Skip to content

RETRACTIONS.md

Latest commit

 

History

History
53 lines (43 loc) · 3.93 KB

RETRACTIONS.md

File metadata and controls

53 lines (43 loc) · 3.93 KB

DM3 Retractions Register

Last updated: 2026-04-28.

This file is the canonical Retractions Register for DM3 under License Section 19.3(d). Retractions remain first-class public artifacts. They stay visible here and in CLAIMS.md. Historical claims are not deleted; they are marked and linked to the killing or superseding evidence.

Status semantics

  • RETRACTED — a previously promoted claim whose pre-registered kill criterion was met. The claim is removed from the active ledger and the killing evidence is recorded.
  • REJECTED-BEFORE-PROMOTED — a candidate claim that was caught and superseded inside the same workstream before it was added to the promoted ledger. Recorded as a process win, not a retraction in the License sense.
  • WEAKENED — a claim whose statement was narrowed in light of new evidence. The earlier wording is preserved in the retired-claims section of CLAIMS.md and linked here.

Current entries

Claim Status First recorded Summary Evidence
H2 RETRACTED / KILLED 2026-04-16 The transformer does not create bistability. Geometry is sovereign; bistability survives Hamiltonian-mode replay without transformer dependence. docs/restart/DM3_MULTI_HYPOTHESIS_FINAL_REPORT.md, docs/restart/DM3_SESSION4_FINAL_REPORT.md
γ RETRACTED 2026-04-17 rot=60° does not uniquely break C3 under asymmetry. Session 5 pooled replication removed the claimed 60°-specific effect. artifacts/phase_P2b_coupling/PHASE_P2b_SUMMARY.md, CLAIMS.md
δ.3 WEAKENED 2026-04-22 The Session 6 reading that R3 "payload-moves along --steps" is weakened. SY-default surface saturates at operational_steps == 10; requested --steps > 20 no longer advances payload. R3 remains structurally unreachable from the exposed CLI. artifacts/phase_S7_P0_receipt_harness_20260418T151800Z/S11_final/, CLAIMS.md
σ REJECTED-BEFORE-PROMOTED 2026-04-25 The Session 8 Phase A draft "coarse 30/40 peak" wording for the exp_k2_scars learning curve was caught and superseded by the A5 one-step sweep before it reached the promoted ledger. docs/restart/DM3_SESSION8_PHASE_A5_B3_A6_FINAL_REPORT_20260425.md, CLAIMS.md retired-claims block
σ′ REJECTED-BEFORE-PROMOTED 2026-04-25 The A5 interim "bimodal s32/s41" wording was caught and superseded by the A6a/A6b fills before it reached the promoted ledger. The active candidate is σ″ (trimodal sawtooth). docs/restart/DM3_SESSION8_PHASE_A5_B3_A6_FINAL_REPORT_20260425.md, CLAIMS.md retired-claims block
θ (Session 7 wording) SUPERSEDED 2026-04-25 The Session 7 wording of θ ("harmonic dynamics layer is substrate-invariant") is reused in the Session 8 ledger for the closed compound-axis R1+R2 flip with r4.transfer_ratio doubling, and the older substrate-null line is now carried under ι. The change is bookkeeping; the underlying evidence is preserved in both anchor cells. CLAIMS.md claims θ and ι, artifacts/phase_S7_P0_receipt_harness_20260418T151800Z/S11_final/

Related lineage notes

  • The σ → σ′ → σ″ lineage is preserved in CLAIMS.md and is promoted as a process win, not buried. The discipline is that draft wordings are caught inside the workstream that produced them.
  • The R3 weakening is paired with the SY-default operational_steps saturation finding; the weakened wording remains visible alongside the strengthened evidence base.

Operating rule

  • A claim enters this register when its pre-registered kill criterion is met, when a stronger receipted result explicitly retracts it, or when a candidate is caught and superseded inside its source workstream before promotion.
  • Retractions are prospective and binding on the public DM3 surface from the moment they are published here.
  • Historical wordings are not deleted; they are marked and linked to the killing or superseding evidence. The lineage stays visible.