Keycutter

One spec, every language — context-aware credential architecture for generating and verifying structured tokens with byte-identical layout and CRC32 tail checksums.

[SystemIdentifier]_[EnvironmentIdentifier]_[DomainPurposeIdentifier]_(Timestamp_)[Entropy][Checksum]

| Segment | Content | Width | |------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|----------------| | Prefix | System, environment, domain purpose identifiers, optional timestamp, delimited by _Restricts every semantic component to lowercase ASCII letters and decimal digits only. Underscores, uppercase letters, and any multi-byte non-ASCII characters are strictly forbidden.BASE36_CHARSET = 0123456789abcdefghijklmnopqrstuvwxyz | Variable (≥ 6) | | High-intensity Entropy | 24 Base62 charactersStrictly the 62-character GMP dictionary, ordered by ASCII code in ascending order. Third-party Base62 dialects that embed internal permutations are prohibitedBASE62_CHARSET = 0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz | 24 | | Tail Checksum | 6 Base62 characters encoding CRC32-IEEE of Prefix + Entropy | 6 |

Physical Seamless Fusion — No separator is permitted between the high-intensity entropy and the tail checksum. The full-stack slice cursor operates as a hard asymmetric dead-lock: the last 6 characters are always the checksum, everything before them is always the base string.

Absolute Unambiguous Prefix — Each semantic component of the prefix is subject to strict character-set constraints that fundamentally eliminate cross-platform parsing ambiguity and the hash inconsistency introduced by Unicode normalization (NFC/NFD).

Unbiased Uniform Sampling — Direct modulo on a low-bit-width integer is prohibited across all platforms. Every random index must be drawn from an OS-level cryptographically secure random number generator (CSPRNG) via rejection sampling, completely eliminating modulo bias.

Structured Threshold Assertion — Perimeter length guards discard hard-coded magic numbers. Minimum valid lengths are derived dynamically from the credential's own topology formula.

Self-Describing Key Versioning (optional) — An optional monotonic Unix-seconds timestamp embeds each credential's own creation instant, imposing a natural total order over key versions so that rotation, age-based expiry, and which-key-is-newer all resolve from the tokens alone, with no central registry lookup.

🚀 Quick Start

Go

go get go.leoweyr.com/keycutter/go/v2

Nodejs

npm install keycutter --save

🏗️ Generation Pipeline

1. Prefix Construction

Concatenate plaintext semantic segments for the given use case. Prefix length is variable and determined by system design, but must conform to a fixed topology.

Example: odc_prod_msk

Render the current Unix epoch seconds in Base36 and append it as a fourth prefix component after the domain purpose identifier. Base36 keeps every digit inside BASE36_CHARSET, and it is variable-width — 6 characters through late 2038, then 7.

[!WARNING]

Embed it only for keys inside a trust boundary, since the creation instant is readable by any holder.

$$ \text{Timestamp} = \text{Base36}(\text{UnixSeconds}) $$

Example: odc_prod_msk_tgny7c

2. Unbiased Entropy Generation

Draw 24 characters from BASE62_CHARSET using an OS-level CSPRNG with a rejection-sampling loop to ensure each index is drawn uniformly from [0, 61]. Direct modulo on a raw random integer is forbidden — it introduces a statistical skew that taints the distribution. The output is a strictly fixed 24-character string.

Example: 7xT2zP9qL4wK1mN8vV5cB3nA

3. Base String Concatenation

Concatenate the prefix from step 1 and the high-intensity entropy string from step 2 directly.

$$ \text{BaseString} = \text{Prefix} + \text{HighIntensityEntropy} $$

Example (no timestamp): odc_prod_msk_7xT2zP9qL4wK1mN8vV5cB3nA

Example (with timestamp): odc_prod_msk_tgny7c_7xT2zP9qL4wK1mN8vV5cB3nA

4. Mathematical CRC32 Mapping

Compute the CRC32-IEEE (reflected form) checksum of BaseString encoded as a UTF-8 byte stream, yielding a 32-bit unsigned integer $\text{Value}$. Convert $\text{Value}$ to a 6-character Base62 string $\text{Checksum}$ using the following right-to-left modulo loop:

$$ \begin{array}{l} \text{for } i = 5 \rightarrow 0: \ \quad \text{Remainder} = \text{Value} \bmod 62 \ \quad \text{Checksum}[i] = \text{Base62Charset}[\text{Remainder}] \ \quad \text{Value} = \left\lfloor \dfrac{\text{Value}}{62} \right\rfloor \end{array} $$

The loop fills from the last position backward. Any value that does not require all 6 digits is naturally zero-padded at the front — no explicit padding logic is needed.

5. Final Assembly

Append the 6-character checksum directly to the end of the base string with no separator:

$$ \text{Token} = \text{BaseString} + \text{TailChecksum} $$

Example: odc_prod_msk_7xT2zP9qL4wK1mN8vV5cB3nA4VHrHM

🛡️ Validation Pipeline

1. Structural Guard & Asymmetric Slice

At every network edge gateway or application entry point, perform hard physical boundary assertions before any business logic runs.

Length check:

$$ \text{MinLength} = \text{Len}(\text{Prefix}) + 24 + 6 $$

If the validator holds a known expected Prefix, the token length must equal exactly $\text{len}(\text{Prefix}) + 30$. Without a known prefix, the absolute minimum token length is 36 characters (each of the three prefix components must be at least 1 character, so the shortest valid prefix is 6 characters). Any token shorter than the derived threshold is discarded immediately.

Atomic slice, ignoring internal underscore structure, using the fixed-width checksum cursor:

$$ \begin{matrix} \text{BaseString} = \text{Token}[0:\text{len}(\text{Token}) - 6] \ \text{ProvidedTailCheckSum} = \text{Token}[\text{len}(\text{Token}) - 6:\text{len}(\text{Token})] \end{matrix} $$

2. Idempotent Verification

Re-execute the generation pipeline step 4 mapping locally on the extracted BaseString to obtain ExpectedTailChecksum. If ExpectedChecksum ≠ ProvidedChecksum, the token is rejected immediately as corrupted or truncated — fail-fast, no further processing.

3. Context Reification

Only after passing step 2 may the system split the prefix portion of BaseString on _. Because BASE36_CHARSET forbids underscores within any component, the split result is uniquely deterministic across every platform and encoding. Extract the system, environment, and domain purpose identifiers and inject them as a typed security context object into downstream operations.

The prefix component count determines whether an optional timestamp is present: the validator is never told in advance, three components carry no timestamp, four mark the trailing one as the Base36 timestamp.

⚖️ Why 24 Characters

On the cryptographic side, 24 Base62 characters carry:

$$ 62^{24} \approx 2^{143} \text{ bits} $$

The practical security floor for API tokens in cloud-native architecture is 128 bits. 24 characters delivers 143 bits — clearing the threshold with margin to spare.

Some vendors push further: GitHub's personal access tokens use 30 entropy characters, reaching ~178 bits. From a pure cryptographic standpoint, that number is unimpeachable. From an engineering leverage standpoint, it is unnecessary — beyond ~140 bits, the marginal security return per additional character converges to zero. The cost, however, is real. Every extra character widens network payloads, inflates database index pages, and — on mobile — turns a token into a string too long to select cleanly with a long-press.

Keycutter locks high-intensity entropy at 24 characters: the precise point where cryptographic surplus meets transmission efficiency and human ergonomics, with nothing wasted on either side.