Acompleted
Saved artifact
3/8checks passed
$0.00000 tokens
unknown
NEW MODEL. OLD WORK. FRESH EVIDENCE.
See what actually
got better.
Hand-authored import demo · no model performance claim
Source: Historical model authorship is unverified; source provenance excluded from this summary.
THE RESULT
Regression detected in at least one candidate
Imported outputs, evaluated locally. Model labels and generation usage are owner-reported and unverified. Zero local model calls does not mean external generation was free. This measures artifact behavior, not model superiority.
2candidate workflowsInspect every comparison
E1completed
Imported candidate 1
8/8checks passed
$0.00000 tokens
Model unverified
E2completed
Imported candidate 2
2/8checks passed
$0.00000 tokens
Model unverified
Every comparison
| Versions | Measured result | New passes | Regressions |
|---|---|---|---|
| A → E1 | Measured improvement | 5 | 0 |
| A → E2 | Regression detected | 0 | 1 |
| E1 → E2 | Regression detected | 0 | 6 |
Each row reads left → right. Any lost pass takes precedence over new passes. One sample per workflow; there is no automatic winner or adoption.
EXTERNAL GENERATION COST
$0.0000
0reported tokens, including cache reads/writes
External generation cost and usage are owner-reported, unverified; local assessment makes no model calls
Original intent Preserved before the revision
Let a product team search feedback titles or tags without caring about case or surrounding whitespace, combine search with priority filters, and export exactly the visible items as valid CSV that preserves the original fields. Counts and reset must remain truthful.
Explicitly included by the exporter · Intent confirmed: True
01 / BEHAVIOR
The same checks. Every version.
| Frozen acceptance check | A | E1 | E2 |
|---|---|---|---|
| Check 1 | PASS | PASS | PASS |
| Check 2 | PASS | PASS | FAIL |
| Check 3 | FAIL | PASS | FAIL |
| Check 4 | FAIL | PASS | FAIL |
| Check 5 | FAIL | PASS | FAIL |
| Check 6HOLDOUT | FAIL | PASS | FAIL |
| Check 7 | PASS | PASS | PASS |
| Check 8HOLDOUT | FAIL | PASS | FAIL |
Check definitions are omitted from the exported request. External generation context and timing are unverified. Holdout cases require inputs unavailable in shared source/data; ordinary visible inputs are not holdouts. Timings include navigation, actions and assertion waits; these are not performance benchmarks. Hover a result for failure details.
02 / VISIBLE CHANGE
Look at the same moment.
Same viewport for each corresponding check; 1200 × 820 by default. Per-check dimensions are in the evidence JSON. Click any image to inspect it. Visual differences are evidence, not an automatic quality score.
Evidence included in this export.
Run identity, costs and comparison evidence
{
"status": "completed",
"manifest": {
"input_result_sha256": "95cb831cdcc449247786b10e7f96ef0d7c1d8b456d571f2363bd93f0d858338e",
"source": "d161974a58956fd915b98dafb0c4924aded8be892d0b34d433e083ebb878ffd0",
"checks": "66b63820c06e197d34f2876a108d07c36ddbd2c418c550cc089344e291e5f08f"
},
"fingerprint": null,
"source_unchanged": true,
"source_context": null,
"model_calls": 0,
"cost": {
"total_usd": 0,
"limit_usd": null,
"known_spend_usd": 0,
"basis": "External generation cost and usage are owner-reported, unverified; local assessment makes no model calls"
},
"comparisons": {
"A_E1": {
"verdict": "improved",
"improvements": [
"check-3",
"check-4",
"check-5",
"check-6",
"check-8"
],
"regressions": []
},
"A_E2": {
"verdict": "regression",
"improvements": [],
"regressions": [
"check-2"
]
},
"E1_E2": {
"verdict": "regression",
"improvements": [],
"regressions": [
"check-2",
"check-3",
"check-4",
"check-5",
"check-6",
"check-8"
]
}
}
}Exported summary JSON · Source, raw prompts/receipts, check definitions, private labels and paths are excluded. Metrics are reported from the input result; this export is not independent verification.