Primary-source evidence
Reinstate Research and Compatibility Evidence
Reinstate publishes source, deterministic synthetic tests, compatibility records, architecture, and physical acceptance methodology so current claims can be checked. It does not yet publish restoration benchmarks or a completed RC8 cross-device compatibility report; those require reproducible results from the open release gates.
- Last verified
- Current release
- v0.1.0-rc.8
- Release status
- pre-1.0 release candidate
- Owner
- Harjot Singh Rana
Evidence levels
| Level | What it can support | What it cannot support alone |
|---|---|---|
| Released code and unit tests | Implemented contracts, parsing, exclusions, transformations, and failure behavior | Real-device compatibility or production reliability |
| Deterministic synthetic fixtures | Repeatable adapter, path-mapping, encryption, and restore regression checks | Distribution claims about private user sessions |
| Physical acceptance result | The exact recorded release, agent versions, devices, directions, and criteria | Every untested version, platform, device, or workload |
| Production field data | The measured population and date window when methodology and denominators are shown | Universal guarantees or causes not isolated by the data |
Current evidence assets
- Compatibility data
- Machine-readable JSON, reviewed
- Fixture commit
4a7ccac3bf9b- Acceptance methodology
- RC8 MacBook + Windows runbook
- Architecture and security
- Architecture and security model
- Encrypted format specification
- Manifest and snapshot format v1, including framing, validation, and pre-1.0 limits
- Synthetic path explainer
- macOS ↔ Windows path-mapping visualizerwith fixed non-private fixtures
Adapter and compatibility evidence
The compatibility matrix links each supported agent range to the exact adapter test source. Versions outside the inclusive stable ranges—including prereleases—are treated asUNTESTED and mutating operations fail closed. The matrix separately marks open physical platform gates instead of borrowing a source-test date as platform certification.
- Claude Code:
2.1.219–2.1.220; test source; source checks and deterministic synthetic fixtures. - OpenAI Codex CLI:
0.133.0–0.145.0; test source; source checks and deterministic synthetic fixtures.
Physical results and current limitation
The repository contains dated RC5 macOS and Windows observations. They are useful historical evidence, but they do not close the RC8 acceptance matrix. The current RC8 runbook requires both agents, both directions, checksum-verifying installers, wrong-passphrase refusal, ciphertext-only remote objects, conflict handling, restore backups, and redacted evidence.
Review theraw acceptance-result directory and the qualified public compatibility summary. Do not infer a stable support claim from an older passing row.
Why the fixtures are synthetic
Real coding-agent sessions can contain proprietary source code, prompts, internal paths, terminal output, or credentials. Committed fixtures are deterministic and synthetic, and the repository's secret checks protect against accidentally turning a compatibility test into a data leak. Public evidence uses versions, exit codes, counts, snapshot identifiers, and redacted paths rather than transcript text.
Benchmark publication contract
Before publishing a speed, success-rate, or session-size result, Reinstate must disclose:
- the question and success criteria;
- hardware, operating systems, agent versions, Reinstate version, and commit;
- the synthetic dataset or safely consented data source;
- exact commands, repetitions, warm-up and cache policy, and raw output;
- every failure, exclusion, limitation, and denominator; and
- a stable canonical page, methodology date, and downloadable evidence.
No benchmark claim yet
This page deliberately reports no restoration speed, success rate, market coverage, or user distribution. A blank metric is more accurate than a number without a reproducible method.