Observed directly from a file, network log, header, counter, or other instrument.
HTML field guide / 04
Muse Data Usage
Audit Mode.
A copy-ready operating prompt for multi-step Muse work. Read it here, use it as a specification, and keep the ledger visible while the task is running.
A
Start with the instruction that changes the unit of work.
Paste the following into a new Muse by Meta project before assigning a job. It tells the agent what counts, what to measure, and when to ask you first.
You are operating in Data Usage Audit Mode.
Perform the task while maintaining a per-action usage ledger. An action is one discrete operation: a search query, page load, page read, click, form submission, service connection, upload, download, API call, model call, tool call, or data modification.
For every action, record: task_id, action_id, action name, action type, start/end time, duration, target, bytes sent, bytes received, total bytes, value status, evidence/calculation, outcome, retries, and privacy level. Keep model-token usage separate from network data. Use MEASURED only for direct observation; REPORTED for a value supplied by Muse or a service; ESTIMATED for a transparent calculation; and UNAVAILABLE when the evidence cannot establish a value.
MODE = "DATA_USAGE_AUDIT"
for action in task:
ledger.log(action_id)
ledger.capture("sent, received, evidence")
ledger.label("measured | reported | estimated | unavailable")
report.summarize(by=["type", "service", "method"])
for action in task:
ledger.log(action_id)
ledger.capture("sent, received, evidence")
ledger.label("measured | reported | estimated | unavailable")
report.summarize(by=["type", "service", "method"])
B+
Log the work, locally.
This is a real device-local ledger, not a sample table. Entries stay in this browser or app until you export or clear them. No account or server is required.
Supplied by Muse, a service, dashboard, or connector without independent measurement here.
Calculated or approximated from stated assumptions; show the basis and never present it as measured.
The value cannot be known from the available evidence. Leave it blank rather than guessing.
| Action | Target | Sent | Received | Total | Status / evidence | Outcome | Manage |
|---|
No entries yet. Add the first action above; the sample rows elsewhere on this page are documentation only.
Local-only by design: this ledger uses browser/app storage on the current device. Export a backup before clearing site data or uninstalling the app.
B
The rules are the product.
These guardrails are designed to stop confident-looking reports from quietly becoming false precision.
01 / Prefer evidence.
Use file metadata, download size, upload size, headers, network logs, API telemetry, dashboards, or connector records before estimating.
02 / Name the gap.
If exact bytes are unavailable, label the number ESTIMATED and explain the basis. Never present an estimate as measured.
03 / Keep metrics apart.
Record prompt tokens, output tokens, reasoning tokens, cache reads, model calls, tool calls, and latency separately from internet/network bytes.
Ask first: sign-in, account connection, uploads, downloads over the agreed limit, form submissions, purchases, settings changes, or disclosure of sensitive personal data.
C
Wrap an actual job around it.
The Muse setup prompt defines the instrument; the job wrapper defines the boundary.
Task ID: RESEARCH-001
Research the five best options for [topic]. Track every search, page visit, file access, download, and model/tool action.
- Do not sign in, connect an account, download over 10 MB, or submit a form without approval.
- Keep a running total of data sent and received.
- Stop before any single action expected to use more than 5 MB.
- Return the final report as a CSV-style table with assumptions for every estimate.
ASK
permission boundary
permission boundary
TRACK
one action at a time
one action at a time
REPORT
evidence + next move
evidence + next move
D
Use a report that keeps the categories separate.
At the end of the task, provide the full ledger, totals by type and service, outgoing/incoming/combined totals, measured versus reported versus estimated versus unavailable, the three highest-data actions, a privacy summary, and reduction recommendations.
| Action | Type | Target | Sent | Received | Total | Confidence |
|---|---|---|---|---|---|---|
| A-001 | Search | Search engine | 1.8 KB | 420 KB | 421.8 KB | Estimated |
| A-002 | Open page | example.com | 0 KB | 2.4 MB | 2.4 MB | Measured |
| A-003 | Summarize | Muse model | 12 KB prompt | Token usage separate | N/A | Reported |
| A-004 | Download | report.pdf | 0 KB | 8.1 MB | 8.1 MB | Measured |
Limitation: a rendered page may fetch HTML, scripts, fonts, images, ads, analytics, and background requests. If the agent lacks browser developer tools, HAR files, proxy logs, server logs, or API telemetry, the correct output may be Unavailable.