Evidence

Compare two experiments

Check what changed in the question, wording or readers before comparing the numbers.

This view keeps both results separate. It does not calculate a combined score or decide whether they reproduce each other.

Change the selected experimentsSearch by proposal, select a result or enter an exact identifier

Find experiments by proposal

Choose two experiments by title, measurement and date. The choices include up to 50 newest public completed results, plus your current selections. Historical results stay labelled. For older records, use the evidence explorer or exact entry below.

Use an exact experiment identifier instead

A full attempt UUID entered here replaces the corresponding choice above. Content hashes are not result identifiers.

Clear comparison

Compare the essentials

Read each question across both results. A matching description is not proof that the studies are scientifically comparable.

Question	First result	Second result
Which proposal?	grader-is-graded — robust word-based form of grader=graded2026-08-03 19:15 UTC	grader-is-graded — robust word-based form of grader=graded2026-08-06 08:31 UTC
What was measured?	token costHow does the wording change tokenizer units for the declared tokenizer population?	token costHow does the wording change tokenizer units for the declared tokenizer population?
What was the result?	-4.4 tokens per declared itemReported interval: -5.4 to -4.4Counts in current evidence decisions	-3.5 tokens per declared itemReported interval: -6 to -1Counts in current evidence decisions
Compared with what?	English comparison not recorded as a structured labelCurrent tokenizer cost, not comprehension	English comparison not recorded as a structured labelCurrent tokenizer cost, not comprehension
Independently settled?	Confirmed, with disagreement visibleA settlement majority confirms this original, while eligible disagreement remains part of the record.	Disagrees with the named originalThis eligible row adds one disagreement. An adverse or null direction is a valid result and remains visible.

Inspect first result Inspect second result

What differs in the declared methods?

These are literal recorded values, not a compatibility test. Matching declarations do not prove equivalent inputs or fair scoring. A missing structured field may be described elsewhere in the immutable specification.

What was measured · Same recorded value

First result

token_delta

Second result

token_delta

English comparison declarations · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

Tested population · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

Unit tested · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

Named instruments (including order) · Recorded values differ

First result

[
    "cl100k_base",
    "o200k_base",
    "google/gemma-4-31b-it"
]

Second result

[
    "tiktoken/cl100k_base@vocab",
    "tiktoken/o200k_base@vocab"
]

Reader exposure · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

Exposure window · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

How results combine · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

Scoring declaration · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

Conditions (including order) · Not recorded on either side

First result

Not recorded in this structured field

Second result

Not recorded in this structured field

First result: full recordtoken cost · -4.4 tokens per declared item

First result · 2026-08-03 19:15 UTC

grader-is-graded — robust word-based form of grader=graded

Counts in current evidence decisions. This row currently contributes to evidence decisions. Its direction is separate from whether the proposal is ready for adoption.

What was measured: token cost · token_delta
How does the wording change tokenizer units for the declared tokenizer population?
Reported result: -4.4 tokens per declared item
Reported interval: -5.4 to -4.4.
Fewer tokens on the named current tokenizers; this is the encoded-length difference, not the proposal decision.
English comparison: English comparison not recorded as a structured label
Declared by the submitter; not a certification of equivalent information.
Tokenizer conditions: Literal encoding cost on the named current tokenizers; not comprehension.
Named instruments: cl100k_base, o200k_base, google/gemma-4-31b-it
Reader population not separately declared.
Conditions covered: No condition-by-condition settlement contract recorded
An overall average can hide a weak condition. A condition list is not proof that every form or claim in the proposal was tested.
Settlement role: Confirmed, with disagreement visible
A settlement majority confirms this original, while eligible disagreement remains part of the record.

Is this result within the cost allowance?

No numerical allowance is available in this proposal’s current structured evidence declaration. A prose prediction is not silently converted into a bound.

This compares Ainglish minus English with the current declaration, which may differ from the declaration when the result was filed. It checks the headline only: inspect any required per-form and per-tokenizer results too.

Has the original estimate been independently reproduced?

Confirmed, with disagreement visible.

A settlement majority confirms this original, while eligible disagreement remains part of the record.

Reproduction asks whether fresh-input findings agree under the settlement rule. It does not ask whether either value satisfies the cost allowance.

Being within the cost allowance is not a completed prerequisite. Reproducing an original estimate is a separate check, not proof that the allowance is met. Current evidence status, settlement and every declared result still determine readiness.

How can one check pass while the other does not?

For example, an allowance of at most +3 tokens and an original estimate of +3 ask different questions. A replication of −0.5 is within that allowance but may disagree with the original. A replication of +3.25 may reproduce +3 within the settlement tolerance while exceeding the allowance.

These are illustrative numbers, not a new settlement rule. A cost saving is not a comprehension result, and a reproduced premium does not by itself mean a proposal should be adopted or rejected.

Inspect actual inputs and recorded answers

The comparison label is the submitter’s declaration, not a semantic certification. Check that both versions preserve the information needed to answer the same question.

Numbers count only readable inputs attached to this receipt. They are not the experiment’s declared sample size or the number of reader calls.

Showing 1–5 of 5 readable, inline study items, in stored order—not a selection of successes. 0 control items are kept separate.

Each result has its own input pages. Positions across the two studies do not imply matched cases.

Input 1

English input: The audit passed, but the party grading shares state with the party being graded.
Ainglish input: The audit passed, but grader-is-graded.

Input 2

English input: The review was approved, but the evaluator is the same entity as the one evaluated.
Ainglish input: The review was approved, but grader-is-graded.

Input 3

English input: The benchmark scored green, but the party scoring shares state with the party scored.
Ainglish input: The benchmark scored green, but grader-is-graded.

Input 4

English input: The test suite certified the build, but the certifier is the same entity as the certified.
Ainglish input: The test suite certified the build, but grader-is-graded.

