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",
    "formula_version": 1,
    "construct": "sanction-allow / sanction-penalize",
    "models": [
        "cl100k_base",
        "o200k_base",
        "p50k_base"
    ],
    "test_set": [
        {
            "form": "allow",
            "ainglish": "sanction-allow(financial-regulator): bank-7 acquires branch-2.",
            "english": "The financial regulator formally permitted bank 7 to acquire branch 2."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(safety-board): reactor-4 restarts after inspection.",
            "english": "The safety board formally permitted reactor 4 to restart after inspection."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(city-council): night-market opens in zone-c.",
            "english": "The city council formally permitted the night market to open in zone C."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(review-panel): team-red releases patch-81.",
            "english": "The review panel formally permitted Team Red to release patch 81."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(customs-office): shipment-552 enters region-west.",
            "english": "The customs office formally permitted shipment 552 to enter the western region."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(data-protection-officer): lab-3 exports dataset-q.",
            "english": "The data protection officer formally permitted Lab 3 to export dataset Q."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(aviation-authority): drone-19 flies above corridor-b.",
            "english": "The aviation authority formally permitted drone 19 to fly above corridor B."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(medical-board): clinic-8 runs trial-nova.",
            "english": "The medical board formally permitted Clinic 8 to run the Nova trial."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(repository-owner): maintainer-k merges change-904.",
            "english": "The repository owner formally permitted maintainer K to merge change 904."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(procurement-chair): vendor-12 submits bid-r.",
            "english": "The procurement chair formally permitted Vendor 12 to submit bid R."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(court-administrator): counsel-5 files exhibit-j under seal.",
            "english": "The court administrator formally permitted counsel 5 to file exhibit J under seal."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(network-operator): peer-77 announces route-6.",
            "english": "The network operator formally permitted peer 77 to announce route 6."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(ethics-committee): lab-aurora enrolls adult volunteers.",
            "english": "The ethics committee formally permitted Lab Aurora to enroll adult volunteers."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(treasury-office): fund-z transfers tranche-4.",
            "english": "The treasury office formally permitted Fund Z to transfer tranche 4."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(standards-board): maker-2 uses mark-x on model-7.",
            "english": "The standards board formally permitted Maker 2 to use mark X on model 7."
        },
        {
            "form": "allow",
            "ainglish": "sanction-allow(building-inspector): tenants occupy wing-d.",
            "english": "The building inspector formally permitted tenants to occupy wing D."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(financial-regulator): bank-9, transfers suspended for 30 days.",
            "english": "The financial regulator formally imposed a penalty on bank 9: transfers are suspended for 30 days."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(sports-league): player-14, fine of 500 credits.",
            "english": "The sports league formally imposed a penalty on player 14: a fine of 500 credits."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(university): lab-6, archive access restricted for one term.",
            "english": "The university formally imposed a penalty on Lab 6: archive access is restricted for one term."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(city-council): vendor-4, excluded from tender-21.",
            "english": "The city council formally imposed a penalty on Vendor 4: exclusion from tender 21."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(customs-office): carrier-8, inspection bond doubled.",
            "english": "The customs office formally imposed a penalty on carrier 8: its inspection bond was doubled."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(data-regulator): broker-3, processing frozen for seven days.",
            "english": "The data regulator formally imposed a penalty on Broker 3: processing is frozen for seven days."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(aviation-authority): airline-k, route-5 slot withdrawn.",
            "english": "The aviation authority formally imposed a penalty on Airline K: its route 5 slot was withdrawn."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(medical-board): clinic-2, new admissions paused.",
            "english": "The medical board formally imposed a penalty on Clinic 2: new admissions are paused."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(repository-owner): bot-r, write access revoked for 48 hours.",
            "english": "The repository owner formally imposed a penalty on bot R: write access is revoked for 48 hours."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(procurement-chair): supplier-m, score reduced by ten points.",
            "english": "The procurement chair formally imposed a penalty on Supplier M: its score was reduced by ten points."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(court): firm-17, filing privilege stayed until Friday.",
            "english": "The court formally imposed a penalty on Firm 17: its filing privilege is stayed until Friday."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(network-operator): peer-31, announcements rate-limited for six hours.",
            "english": "The network operator formally imposed a penalty on peer 31: announcements are rate-limited for six hours."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(ethics-committee): lab-comet, recruitment suspended pending review.",
            "english": "The ethics committee formally imposed a penalty on Lab Comet: recruitment is suspended pending review."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(treasury-office): fund-y, disbursements capped at 20 credits.",
            "english": "The treasury office formally imposed a penalty on Fund Y: disbursements are capped at 20 credits."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(standards-board): maker-8, certification mark withdrawn.",
            "english": "The standards board formally imposed a penalty on Maker 8: its certification mark was withdrawn."
        },
        {
            "form": "penalize",
            "ainglish": "sanction-penalize(building-inspector): landlord-5, occupancy permit restricted to floor-1.",
            "english": "The building inspector formally imposed a penalty on landlord 5: the occupancy permit is restricted to floor 1."
        }
    ],
    "seed": "none — deterministic tokenizer counts",
    "population": "32 fresh complete authority-and-act pairs, balanced 16 sanction-allow and 16 sanction-penalize; form labels are diagnostic only",
    "selection": "All authorities, targets, formal acts, complete renderings, and the 16/16 balance were authored before target-manifest inspection or tokenizer import. Each marker arm states an explicit resolved authority and bounded act or penalty; each comparator is the proposal's complete careful-English formal-permission or formal-penalty disclosure. Capability, prediction, informal tolerance, unexecuted penalty inference, authority validity, and legal-compliance claims are excluded. Exact pair-set disjointness from the target is checked before mint; no item is selected using token outcomes.",
    "method": "After stored-manifest mint, compute len(encode(ainglish))-len(encode(english)) for every pair and tokenizer. Average all 32 items equally for each tokenizer. Report the maximum tokenizer mean as the least-favourable aggregate headline and the tokenizer span as value_lo/value_hi. Form-specific means are descriptive only because the target original is aggregate-only. File every finite result once regardless of agreement, sign, or proposal consequence.",
    "estimand": {
        "population": "the 32 frozen target-disjoint formal-sanction pairs",
        "aggregation": "equal-item aggregate mean per tokenizer; headline is maximum tokenizer mean",
        "comparator": "complete careful English reporting formal permission or a formally imposed penalty by the same named authority",
        "comparator_class": "careful_expansion",
        "unit": "tokens per complete claim"
    },
    "environment": {
        "tiktoken": "0.13.0",
        "python": "3.12.3"
    },
    "freeze": "The final target-matched runtime and exact target-disjoint pair set are stored by the API before tokenizer import or count exposure.",
    "replicates_hash": "6ed658d542b3e97ce709660bde30c77bac87fa0ea3bc7c281917c672393ab6dc"
}