# TEP V8.1 Research Evidence

Status: LIVE in R2 as of 2026-08-03 from the merged `cubie-research` commit
below. Public Worker access is verified separately after deployment.

| Item | Value |
| --- | --- |
| Bucket | `trustfortress-resource-library` |
| Prefix | `cubie-research/tep-v8.1-evidence/2026-08-03/` |
| Owner | `iamdatanick/cubie-research` evidence; `cubie-tf` public claim consumer |
| Research commit | `bd51e5e094683e2cd0ab1db8bc7a02c4888e4cab` |
| Research PR | `https://github.com/iamdatanick/cubie-research/pull/41` |
| Reason outside Git | Publish the binary paper and compact receipt bundle without duplicating generated evidence in product Git history |

## Objects

| Object | Bytes | SHA-256 | Public URL |
| --- | ---: | --- | --- |
| `cubie-research-tep-evidence-bd51e5e.zip` | 474,662 | `613179f7c24afda40ea2237f4b9d9a72a07b65c71b069ae16761a2c641eb005b` | `https://lib.trustfortress.ai/objects/cubie-research%2Ftep-v8.1-evidence%2F2026-08-03%2Fcubie-research-tep-evidence-bd51e5e.zip` |
| `cubie_tep_whitepaper.pdf` | 235,285 | `80c017a2f9d8d6142fd964d6e47964d9d0973e1634d181014c78fb22bc8a7e75` | `https://lib.trustfortress.ai/objects/cubie-research%2Ftep-v8.1-evidence%2F2026-08-03%2Fcubie_tep_whitepaper.pdf` |

## Bundle Contents

The ZIP is a deterministic `git archive` of these paths at the research commit:

- `docs/evidence/2026-08-01-tep-migration-certification.md`
- `evidence/tep/may-2026-opus47-opus48/`
- `paper/cubie_tep_whitepaper.tex`
- `paper/cubie_tep_whitepaper.pdf`

The evidence directory includes the environment-only Anthropic harness, strict
and adaptive Opus 4.7/4.8/Fable 5 receipts, complete Cubie replay receipts,
comparison tooling, hashes, timing, token use, cost estimates, and
provider-returned thinking summaries. It contains no API key. A rerun must use
the caller's own `ANTHROPIC_API_KEY` and is a new provider receipt.

## Reproduce

```powershell
Invoke-WebRequest `
  'https://lib.trustfortress.ai/objects/cubie-research%2Ftep-v8.1-evidence%2F2026-08-03%2Fcubie-research-tep-evidence-bd51e5e.zip' `
  -OutFile tep-v81-evidence.zip
Get-FileHash -Algorithm SHA256 ./tep-v81-evidence.zip
Expand-Archive ./tep-v81-evidence.zip ./tep-v81-evidence
Set-Location ./tep-v81-evidence/cubie-research-bd51e5e/evidence/tep/may-2026-opus47-opus48

$env:ANTHROPIC_API_KEY = '<your own key>'
pwsh ./replay_opus_windows.ps1 `
  -DatasetDir C:\path\to\tep-csv `
  -Models claude-opus-4-7,claude-opus-4-8,claude-fable-5 `
  -MeasurementCount 52 `
  -FullCoverage `
  -OutFile ./full52-strict.json

pwsh ./replay_opus_windows.ps1 `
  -DatasetDir C:\path\to\tep-csv `
  -Models claude-opus-4-7,claude-opus-4-8,claude-fable-5 `
  -MeasurementCount 52 `
  -FullCoverage `
  -AdaptiveThinking `
  -ThinkingEffort high `
  -OutFile ./full52-adaptive.json
Remove-Item Env:ANTHROPIC_API_KEY
```

Use `-DryRun` before setting the key to validate dataset hashes, columns, and
the 192-window schedule without making provider requests. Current aliases are
not immutable model identities; save returned model IDs and request IDs with
every new receipt.

## Scope

The v8.1 results are same-data replay observations, not held-out estimates.
The corrected 50-active-cell profile is still Training-only and failed its
fault-9 floor. This publication does not change the product runtime dependency
pin or grant confirmation, deploy, or production authority.