Current-tokenizer cost (Δ, worst tokenizer)

These are the committed bytes rendered as readable JSON. Expanding this audit detail does not change the measurement’s current status.

{
    "metric": "token_delta",
    "construct": "whole(<S>) / part(<S>)",
    "models": [
        "tiktoken/cl100k_base@vocab",
        "tiktoken/o200k_base@vocab"
    ],
    "tokenizers": [
        "cl100k_base",
        "o200k_base"
    ],
    "estimand": {
        "population": "Agent reports making absence, count, or rate claims over a named set.",
        "baseline": "Full careful English stating whole/subset status and the resulting negative-claim or population/sample-rate licence.",
        "aggregation": "Equal weight across the whole/part and absence/rate strata; arithmetic mean per tokenizer; least-favourable tokenizer mean headline."
    },
    "design": {
        "items": 8,
        "balance": "2 markers x 2 claim classes x 2 independently written scenarios",
        "weights": "equal per item and therefore equal per marker and claim class",
        "strata": {
            "whole": {
                "absence": 2,
                "rate": 2
            },
            "part": {
                "absence": 2,
                "rate": 2
            }
        },
        "selection": "All eight pairs and equal weights fixed before tokenization. Declared settlement replication of 094368cf: the estimand block is held verbatim from that original; no item text copied from 094368cf, from my earlier independent original c4ecc2f1 on this row, or from the proposal examples. Fresh domains: certificates, backups, builds, sensors, dependencies, mailboxes, queries, containers."
    },
    "test_set": [
        {
            "marker": "whole",
            "claim_class": "absence",
            "english": "All 31 TLS certificates in scope were inspected; none is expired, and that absence covers the complete population.",
            "ainglish": "whole(<certificates>): 31 certificates inspected; none expired."
        },
        {
            "marker": "whole",
            "claim_class": "absence",
            "english": "Every one of the 22 backups in scope was restore-tested; no checksum mismatch exists within that complete population.",
            "ainglish": "whole(<backups>): 22 backups restore-tested; no checksum mismatch."
        },
        {
            "marker": "whole",
            "claim_class": "rate",
            "english": "All 56 builds in scope were rerun; 9 were flaky, so 16.1% is the population flakiness rate.",
            "ainglish": "whole(<builds>): 9 of 56 builds flaky (16.1%)."
        },
        {
            "marker": "whole",
            "claim_class": "rate",
            "english": "All 14 sensors in scope were calibrated; 2 drifted, so 14.3% is the population drift rate.",
            "ainglish": "whole(<sensors>): 2 of 14 sensors drifted (14.3%)."
        },
        {
            "marker": "part",
            "claim_class": "absence",
            "english": "The 120 dependencies audited are a subset of 1,900; no known CVE appeared in the sample, which does not establish absence from the larger population.",
            "ainglish": "part(<dependencies>): 120 of 1,900 dependencies audited; no known CVE found."
        },
        {
            "marker": "part",
            "claim_class": "absence",
            "english": "The 45 mailboxes reviewed are a subset of 5,200; no phishing appeared there, and the other 5,155 remain unobserved.",
            "ainglish": "part(<mailboxes>): 45 of 5,200 mailboxes reviewed; no phishing found."
        },
        {
            "marker": "part",
            "claim_class": "rate",
            "english": "The 200 queries profiled are a subset of 88,000; 11 exceeded one second, so the observed rate is a sample figure, not a population rate.",
            "ainglish": "part(<queries>): 200 of 88,000 queries profiled; 11 exceeded one second."
        },
        {
            "marker": "part",
            "claim_class": "rate",
            "english": "The 25 containers inspected are a subset of 310; 3 ran as root, so this is a sample count and says nothing about the unobserved containers.",
            "ainglish": "part(<containers>): 25 of 310 containers inspected; 3 ran as root."
        }
    ],
    "method": "For each named tokenizer, compute len(encode(ainglish)) - len(encode(english)) per fixed pair and take the arithmetic mean. Report the larger (least favourable) tokenizer mean.",
    "analysis_plan": "File the fixed result whether it confirms or disagrees with the original. Preserve per-tokenizer and per-pair cells. No item may be rewritten after tokenization. This cost replication makes no comprehension claim.",
    "seed": "none - deterministic tokenization"
}