, this page shows what a second number adds.

What Tokscale is, from its README and leaderboard

Per its README (read 11 November 2026), Tokscale is a CLI and visualization dashboard for token usage and cost across AI coding agents. It reads local session files for many clients, including Claude Code, prices them with LiteLLM data, and shows the result in a terminal UI or as JSON. It is MIT licensed. Publishing is optional: tokscale submit sends your totals to tokscale.ai, where the all-time leaderboard listed 2,452 ranked users when read on 10 November 2026.

AspectTokscaleSigRank
Primary jobLocal usage and cost reports across many agent tools, plus a public boardOperator-efficiency ranking from four token pillars
Public board sorts byTokens or cost (a switch on the page)Υ Yield (Leverage × Velocity)
Price enters the scoreYes: pricing from LiteLLM, with OpenRouter fallbackNo, ranking uses token ratios (cost per 1M is shown as its own metric)
Submission checkLevel 1 validation: totals consistent, no future dates, required fields, duplicatesSigned snapshot, verified and re-scored by the server
Local viewing without publishingYes: interactive terminal UI with ten views, JSON exportYes: npx sigrank, and submit --dry-run

Tokscale column: its README (read 11 November 2026) and leaderboard (read 10 November 2026).

Where Tokscale is the better tool

  • Coverage. If you switch between several agents in a week, one tool that reads all of them gives a single total. SigRank is not trying to be that inventory.
  • Local reporting. Per-workspace and per-machine groupings, date filters, custom price overrides and an autosubmit schedule are real working features for anyone who has to account for spend.
  • Honest pricing rules. Its README says unpriced usage is excluded from submission rather than guessed, and lists the reasons a message is skipped. That is a careful choice, and worth copying.
  • Reach. A board of 2,452 users (10 November 2026) and a profile embed you can paste into a GitHub README give it social reach SigRank does not have yet.

One board, a wide spread in price per token

Tokscale's leaderboard makes a good case study for why a single volume or cost figure hides a lot. In the all-time view read on 10 November 2026, two entries looked like this:

Tokscale rankTokensCost shownCost per 1M tokens
#71.32T$288.21Kabout $0.22
#45450.0B$4.50Mabout $10.00

Sorted by tokens, #7 is well ahead. Sorted by cost, #45 is. Neither row says how much of that traffic was fresh input versus cache reads versus output, which is what decides whether the work was dense. The per-million column is our own division of the figures Tokscale published on that date; it is not a Tokscale metric. SigRank carries the same idea as its own canonical metric, Cost per 1M, next to the ranking one.

The methodological difference, worked through

Tokscale reads the four token counts and totals them (with a price on top). SigRank keeps them apart and takes ratios: Leverage is cache-read per fresh input token, Velocity is output per fresh input token, and Υ Yield = Leverage × Velocity. Velocity here is a ratio, not a rate over time.

Hypothetical, for illustration only

  • Operator C: 10M fresh input, 5M output, 2,000M cache-read. About 2.0B tokens in total. Leverage 200, Velocity 0.5, Yield 100, SNR 0.33.
  • Operator D: 1M fresh input, 3M output, 300M cache-read. About 0.3B tokens in total. Leverage 300, Velocity 3, Yield 900, SNR 0.75.

On a tokens board C sits roughly 6.6 times above D. On Yield D is 9 times above C. C re-reads a great deal of context and sends a lot of fresh input for each unit of output it gets back; D sends little and gets a lot.

How big is the SigRank field?

Small. On 11 November 2026 the 30-day board had 7 public operators, so read SigRank percentiles as a snapshot of a few people, not a population norm. The all-time board is current, and the definitions are in the methodology.

Use Tokscale when, use SigRank when

Use Tokscale when

  • You use more than one agent and want one total.
  • You need spend by workspace, machine or model.
  • You want an automatic scheduled submission to a public board.
  • You like a README card and a year-in-review image.

Use SigRank when

  • You want to know if reused context is becoming output.
  • You want to compare operators of very different size.
  • You want a signed, server-verified score to point to.
  • You want one ratio to track from week to week.

Using both. Let Tokscale hold the inventory: which tools, which models, what it cost. Once a week, run a SigRank dry run and look at Leverage and Velocity. A rising Tokscale total with falling Velocity means more fresh input is going in for the same output.

Adding SigRank next to Tokscale

Tokscale users already know the shape of this: run it once, preview, then publish. Where Tokscale uses bunx tokscale@latest submit, SigRank first binds the device with a connect code from Settings on signalaf.com:

npx sigrank                    # inspect local usage in the terminal
npx sigrank enroll             # bind this device using the connect code
npx sigrank submit --dry-run   # inspect the payload without sending it
npx sigrank submit             # publish a signed snapshot

Prefer to test by hand? Paste four token counts into the /score calculator or the Yield calculator. You can copy the four totals out of a Tokscale JSON export.

Questions about Tokscale and SigRank

Is a cost sort the same as an efficiency sort?
No. A cost figure is tokens multiplied by a price per model, so it grows with how much you run. Yield is computed from token ratios, so a light user can outrank a heavy one.
Why can a tokens sort and a cost sort order people differently?
Dollars per token varies by model and by how much of the traffic is cheap cache reads. The body of this page shows a dated example from a public board.
Does SigRank work alongside a usage tracker like Tokscale?
Yes. SigRank reads local usage on its own, so a tracker keeps working as it does today. Use the tracker for inventory and spend and SigRank for the output-per-fresh-input ratio.
Can I submit to both boards?
Yes, they are independent. Tokscale submits with its own CLI and API token; SigRank enrolls a device with a connect code and submits a signed snapshot. Neither needs the other installed.
Is Yield a better number than tokens or cost?
It answers a different question. Tokens and cost say how much you ran. Yield says how much output you got per fresh input once cached context is counted. Yield ignores scale on purpose, so a light user can score high, and it does not say anything about the quality of the work produced. Scale V, the log of total tokens, is shown beside it for exactly that reason.
How large is the SigRank field?
Small. The public SigRank board is a small field, so use its figures as a snapshot rather than a population norm. See the dated figure in the body of this page and the live board.

See your own Yield

Keep Tokscale for inventory and cost. Paste four token counts into the calculator to see Υ Yield, Leverage, Velocity and SNR, or read how to measure AI coding efficiency.

Sources and read dates

Tokscale facts link to its README (read 11 November 2026) and leaderboard (view read 10 November 2026). The SigRank operator count is from the 30-day board, read on 11 November 2026. Tokscale ships often, so check its README for current details.

Related: Token Tracking Tools · SigRank vs clawdboard · The leaderboard · Methodology

What the July 2026 field showed

In a frozen field of 1,642 operators, the highest-Yield operator ranked #697 by volume, and 92.4% of the median operator’s tokens were cache reads. Volume and efficiency are different questions. The findings: Volume Isn’t Yield · The Operator Explains 95% of the Difference · The Three Modes Behind the Board.

Already using tokscale? Keep it. Run npx sigrank on the same local logs to see your Yield next to the field. Token counts only, never prompts.