Skip to content

STATE.md

Latest commit

 

History

History
57 lines (38 loc) · 2.97 KB

STATE.md

File metadata and controls

57 lines (38 loc) · 2.97 KB
Raw

State

Last updated: 2026-07-29

Phase

No active study / Study 008 activation NO-GO

Laboratory

  • Public operator: Templex Tsukino / 月野テンプレクス
  • Laboratory: TEMPLEX/0
  • Repository: yo4e/templex-zero
  • Execution model: governance/APPROVAL_DRIVEN_EXECUTION.md
  • Active issue: None

Closed studies

Studies 001 through 007 are closed.

  • Study 006: valid partial Python tar extraction boundary-conformance result; H1 and H2 supported, H3 unsupported.
  • Study 007: negative SQLite setup result; hand gate failed before protected execution.

Study 008 activation decision

  • Frozen proposal: research/proposals/008-summeval-dimension-specific-proxy-reliability.md
  • Activation decision: research/decisions/2026-07-29-study-008-activation.md
  • Disposition: NO-GO; Study 008 was not activated
  • Inspected official upstream: Yale-LILY/SummEval
  • Pinned inspected upstream commit: 81b59ad53d63cb6009764240853c91235a44e238
  • Repository license: MIT
  • Official human annotation referent: model_annotations.aligned.jsonl
  • Official scored annotation referent: Google Drive file ID 1d2Iaz3jNraURP1i7CfTqPIj8REZMJ3tS

The human-only annotation file is reported in official Issue #18 not to contain the required metric_scores_* fields. Official Issue #56 reports the scored Google Drive file unavailable and remains open without an official replacement.

The frozen proposal forbids mirror substitution, reconstructed metrics, guessed directions, reduced denominators, and source-article reconstruction. Therefore the scored file SHA-256, byte length, schema, exact scored denominator, complete metric inventory, numeric coverage, and score directions could not be frozen. NO-GO was mandatory.

No dataset was acquired, no active-study issue was opened, no instrument was implemented, no metric-human relationship or rank was inspected, and no H1–H3 disposition exists.

self/FAILURE_MODES.md now records FM-010 — Referent availability optimism. A paper or README naming an artifact is not evidence that the exact official bytes remain retrievable and pinnable.

Next bounded work

The next exact 承認 may perform one post-Study-008-NO-GO portfolio decision only:

  1. re-read all seven closed studies, the Study 008 proposal and NO-GO decision, governance, self-model, failure modes, open issues, and recent commits;
  2. freeze a revised selection threshold before candidate research or scoring;
  3. make outcome-blind official referent availability a hard feasibility gate for indispensable external artifacts;
  4. compare materially distinct directions plus inactivity;
  5. select at most one frozen inactive proposal or remain inactive;
  6. update decision, state, restart, and intervention records;
  7. stop before activation, dataset acquisition beyond availability preflight, implementation, protected analysis, outcome inspection, or external action.

Human action currently needed

None beyond a later exact 承認 for the bounded post-NO-GO portfolio decision.