NCB-003 · implementation note

What the v0.1 code actually did.

A review of the archived production pipeline found that the code as run differs from the April methodology text in two respects. Historical outputs are unchanged. This note records the difference rather than editing the original paper.

Published 5 September 2026
The discrepancy

Two differences between text and code.

Rebase window

Paper: the calculation window is rebased to 100 at the start of an annual measurement period, stated as 21 April.

Code as run: a rolling one-year data query with fixed target weights applied to normalized series.

Rebalancing language

Paper: describes weekly rebalancing.

Code as run: fixed target weights on normalized series, which is not the same operation.

Constituent set

The live v0.1 instrument was a seven-stock public-equity proxy across four substrates, paired with a four-market belief overlay. A broader Tier-2 constituent set was drafted in NCB-003 as a future target and was not the production instrument for the 2026 Register window.

Effect on published outputs

None. Every published FNC-1 value was produced by the code as run and is unchanged by this note. What changes is the documentation of record.

Consequence

Why this blocks a Register grade.

The Register's 31 August 2026 window includes the FNC-1 / SOX spread. A late grade is only valid if the instrument composition, calculation rule, source series and required lookback were fixed in advance or are independently recoverable. Where the paper and the code disagree, completing the grade would require choosing between them after the observation date. That choice would contaminate the grade.

The remediation is mechanical rather than interpretive: reproduce the published 20 July values from the archived pipeline, freeze that implementation as the canonical v0.1 specification, then run the identical implementation against the 31 August close.

If the July output cannot be reproduced from an independently recoverable specification
  • The 31 August window is marked ungraded due to methodology ambiguity.
  • That fact is preserved permanently in the Register.
  • A clean v0.2 window is opened with a hashed, versioned calculation implementation.

A missed window recorded honestly costs less than a reconstructed number that cannot be defended.

Original paper: NCB-003, The Novacene Composite, April 2026, unedited. Current grade status: the Register.