sigrank-agent 0.1.1

SigRank local telemetry agent — scans AI-platform logs, computes the 11-metric stack, publishes signed snapshots.

sigrank-agent

SigRank local telemetry agent. Scans your AI platform logs on-device, computes your token cascade, and publishes a signed snapshot to the SigRank leaderboard.

Token counts only. Nothing else ever leaves your machine.

→ signalaf.com

---

What this is for

sigrank-agent is for operators who want to get their real usage data onto the leaderboard — not a paste, not an estimate, but a locally-computed, cryptographically signed snapshot from your actual log files.

Once published, your profile appears on signalaf.com with your full cascade across all four measurement windows (7d / 30d / 90d / all-time). Your profile workspace on signalaf.com will list every cascade you've published, organized by model and window, so you have a full history of your efficiency over time.

The companion tool sigrank-mcp lets you check the leaderboard, see your rank, and compare operators from your terminal — no publish needed.

---

Install

pip install sigrank-agent

Requires Python ≥ 3.11. Or install isolated with pipx (recommended):

pipx install sigrank-agent
sigrank-agent version

---

Quick start

Five commands from zero to published:

# 1. Initialize — creates ~/.sigrank/ with keypair, database, config
sigrank-agent init

# 2. Point at your Claude logs (the default path)
sigrank-agent source add claude-code ~/.claude/projects

# 3. Scan — reads your logs, loads into local database
sigrank-agent scan

# 4. Compute — runs the metric stack for a window
sigrank-agent compute --window 30d

# 5. Preview — inspect what will be published
sigrank-agent preview

# 6. Publish — sign with ed25519 and POST to signalaf.com
sigrank-agent publish

---

All commands

sigrank-agent init

Initialize ~/.sigrank/ — keypair, config, database, subfolders. Run once.

Codename (3-32 chars, letters/digits/-/_): YourCodename
Primary platform [claude]:
Server URL [https://signalaf.com]:

✓ Initialized /Users/you/.sigrank
✓ Generated device keypair: ed25519:xxxx… (signing OFF by default)
✓ Codename: YourCodename
✓ Ready. Run sigrank-agent source add to point at a data source.

---

sigrank-agent source

Manage which log directories the agent scans.

sigrank-agent source add claude-code ~/.claude/projects   # add Claude Code logs
sigrank-agent source list                                  # show configured sources
sigrank-agent source remove claude-code                   # remove a source

Currently supported source type: claude-code — reads ~/.claude/projects/*/*.jsonl

More adapters coming (Codex, Amp, Gemini, and others are supported in sigrank-mcp's reader and will land in the agent as the platform list expands).

---

sigrank-agent scan

Parse configured sources into the local SQLite database.

sigrank-agent scan                          # scan all sources
sigrank-agent scan --source claude-code     # scan one source only
sigrank-agent scan --since 2026-06-01       # only messages on/after this date

✓ Scanned claude-code  ·  1,208 files  ·  84,302 messages

Token counts, model IDs, and message timestamps are stored locally. No content is read.

---

sigrank-agent compute

Run the 11-metric stack over your scanned data for a measurement window.

sigrank-agent compute                       # default: 30d
sigrank-agent compute --window 7d
sigrank-agent compute --window 90d
sigrank-agent compute --window all_time

Windows: today · 7d · 30d · 90d · all_time

✓ Computed snapshot  ·  window: 30d  ·  2026-06-24T06:00:00Z

  Υ Yield       12,847.2
  SNR               92.3%
  Leverage       2,041.1x
  10xDEV             3.31
  Velocity           9.0x
  Class         TRANSMITTER

  Pillars
  input          1,251,211
  output        11,296,121
  cache_create 128,196,310
  cache_read 2,555,179,769

---

sigrank-agent preview

Inspect the computed snapshot before publishing. Shows the full payload.

sigrank-agent preview           # formatted display
sigrank-agent preview --json    # raw JSON payload

---

sigrank-agent publish

Sign the snapshot with your device keypair (ed25519) and POST it to signalaf.com.

sigrank-agent publish

  Signing       ✓
  Posting       ✓
  Validated     ✓

✓ Published. Visit https://signalaf.com/user/YourCodename shortly.

The server recomputes the canonical hash and verifies your signature before accepting the snapshot. Your profile updates within seconds.

---

sigrank-agent history

Show your local publish history.

sigrank-agent history           # last 20
sigrank-agent history --limit 5

  #   published_at              window    Υ Yield    class
  ─────────────────────────────────────────────────────────
  1   2026-06-24 06:00:00 UTC   30d       12,847.2   TRANSMITTER
  2   2026-06-23 18:00:00 UTC   7d         8,231.4   TRANSMITTER
  3   2026-06-22 12:00:00 UTC   all_time  18,436.98  TRANSMITTER

---

sigrank-agent verify

Re-verify the signature and hash of the most recent snapshot.

sigrank-agent verify

---

sigrank-agent config

Get or set configuration values.

sigrank-agent config --list                         # show all config
sigrank-agent config codename                       # get one value
sigrank-agent config signing_enabled true           # enable ed25519 signing
sigrank-agent config server_url https://signalaf.com

---

sigrank-agent version

sigrank-agent version

sigrank-agent     │ 0.1.0
schema_version    │ 1.0
ruleset_supported │ 1.0
python            │ 3.11.x
platform          │ darwin / arm64
signing           │ off

---

Your profile on signalaf.com

Once you publish, your profile at signalaf.com/user/your-codename shows:

• Your current rank and class tier
• Cascade metrics across all four windows (7d / 30d / 90d / all-time)
• Your full publish history — every cascade you've submitted, by model and window

Profile workspace (coming): a full operator workspace where you can manage all your submitted cascades, see model-by-model breakdowns, and track your efficiency over time. Future support for importing full sessions and, eventually, chats.

---

How it works

scan (read logs) → compute (metric stack) → preview → publish (sign + POST)

• Local database — all parsed data lives in ~/.sigrank/db.sqlite. Nothing leaves your machine until you explicitly publish.
• Deterministic — given the same logs, the same snapshot is always produced. No randomness, no wall-clock reads in the metric pipeline.
• Signed — each snapshot is signed with a per-device ed25519 keypair generated at init. The server verifies before accepting.
• Token-only — content character lengths are used as a proxy for one metric (Prompt Complexity). Content text is never stored, transmitted, or logged.

Metric stack

Metric	What it measures
Υ Yield	(cache_read × output) / input² — overall cascade efficiency
SNR	Signal-to-noise ratio — cache pull vs total push
Leverage	cache_read / input — memory amplification
10xDEV	log₁₀(Leverage) — orders of magnitude above baseline
Velocity	output / input — generation rate

---

Configuration reference

~/.sigrank/ holds everything:

File	Contents
config.json	Codename, server URL, signing flag, platform
keypair.json	ed25519 device keypair (never transmitted)
db.sqlite	Parsed telemetry, computed snapshots, publish history
imports/	Reserved for future session import
cache/	Scan state cache
exports/	Local snapshot exports

Override the home directory: SIGRANK_HOME=/path/to/.sigrank

Other env vars: SIGRANK_SERVER · SIGRANK_TIER · SIGRANK_LOG_LEVEL · SIGRANK_NO_COLOR

---

Privacy

• No content, ever. The agent reads token counts, model IDs, timestamps, and content lengths only. Message text never leaves your device.
• Local by default. All data stays in ~/.sigrank/ until you run sigrank-agent publish.
• You control the keypair. Signing is off by default. Enable it when you're ready: sigrank-agent config signing_enabled true.
• Open source. The full agent source is at github.com/SunrisesIllNeverSee/sigrank-agent.

---

Links

• Leaderboard: signalaf.com
• Terminal client (view + compare): sigrank-mcp on npm
• PyPI: pypi.org/project/sigrank-agent
• Source: github.com/SunrisesIllNeverSee/sigrank-agent

---

License

CC-BY-NC-4.0 — © 2026 Deric J. McHenry / Ello Cello LLC