Skip to content

SPACE vNext: typed automatic retrieval reframes and source-conflict independence - #57

Open
Kitahl wants to merge 1 commit into
upgrade/challenge-vnext-meditatefrom
upgrade/challenge-vnext-space
Open

SPACE vNext: typed automatic retrieval reframes and source-conflict independence#57
Kitahl wants to merge 1 commit into
upgrade/challenge-vnext-meditatefrom
upgrade/challenge-vnext-space

Conversation

@Kitahl

@Kitahl Kitahl commented Aug 29, 2026

Copy link
Copy Markdown
Owner

Scope

Upgrades SPACE/scoutbot with typed retrieval-failure diagnosis, one automatic bounded mechanism-level reframe, query/challenge lineage, and source-conflict/derivative-copy independence handling. Production discovery remains automatically routable across the frozen query/source matrix. Retrieval and challenge state remain separate from claim-scoped source evidence.

Main changes

  • adds egrt.space.v2, QueryClass, QueryHypothesis, and QueryRoundState;
  • covers all ten registered retrieval-gap classes;
  • executes at most one frozen, non-redundant reframe after a load-bearing zero-yield result;
  • rejects exact-repeat fake rounds and multiple order-dependent reframes for one parent;
  • records parent query/challenge, task, obligation, scope, candidate, outcome, and novel-yield lineage;
  • preserves UNAVAILABLE for all-adapter failure and UNKNOWN / NOT_FOUND_WITHIN_SCOPE for successful scoped zero-yield search;
  • never turns search failure into a global nonexistence claim;
  • groups byte-identical and shared-provenance copies so derivative sources do not become independent support;
  • opens neutral native SOURCE_CONFLICT challenges for assessed support/refutation conflicts and provenance collisions;
  • stores query hashes rather than raw queries in generic runtime state;
  • updates the SPACE skill/spec and adds adversarial regression tests.

Changed paths

  1. tools/space_runtime.py
  2. tests/test_space_challenge_runtime.py
  3. docs/specs/SPACE_ENGINEERING_SPEC.md
  4. skills/scoutbot/SKILL.md

Exact published revision

  • base: upgrade/challenge-vnext-meditate
  • parent: abeecf8542751a8538ef1642a5b62c4cd532badc
  • head: 01c07faf1848284bda3c13d1c1eec972629be9c4
  • source tree: 2c7ca4c022ad30cfd68b608f468c5816d797eca9
  • commit count over base: 1
  • changed paths over base: 4

Exact-head validation

Exact head 01c07faf1848284bda3c13d1c1eec972629be9c4 was temporarily presented against main only because the repository's official workflows are configured to trigger for pull requests whose base is main. The branch commit was unchanged during validation and this PR has been restored to its required Meditate base.

  • Research software validation run 33229631786: PASS
    • Ruff, runtime/layout tests, public invariants, typed-runtime validator, showcase validator, compile/syntax, and repository metadata passed.
  • Runtime portability run 33229631874: PASS
    • Ubuntu, macOS, and Windows passed.
  • CodeQL run 33229631834: PASS.
  • Security run 33229631824, attempt 2: PASS
    • full-history Gitleaks: PASS;
    • frozen/hash-locked dependency verification: PASS;
    • Python dependency audits: PASS.

Repository-wide secret-scan repair

The first security attempt exposed 834 third-party findings from the copied Hermes source snapshot introduced by ancestor a24a952133d03c43872dbec034a5a8ba5515a32f on work/native-hermes-fastpath; all were under vendor/hermes-agent/, with 0 findings in the four SPACE paths.

That root cause was repaired rather than suppressed:

  • work/native-hermes-fastpath was rewritten off the copied-snapshot ancestor;
  • Hermes source is now represented as a mode-160000 gitlink/submodule pinned to upstream commit 5fc308a70719a83cccdbba4c0e39c23f5a8239d5;
  • no Gitleaks allowlist was added and the first-party secret scanner was not weakened;
  • the current Hermes repair head is dda78eabc216749cb1a3010d8e533428b78721d0, tree 0ad6d8641561de46bd729a5115500d4fd7bb0483;
  • Hermes verification run 33230839031: PASS, including exact gitlink verification, dependency installation, and one agent-visible read-only status call.

The SPACE commit/tree did not change during this repository-level repair.

Authority and evidence boundary

  • SPACE owns DISCOVERY evidence only.
  • Retrieval is not source assessment.
  • Challenge proposals and resolutions are not target-domain evidence.
  • Derivative copies are not independent support.
  • NOT_FOUND_WITHIN_SCOPE is not proof of nonexistence.
  • Green tests establish the named mechanical invariants, not improved discovery performance or complete recall.

Completion state

SPACE is mechanically COMPLETE for this slice: its exact head is published, its four-path diff remains bounded, and Research validation, portability, Security, and CodeQL are green. No claim of improved discovery efficacy or complete recall is made.

Stop rule

This PR contains SPACE only. POWER is not included.

@Kitahl
Kitahl changed the base branch from upgrade/challenge-vnext-meditate to main August 29, 2026 01:40
@Kitahl Kitahl closed this Aug 29, 2026
@Kitahl Kitahl reopened this Aug 29, 2026
@Kitahl
Kitahl force-pushed the upgrade/challenge-vnext-space branch 2 times, most recently from f5e1b87 to 4de25fc Compare August 29, 2026 02:35
@Kitahl Kitahl changed the title Space vNext: bounded retrieval diagnosis and source-conflict handling SPACE vNext: typed automatic retrieval reframes and source-conflict independence Aug 29, 2026
@Kitahl
Kitahl changed the base branch from main to upgrade/challenge-vnext-meditate August 29, 2026 02:36
@Kitahl
Kitahl force-pushed the upgrade/challenge-vnext-space branch from 4de25fc to 01c07fa Compare August 29, 2026 02:39
@Kitahl Kitahl closed this Aug 29, 2026
@Kitahl Kitahl reopened this Aug 29, 2026
@Kitahl
Kitahl changed the base branch from upgrade/challenge-vnext-meditate to main August 29, 2026 02:42
@Kitahl Kitahl closed this Aug 29, 2026
@Kitahl Kitahl reopened this Aug 29, 2026
@Kitahl
Kitahl changed the base branch from main to upgrade/challenge-vnext-meditate August 29, 2026 02:56
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant