[CaffeinatedCoder]

Closes #97

Replaces #99, the same commits on a branch renamed from chore/oracles: chore/ is no branch type PR hygiene accepts. The decision record is 0112, after #94's 0110 and #96's 0111, both merged.

Problem

The prober cost more than it paid. prober seeds built the probe harness once per seed, 170 seeds at about four minutes each, so a release's seeds --all took about eleven hours. One change to the fault responses (#85) staled 23 seeds of five probes, about 90 minutes. A checkout whose ledger had never recorded a catch wanted all 170. The replay, the footprints, the ledger, the covering arrays and the release gate carried the same weight.

Change

Retired:

• tools/Prober and its tests;
• the corpus under tools/probe/: 38 probes, their rows, 170 seeds;
• the probes and probe-status jobs of dotnet.yml, and the replay and gate steps of release.yml;
• .claude/probe-axes.md, the probe-triage skill and the five probe conventions tests;
• DocsQuery's check of the facts the corpus cited. The verified facts and their fixtures stay.

Added, CodoMetis.ValueRanges' kit ported to xunit (test/oracles/README.md is the manual):

• claims.json: the prober's 169 obligations as 150 claims, one per distinct sentence, each quoted verbatim. 112 are held by a named test class; the 38 with an empty heldBy are the worklist.
• surface-waivers.json: every public member of the five run-time libraries is claimed or waived. Members of documents the prober never read are waived as unread:, and Oracle catalogue: read the documents the prober never read for claims #98 reads them.
• oracles.json and seeds/: five oracles that already existed (JsonParityTests, NullDelegateTests, AsynchronousContextTests, NotWireTypeTests, KeyGenerationVariantTests), each with a seed in src/.
• candidates.json (empty) and axes.md (hand-written).
• test/CodoMetis.TypeKit.Oracles.Tests holds all of it in CI without building anything else. It is a project of its own because the surface is read by reflection, and the conventions project references no library.
• tools/Oracles, run locally: seeds [filter] copies the tree, applies one seed, builds one project and runs one class; status prints the worklist.

Decision 112 supersedes decisions 96, 99 to 104 and 108, decision 107 but for its fixture rule, and decision 97's gate. AGENTS.md (commands, workflow step 7, Releasing), CONTRIBUTING.md (release steps, label legend), design/plan.md, the PR template and the verify-the-guard skill follow.

Tests

• oracles seeds: all five seeds caught, about 10 s each.
  • json-key-written-by-to-string: 71 of 157 JsonParityTests fail.
  • ushort-key-not-generated: 19 of 547 KeyGenerationVariantTests fail.
  • result-reads-as-json: 15 of 99 NotWireTypeTests fail.
  • patch-traverse-takes-a-null-selector and traverse-async-resumes-on-the-caller-s-context: one row each of NullDelegateTests and AsynchronousContextTests.
• The kit's own guards, seeded:
  • a reworded promise in docs/json.md fails Every_claim_quotes_its_source_verbatim;
  • a waiver removed fails Every_public_member_is_claimed_or_waived.
• Full suite: passes (the Oracles, Conventions and DocsQuery projects re-run after the last fix).
• main's required checks name no probe job, so nothing waits on a check that no longer runs.

Merge notes:

• feat: a fault keeps the exception that caused it, for the log and never the problem #89 and feat: assertions on a Task or ValueTask, and CMTK0003 reports one left unawaited #95 still touch seed patches under tools/probe/; resolve those as deletions.
• Each new public member they add falls under an existing claim (Fault.*, OptionAssertions.*).

🤖 Generated with Claude Code