Live Fire Strand · LF-08 · Training Simulator

THE
THROTTLE
operate inside
the limits.

Two clocks govern every Claude subscription: a 5-hour rolling window and a weekly cap — one shared bucket across chat, Claude Code, and Cowork. Most teams discover this on a Wednesday, mid-refactor, at 100%. THE THROTTLE trains the discipline before the wall does.

What
Read the doctrine, then launch the simulator to train limit discipline and CLAUDE.md craft
How
Scroll to learn the two-layer limit model, then click Launch the Simulator above
When
Before your first real Claude session — or after hitting a wall mid-week
Next
Simulator → AI Debrief → Gatekeeper (your real file) → Verifier
The Doctrine

Tokens are the meter.
Judgment is the throttle.

The limit system isn't an obstacle — it's a budget, and budgets reward operators. Context carried, tool calls fired, retries burned: everything spends. THE THROTTLE compresses months of expensive lessons into forty minutes of scored, hash-chained decisions.

2
Stacked limits — a 5-hour rolling window and a weekly cap. Hit either, you stop.
1
Shared bucket across Claude.ai, Claude Code & Cowork. Burn in one, lose in all.
4
Scored drills — a live sprint, conviction wagers, a triage, and a full file audit.
480
Attainable points, sealed by SHA-256 into a tamper-evident session ledger.
Two Wings

The Meter. The Manifest.

Half the burn happens in the moment — the model you pick, the chat you refuse to close. The other half is decided before you type a word, by the CLAUDE.md loaded into every session. One simulator, both halves.

WING I

The Meter

Operate a full working day against live gauges. Learn the two-layer limit model as it stands after the May 2026 changes — and the eight decisions that decide whether your week survives you.

  • /clear vs /compact — the rule pair most operators get backwards
  • Model routing as a budget decision, not a preference
  • Projects as cached knowledge vs the daily re-upload tax
  • Why CI pipelines get API keys, never your seat
WING II

The Manifest

Build a CLAUDE.md that earns its context cost. The hierarchy, the @import system, the rule of admission — then prove it under review, line by flagged line.

  • Enterprise → project → user → nested: what loads, and when
  • KEEP / MOVE / CUT — triage ten candidate lines
  • Audit a shipped file with seven planted violations
  • Leave with the Manifest Blueprint, annotated and copyable
The Drills

Scored. Staked. Sealed.

DRILL 01 · THE SPRINT

One working day, one meter

Eight sequential decisions drain the HUD gauges in real time. Finish with fuel — or watch the weekly cap die on a Tuesday and read the verdict.

80 pts attainable
DRILL 02 · CONVICTION WAGERS

How sure are you?

Five claims about the limit system. Stake 10, 25, or 50 before answering. The chain records wins and losses with equal indifference.

250 pts attainable
DRILL 03 · THE TRIAGE

Ten lines, three fates

KEEP in CLAUDE.md, MOVE to a skill or @import, or CUT entirely. The editorial judgment that separates a manifest from a landfill.

80 pts attainable
DRILL 04 · THE AUDIT

Judgment, line by line

A shipped CLAUDE.md with seven violations — a live secret, a pasted spec, a rule written in anger. Correct flags pay +10. False flags cost −5.

70 pts attainable
The Chain

Every decision, hashed forward.

Each scored event is SHA-256 chained to the one before it. Alter any entry and every hash after it breaks — the same tamper-evidence discipline an EU AI Act Article 4 training record deserves. The session closes with a Certificate of Judgment sealed by the final hash.

initializing chain…
Conclusion · The Debrief

THE EXAMINER will see you now.

Finish a session, bring your certificate numbers here. The Studio's debrief officer reads the record and returns a verdict: what held, what bled, and the three prescriptions that change your next hundred sessions. Arriving from the simulator? Your numbers are already loaded.

Session Record

STUDIO DEBRIEF · CLASSIFICATION: CANDID
The Examiner is waiting on your record. Enter the session numbers — or arrive here directly from the simulator's close screen — and request the debrief. The verdict is generated live against your actual performance profile: fuel discipline, conviction calibration, editorial judgment, audit precision.