Source: https://guthlabs.ai/docs/instruments/cpu-cache-hit-rate-amat-calculator/

instrument / verified-clean

# CPU Cache Hit-Rate / AMAT Calculator

The current ledger marks this instrument verified clean within the source-bound verification corpus.

Status: **VERIFIED_CLEAN**Lifecycle: **verified-clean**

## Inspectable records

[01**Overview**Purpose and lifecycle](<https://guthlabs.ai/docs/instruments/cpu-cache-hit-rate-amat-calculator/>)[02**Mechanism**Method and assumptions](<https://guthlabs.ai/docs/instruments/cpu-cache-hit-rate-amat-calculator/mechanism/>)[03**Implementation**Binding and validation](<https://guthlabs.ai/docs/instruments/cpu-cache-hit-rate-amat-calculator/implementation/>)[04**Qualification**Evidence and errata](<https://guthlabs.ai/docs/instruments/cpu-cache-hit-rate-amat-calculator/qualification/>)

## Ledger state

The current ledger marks this instrument verified clean within the source-bound verification corpus.

**Area**: Computing & Information

**Tranche**: tranche-4

**Evidence ID**: t4-288

**Ledger rows**: 1

No ledger note is attached.

## Review findings

Core AMAT arithmetic is valid under the stated unit and miss-penalty conventions. The instrument cannot honestly claim to output effective speedup until it defines the speedup formula and adds an explicit no-cache baseline input. It also needs an unambiguous multilevel parameter convention, validation for miss-rate bounds and level count, and a defined terminal-memory penalty.

### Formula and scope findings

- The AMAT formula AMAT = hit_time + miss_rate × miss_penalty is standard for a single cache level when miss_penalty means the additional time after a miss, excluding hit_time.

- The multilevel formula AMAT_L1 = hit_L1 + miss_L1 × AMAT_L2 is standard when AMAT_L2 includes all lower-level access time and when each level's hit time is charged only when that level is reached.

- The recursive definition for levels beyond L2 is not specified explicitly, including how the final-level memory-access penalty is represented.

- The claimed effective speedup versus a no-cache baseline has no formula and no no-cache-baseline input, so that output cannot be computed or verified.

- The term miss_penalty is ambiguous for multilevel inputs: treating it as inclusive of lower-level access time would double-count time in the stated recursive formula; treating it as exclusive makes the formula valid.

### Unit assumptions

- All hit times, miss penalties, and recursively computed AMAT values must use the same time unit, such as nanoseconds.

- Miss rates must be supplied as fractions from 0 to 1, not percentages from 0 to 100; for example, 5% must be entered as 0.05.

- If timing is given in processor cycles, a clock frequency or cycle-time conversion is required before combining it with nanosecond values.

- Hit times and miss penalties must be nonnegative.

## Boundary hardening

**BREAK_CLAIMED** · SELF_FLAGGED

The extreme-magnitude vector overflows finite JavaScript arithmetic to Infinity, and the convention probe produces radically different results unless percentage versus fractional miss-rate input is explicitly disambiguated.

## Errata and corrections

No applied erratum is attached to this unique-name record.

## Inspectable example

4 review vectors are recorded. One source-bound sample is shown below.

````
{
  "inputs": {
    "levels": 1,
    "hit_time_ns": 1,
    "miss_rate": 0.05,
    "miss_penalty_ns": 80
  },
  "expected_outputs": {
    "amat_ns": 5,
    "effective_speedup_vs_no_cache": null
  },
  "working": "AMAT = hit_time + miss_rate × miss_penalty\n= 1 ns + 0.05 × 80 ns\n= 1 ns + 4 ns\n= 5 ns.\nSpeedup is unavailable because no no-cache baseline is specified."
}
````

## Related workflows

No source-defined multi-instrument workflow membership is recorded.

## Sources and evidence

These pointers bind the page to evidence. They do not grant authority or collapse verification into deployment.

- `VERIFIED_LEDGER.json · VERIFIED_CLEAN`

- `evidence-tranche4/tranche4-results.json`

- `BOUNDARY_HAZARD_CATALOG.json`