Skip to content
Induscript

Induscript · Structural Atlas v1.0

Days 6–7 · crash-test corpus

Prove the gates before Mohenjo-daro

Five synthetic records. If directional weights do not sum to one, or a cast inflates n, or a poison-pill label leaks into a model log, the pipeline fails closed.

Pass · ART-002 partitionPass · FAM-001 n_eff = 1Pass · ART-003 silentPass · granary_tax_override absent

  • ART-001

    The perfect baseline

    Securely excavated, complete, 100% forward.

    q_p = 1.0, π_forward = 1.0. Independent until the cast is linked.

    q_p 1 · π_fwd 1 · π_rev 0 · sum 1

  • ART-002

    The contested fragment

    Legacy excavation. Two scholarly direction claims.

    Scenarios must partition weight: π_fwd = 0.8/1.2, π_rev = 0.4/1.2. Sample size must not inflate.

    q_p 0.75 · π_fwd 0.6667 · π_rev 0.3333 · sum 1

  • ART-003

    The auction block

    No provenance, unstratified.

    q_p = 0. Active weight is zero. It cannot confirm anything.

    q_p 0 · π_fwd 1 · π_rev 0 · sum 1

  • ART-004

    The cast

    Museum plaster cast of ART-001.

    Same duplicate family as ART-001. Effective n of the family must remain 1.0.

    q_p 1 · π_fwd 1 · π_rev 0 · sum 1

  • M-CODE-999

    granary_tax_override

    Synthetic trap. If this string appears in a model manifest, the run is void.

    Scan: indus-atlas-model-synthetic-0.1.0 · clean

ART-002 arithmetic

π_forward = 0.8 / (0.8 + 0.4) = 0.6667 · expected 0.6667

π_reverse = 0.4 / (0.8 + 0.4) = 0.3333 · expected 0.3333

sum = 1 · scenarios 2