MModelapseAI Model Test & Evolution Archive
Archive/Research

ARCHIVE v0.26 · PUBLIC RESEARCH

Explore the evidence, not a leaderboard.

Inspect sealed first-party and documented execution records across Providers. Every result links to its exact Model, Test Case and Run. Filters select records, not winners, and calibration canaries are excluded.

RESEARCH FILTERS

Build a reproducible evidence slice

newest completed first · 20 per page
Filters and cursor live in the shareable URL. Each query is bounded and sorted by completed timestamp + Run ID.

Clear filters · Missing access, pricing or qualification evidence is not silently interpreted as free or normal. Research slices are not statistical samples.

ARCHIVE v0.27 · CURATED RESEARCH

Save an immutable collection

Browse saved collections →

Research filters above are live and can find new Runs. Saving creates a public immutable snapshot of the current matching Run IDs (1–50 only). Capture requires operator access. Empty or oversized slices are rejected rather than silently truncated. Collections never include calibration Runs.

Operator · capture this filtered slice
Saved manifests store no Provider credentials or account identifiers.

ARCHIVE v0.28 · OBSERVED FACETS

Evidence-backed filter counts

2 matching Runs · all pages

Each count represents sealed public non-calibration Runs satisfying all current filters, not just this page. Facets count archival observations, not unique models or a quality score. Up to 25 values are displayed per facet, with overflow explicitly marked. Unknown context is never inferred.

Region1 visible value(s)
unknown / not recorded (2)
Account tier1 visible value(s)
unknown / not recorded (2)
Service tier1 visible value(s)
unknown / not recorded (2)
Assurance (read-only)1 visible value(s)
unknown / not recorded (2)

SEALED PUBLIC RECORDS

Observed Runs

2 record(s) on this page

Public non-calibration, sealed completed Runs only. No composite score, winning order, or quality inference. A newer record is not necessarily a better or more representative result.

DeepSeek · DeepSeek Flash2026-09-24 11:30:47.704Z · 4c56f78a
exact matchEvidence E4

modelapse-direct-smoke · exact-modelapse · v1.0.0

first_party_direct · region unknown · account unknown · service unknown · assurance unknown

Provider access unknown · Runner access unknown · Calibration eligibility: not evaluated in this research index

Native cost estimate: unknown (not free) · pricing source missing

Qualification: qualification_missing · Cost: cost_evidence_missing

Run evidence →Model →Test →History →Compare same Test →
DeepSeek · DeepSeek Flash2026-09-24 08:48:26.999Z · f702d742
evaluation unavailableEvidence E4

modelapse-direct-smoke · exact-modelapse · v1.0.0

first_party_direct · region unknown · account unknown · service unknown · assurance unknown

Provider access unknown · Runner access unknown · Calibration eligibility: not evaluated in this research index

Native cost estimate: unknown (not free) · pricing source missing

Qualification: qualification_missing · Cost: cost_evidence_missing

Run evidence →Model →Test →History →Compare same Test →
Records are chronological. Pagination only traverses older Run IDs; the cursor is not an eligibility or comparability verdict.End of this slice