# RemakeBench Subscription Capacity Index — Kimi Allegretto v2

Status: **Provisional measurement**
Receipt received: **July 20, 2026 at 20:13:56 UTC**
Record: **Kimi Allegretto · Kimi K3 · max reasoning**
Supersedes: **v1 for the active presentation; immutable v1 remains published unchanged**

## What changed in v2

V2 attaches a durable, metadata-stripped first-party quota receipt. The receipt displays total usage of 10.73% for the monthly-reset plan meter and Code usage of 51.17% for the seven-day meter. The 51.17% reading corroborates the rounded 51% weekly observation in v1; it does not replace v1's frozen weekly arithmetic.

The receipt is post-run evidence. No pre-run meter receipt was retained, so this record remains provisional and still depends on operator attestation for workload attribution.

## Receipt provenance and privacy

The operator submitted the first-party Kimi Allegretto quota screenshot to RemakeBench, which received it at 2026-07-20T20:13:56Z. That `receivedAt` value is the RemakeBench attachment receipt time, not a provider-signed capture timestamp. The provider UI does not display a capture time or reset timezone. The original filename and raw bytes are not published. For reproducibility, the v2 measurement records the raw source SHA-256, dimensions and received time.

The public artifact was decoded and losslessly re-encoded with Sharp 0.34.5, then every ancillary PNG chunk was removed. Only IHDR, IDAT and IEND chunks remain. A decoded RGBA pixel SHA-256 proves pixel identity across the private source and public artifact. The sanitized image was reviewed for credentials, personal identity, private URLs and local filesystem paths; none are visible or embedded.

## Weekly estimate retained from v1

- Recorded API-equivalent retail value: $39.9267894
- Frozen v1 weekly observation: 51%
- Estimated weekly capacity: $39.9267894 / 0.51 = $78.2878224
- Frozen v1 display-rounding interval: $77.5277464–$79.0629493
- Receipt corroboration: Code 51.17%

The weekly point and interval remain byte-for-byte grounded in v1. V2 records the more precise receipt reading only as corroboration.

## Monthly-cycle estimate added in v2

- Receipt observation: total usage 10.73%
- Estimated monthly-cycle capacity: $39.9267894 / 0.1073 = $372.1042815
- Display-rounding interval: $39.9267894 / 0.10735 through / 0.10725 = $371.9309679–$372.2777566

Calculations use integer fixed-point dollars at 10^-7 precision. Because the monthly meter displays two decimal places, the uncertainty interval treats 10.73% as a rounded value from 10.725% through 10.735%.

## Interpretation and limitations

- API-equivalent retail value is not Kimi's internal serving cost or a subscription invoice.
- Both capacity estimates are workload-specific linear extrapolations, not guaranteed allowances.
- The measured workload was extremely cache-heavy; other workload mixes can produce materially different API-equivalent capacity.
- This is one Allegretto plan, one account and one observed quota window. Reliability across recurring windows is not established.
- No pre-run meter screenshot was retained. The attached screenshot is a post-run receipt, so workload attribution still depends on operator attestation and displayed reset timing.
- The provider UI does not display a capture time or reset timezone. Receipt arrival time must not be presented as provider-signed capture time.
- One MacBook API request lacks a reconciled usage record; the nine-task public-ledger subtotal remains a lower bound.

Published v1 files remain immutable. Published v2 core files and its content-addressed artifact are also immutable; any correction must be a new evidence version.