Proof runs

The terminal blocks on /verify/ are run every day and compared with what proof-claims.json says they should print. The expected output lives in that file, not in the page — so a badge on /verify/ cannot say “verified” about a run that disagreed with it.

Skipped is not passed and is not failed. A rate limit or a DNS failure is not a disagreement about what the surface does; it is recorded with its reason and leaves the previous badge alone.

• Proof run 2026-10-06

  10 verified, 0 stale, 0 skipped

  website commit

  ea8e2b13a7e1

  app server

  CodeRifts 1.0.3

  app commit

  not recorded — the app exposes no commit on any public surface — /health and MCP initialize both return a version string and no SHA. surface_source_ref_commit is the coderifts/api-governance commit that pins the published tools[], not the server build.

  surface commit

  267f735e180b

  runner

  Linux, node v20.20.2, Python 3.12.3

• Proof run 2026-10-05

  10 verified, 0 stale, 0 skipped

  website commit

  7f524d5bd31a

  app server

  CodeRifts 1.0.3

  app commit

  not recorded — the app exposes no commit on any public surface — /health and MCP initialize both return a version string and no SHA. surface_source_ref_commit is the coderifts/api-governance commit that pins the published tools[], not the server build.

  surface commit

  267f735e180b

  runner

  Linux, node v20.20.2, Python 3.12.3

• Proof run 2026-10-04

  10 verified, 0 stale, 0 skipped

  website commit

  d730ca7b117e

  app server

  CodeRifts 1.0.3

  app commit

  not recorded — the app exposes no commit on any public surface — /health and MCP initialize both return a version string and no SHA. surface_source_ref_commit is the coderifts/api-governance commit that pins the published tools[], not the server build.

  surface commit

  267f735e180b

  runner

  Linux, node v20.20.2, Python 3.12.3

• Proof run 2026-09-27

  10 verified, 0 stale, 0 skipped

  website commit

  23d4ed23d00f

  app server

  CodeRifts API Governance 1.0.3

  app commit

  not recorded — the app exposes no commit on any public surface — /health and MCP initialize both return a version string and no SHA. surface_source_ref_commit is the coderifts/api-governance commit that pins the published tools[], not the server build.

  surface commit

  d68d5c9d87d3

  runner

  darwin-x64, node v24.12.0, Python 3.9.6

• Proof run 2026-09-23

  10 verified, 0 stale, 0 skipped

  website commit

  5bc2f692b837

  app server

  CodeRifts API Governance 1.0.3

  app commit

  not recorded — the app exposes no commit on any public surface — /health and MCP initialize both return a version string and no SHA. surface_source_ref_commit is the coderifts/api-governance commit that pins the published tools[], not the server build.

  surface commit

  d68d5c9d87d3

  runner

  darwin-x64, node v24.12.0, Python 3.9.6

Assurance captures

The published conformance suite, run under BOTH evidence tiers. Coverage is whether the claim is proven; evidence is whether it was proven now or replayed from a pinned capture. For each profile that is not LIVE, the capture records the condition that would take it there — in the suite’s own words, because a paraphrase of “what would make this live” is how a roadmap sentence is born.

• Assurance capture 2026-09-23

  7 profile(s), 2 on LIVE evidence, 5 on RECORDED

  measured by

  @coderifts/conformance@0.8.11

  pinned version

  0.8.11

Guard-control captures

Every guard’s own self-test, run and recorded: how many planted defects it carries and whether they all still went red. A rule that cannot be made to fail is not a rule, and this is where that stops being an assertion. The numbers are measured — the first attempt counted call sites in the source and was short by exactly one per guard.

• Guard controls 2026-10-04

  36 guard(s), 24 with a self-test, 140 control(s), all passed: true

  self-test flags

  --self-test, --selftest

  node

  v24.12.0

• Guard controls 2026-10-03

  33 guard(s), 21 with a self-test, 132 control(s), all passed: true

  self-test flags

  --self-test, --selftest

  node

  v24.12.0

• Guard controls 2026-09-29

  32 guard(s), 19 with a self-test, 110 control(s), all passed: true

  self-test flags

  --self-test, --selftest

  node

  v24.12.0

• Guard controls 2026-09-28

  32 guard(s), 19 with a self-test, 110 control(s), all passed: true

  self-test flags

  --self-test, --selftest

  node

  v24.12.0

• Guard controls 2026-09-26

  30 guard(s), 17 with a self-test, 102 control(s), all passed: true

  self-test flags

  --self-test, --selftest

  node

  v24.12.0

• Guard controls 2026-09-24

  28 guard(s), 15 with a self-test, 90 control(s), all passed: true

  self-test flags

  --self-test, --selftest

  node

  v24.12.0

The table form of the same thing, with what each guard measures and when it runs: the guards.

External measurements

Runs made outside this repository’s own proof job, indexed here so a note in docs/evidence/ cannot land unseen. A figure the note marks as not re-run was not re-run.

• External measurements 2026-09-26

  Website-side index. Two doc bugs re-measured here. Gemini timing, the doc-smoke table, and the Claude Code stage numbers were not in a file this session could read, so they are not re-stated as results.

Per-release measurements live next door: release transcripts — what each release run measured, with a value that was not measured recorded as null and its reason.

One receipt, read as a line of evidence for the record-keeping part of a change under DORA, the AI Act and NIS2: Record-keeping.

What a proof run does and does not prove is in every artifact’s own does_not_prove block. The wider honesty table is What each proof proves.