v0.1.0 — the version the Quickstart already named

First tagged release. docs/Quickstart_OSCTL_v1.md has named v0.1.0 since it
was written; until now no such tag existed. This makes that reference true.

What this is

An evaluation harness where a run produces a manifest, a decision log and a
verify report, and where verification refuses a run whose stored verdict its
own event log does not support.

The workloads are toys. The refusal is not — of the four runs make produces,
one fails, by name:

$ make verify-all
{"status": "PASS", "run_id": "FORBIDDEN_RESEARCH_001"}
{"status": "PASS", "run_id": "MODERATION_DEMO_RUN"}
{"status": "PASS", "run_id": "OSS_TOY_REJECT_RUN"}
{"status": "FAIL", "run_id": "OSS_TOY_RUN", "detail": "fail: forensic_reconstruction"}

OSS_TOY_RUN declares 17 checks; exactly one of them is FAIL, and it is
forensic_reconstruction. That output is the literal result of a fresh clone of
this tag, not retyped from memory.

What changed since publication

Most of the history is retraction. This release is the point at which the
repository's claims and its behaviour agree, so the list is what was wrong:

• Merge-conflict markers that shipped in the public tree from the first
  commit — 206 days in README.md (2026-01-06 to 07-31) and 207 in LICENSE
  (to 08-01), both measured from the commits that introduced and removed them.
  Alongside twelve claims that measurement refuted — named theorems that do not exist, a file
  that does not exist, a proof status that overstated what the scripts prove.
  Withdrawn in the README rather than quietly deleted.
• A verdict that read a list nothing ever wrote to, so a report could carry
  a FAIL check under a PASS verdict.
• An audit step that emitted hardcoded scores — composite: 8.2,
  recommendation: ACCEPT, computed from nothing. Deleted rather than repaired:
  repairing it would have turned a red X into a green badge certifying invented
  numbers.
• Documented commands that called an interpreter a clean machine does not
  have. CI now runs the interface the README documents, on Ubuntu, against
  Python 3.10 and 3.13, with python shadowed by an exit-127 stub so the
  documented python3 path is the one actually exercised.
• Gates that compared an exit code and never a check name. osctl verify
  collapses 17 named checks onto four exit codes, so exit 1 said "something
  failed" and no more. Measured: storing REJECT in OSS_TOY_RUN/govdec.json
  makes forensic_reconstruction PASS — the demonstration gone — while a
  different check fails, and both gates stayed green. They now assert the name.
• A sign-off rule that lived only in prose. The README has asked for
  git commit -s since publication while nothing checked it and no commit
  carries the trailer. .github/workflows/dco.yml now enforces it on pull
  requests, with 17 controls in tools/dco_check.py --selftest.

Known limits, stated rather than left to discovery

• The DCO gate runs on pull requests only, so a commit pushed straight to main
  by the maintainer is not checked. Branch protection would close that; it is
  configured with enforce_admins false, so it currently binds nobody.
• proofs/ holds hand-written Lean 4, Isabelle/HOL and Dafny scripts. The
  README states what they do and do not establish. They are not machine-checked
  in CI.
• formal_proof_check_stub is UNKNOWN by construction — nothing in this build
  computes a formal status — and is named in the accept list rather than
  suppressed globally.

Apache-2.0. Requires Python 3.10+.