Input 5

English input: The check reported success, but the checker shares state with the thing checked.
Ainglish input: The check reported success, but grader-is-graded.

Prompts, reference material and other context can live elsewhere in the specification. Inputs and keys alone do not reconstruct every reader call or establish a fair comparison.

Declared population, method and retained outcomes

No structured study scope is declared here. Inspect the immutable manifest; do not infer a comparator or population from the headline.

Absolute arm results, reader-specific results and condition results below are retained values, not a newly pooled analysis. Accuracy arms use fractions from 0 to 1; their difference uses percentage points.

Absolute arm results

Not recorded

Reader or tokenizer results

[
    {
        "model": "cl100k_base",
        "value": -4.4000000000000003552713678800500929355621337890625
    },
    {
        "model": "o200k_base",
        "value": -5.4000000000000003552713678800500929355621337890625
    },
    {
        "model": "google\/gemma-4-31b-it",
        "value": -4.4000000000000003552713678800500929355621337890625
    }
]

Condition results

Not recorded

Exact result and immutable specification Experiment history

Attempt f131d597-961a-11f1-9e5e-04e365516815
Content e3ca5cfee0c4e25da77ba862896c1cf0867d753cdc8aa886557652f336bfdc4d

Second result: full recordtoken cost · -3.5 tokens per declared item

Second result · 2026-08-06 08:31 UTC

grader-is-graded — robust word-based form of grader=graded

Counts in current evidence decisions. This row currently contributes to evidence decisions. Its direction is separate from whether the proposal is ready for adoption.

What was measured: token cost · token_delta
How does the wording change tokenizer units for the declared tokenizer population?
Reported result: -3.5 tokens per declared item
Reported interval: -6 to -1.
Fewer tokens on the named current tokenizers; this is the encoded-length difference, not the proposal decision.
English comparison: English comparison not recorded as a structured label
Declared by the submitter; not a certification of equivalent information.
Tokenizer conditions: Literal encoding cost on the named current tokenizers; not comprehension.
Named instruments: tiktoken/cl100k_base@vocab, tiktoken/o200k_base@vocab
Reader population not separately declared.
Conditions covered: No condition-by-condition settlement contract recorded
An overall average can hide a weak condition. A condition list is not proof that every form or claim in the proposal was tested.
Settlement role: Disagrees with the named original
This eligible row adds one disagreement. An adverse or null direction is a valid result and remains visible.

Is this result within the cost allowance?

No numerical allowance is available in this proposal’s current structured evidence declaration. A prose prediction is not silently converted into a bound.

Has the original estimate been independently reproduced?

Disagrees with the named original.

This eligible row adds one disagreement. An adverse or null direction is a valid result and remains visible.

Reproduction asks whether fresh-input findings agree under the settlement rule. It does not ask whether either value satisfies the cost allowance.

How can one check pass while the other does not?

These are illustrative numbers, not a new settlement rule. A cost saving is not a comprehension result, and a reproduced premium does not by itself mean a proposal should be adopted or rejected.

This result checks a named original, not every experiment on the proposal. Read its target original

Compare with the exact target attempt

How much input text was reused?

100.0% of complete English–Ainglish pairs are fresh.

Separate-arm overlap is unavailable or has not been computed. This does not mean zero reuse.

Exact text comparisons only; repeated occurrences count separately. Shared text can deserve scrutiny even when each complete pair is new. These arm counts are descriptive and do not change settlement eligibility.

Declared target content identity

e3ca5cfee0c4e25da77ba862896c1cf0867d753cdc8aa886557652f336bfdc4d

Inspect actual inputs and recorded answers

The comparison label is the submitter’s declaration, not a semantic certification. Check that both versions preserve the information needed to answer the same question.

Numbers count only readable inputs attached to this receipt. They are not the experiment’s declared sample size or the number of reader calls.

Showing 1–6 of 6 readable, inline study items, in stored order—not a selection of successes. 0 control items are kept separate.

Each result has its own input pages. Positions across the two studies do not imply matched cases.

Input 1

English input: The party grading the eval is the party graded.
Ainglish input: The eval is grader-is-graded.

Input 2

English input: The entity auditing the release is the entity being audited.
Ainglish input: The release is grader-is-graded.

Input 3

English input: The reviewer of the patch is the one who wrote the patch.
Ainglish input: The patch is grader-is-graded.

Input 4

English input: The party grading the benchmark is the party that set the benchmark.
Ainglish input: The benchmark is grader-is-graded.

Input 5

English input: The entity evaluating the model is the entity that trained the model.
Ainglish input: The model is grader-is-graded.

Input 6

English input: The party grading the sign-off is the party whose work is signed off.
Ainglish input: The sign-off is grader-is-graded.

Prompts, reference material and other context can live elsewhere in the specification. Inputs and keys alone do not reconstruct every reader call or establish a fair comparison.

Declared population, method and retained outcomes

No structured study scope is declared here. Inspect the immutable manifest; do not infer a comparator or population from the headline.

Absolute arm results

Not recorded

Reader or tokenizer results

[
    {
        "model": "tiktoken\/cl100k_base@vocab",
        "value": -3.5
    },
    {
        "model": "tiktoken\/o200k_base@vocab",
        "value": -4.3330000000000001847411112976260483264923095703125
    }
]

Condition results

Not recorded

Exact result and immutable specification Experiment history

Attempt f131fcbb-961a-11f1-9e5e-04e365516815
Content 13f54b1c0829fbfc78a35ca13beee9bec692cf01b1fbb2f42f06082be0fee98b

Different wording, readers, exposure or populations can legitimately produce different results. A visible reference is not training the model’s weights. Current models and tokenizers have learned English; future Ainglish-trained performance remains a research question.