01 / Version and scope
This release uses methodology 0.2.0-research. The illustrative base remains 01 January 2021 = 100.00000. It has not been approved for contractual settlement.
Version 0.2 adds an explicit register of official Treasury contingency inputs and separates publication records from reconstructed history. Version 0.1 remains available as the BLS-only calculation. A version change does not rewrite earlier fixing records.
02 / Public parameters
The parameters below govern the calculation in this release. The machine-readable specification carries the same version identifier as each fixing record.
| Methodology | 0.2.0-research |
|---|---|
| Status | Research preview; not approved for settlement |
| Series | CUUR0000SA0 · U.S. CPI-U, U.S. city average, all items, NSA |
| Base date | 2021-01-01 (illustrative) |
| Base value | 100.00000 |
| Base Reference CPI | 260.38800 |
| Valuation time zone | UTC |
| Input months | M−3 and M−2; only M−3 is needed on day 1 |
| Day weight | (dayOfMonth − 1) / actualDaysInMonth |
| Reference CPI rounding | Truncate to 6 decimal places, then round half-up to 5 |
| Index rounding | Round half-up to 5 decimal places after multiplying the reference CPI ratio by 100 |
| Floor / cap | None / None |
| Prediction model | None |
| Target daily check | 15:30 UTC; best effort. Late runs retain their actual timestamp. |
| Overdue threshold | 16:30 UTC when that day has no successful publication |
03 / Monthly observations
The underlying series is the U.S. Consumer Price Index for All Urban Consumers: U.S. city average, all items, 1982–84 = 100, not seasonally adjusted. BLS series identifier: CUUR0000SA0. Annual averages and seasonally adjusted data are excluded.
The BLS dataset is retained as retrieved, including missing observations. The only registered contingency in version 0.2 is the October 2025 CPI input of 325.604, formally announced by the U.S. Treasury on 26 November 2025. It is stored separately from the BLS observations and remains the input for this methodology even if BLS later supplies that month.
The October input affects valuations from 02 December 2025 through 31 January 2026. Each affected calculation identifies Treasury as the source. No other missing observation is estimated or filled by inference.
04 / Daily Reference CPI
For a date in calendar month M, the first input is CPI for M−3 and the second is CPI for M−2. The day weight is (d−1)/D, where d is the day of the month and D is the actual number of calendar days in that month. Valuation dates use UTC. On the first of a month, only the M−3 observation is required.
Reference CPI is truncated to six decimal places and then rounded half-up to five. Leap days and month lengths are part of the calculation; business-day calendars do not alter the reference date.
05 / From Reference CPI to IPU
The index divides the five-decimal Reference CPI for the valuation date by the five-decimal base Reference CPI, multiplies by 100, then rounds half-up to five decimal places. The base uses the same Reference CPI method.
The base Reference CPI is 260.38800. IPU does not use the separately rounded index ratio for any individual Treasury security. It has no floor, cap, forecast or discretionary price adjustment. A negative price-level change can reduce the index.
06 / Publication and reconstruction
A published daily fixing is a stored record with a valuation date, actual publication time, methodology version, source dataset hash and record hash. The original record uses revision 0 and is retained when the daily task runs again.
Reconstructed history applies a chosen methodology to a specified source snapshot. Its dates can predate the first live publication. It is labelled separately and never treated as an archive of values published at the time.
The task targets 15:30 UTC each day, with best-effort scheduling. A day without successful publication is flagged after 16:30 UTC. A late run retains its actual timestamp. A missed day does not acquire a backdated publication record.
07 / Missing data and corrections
An unregistered missing input makes the affected date unavailable. Publication stops for that date; the previous fixing remains visible with its own date and status. Values are not carried forward as though a new fixing had been produced.
A change to an existing numeric BLS observation is quarantined for review. The accepted dataset and earlier fixings remain intact. A correction must have its own revision ID, reason and link to the superseded record. The original-fixing endpoint continues to identify the original publication.
A newly supplied BLS observation may enter a new source dataset. It does not remove an explicitly registered Treasury contingency from this methodology. Any change to that rule requires a new method version.
08 / Reproduce and compare
Start with a fixing ID or a reconstructed-history response. Download its method specification and dataset by identifier. Select the dated input months, apply any enumerated contingency, calculate and round Reference CPI, then calculate and round the index.
SHA-256 identifiers use the canonical JSON representation: object keys sorted recursively, array order preserved, UTF-8 encoding and no formatting whitespace. The record hash excludes the recordHash field itself. A checksum detects a change to the recorded content; it is not a digital signature or independent assurance.
Compare Reference CPI with the corresponding Treasury Reference CPI table. A security-specific Treasury index ratio has a different base and is not an appropriate direct comparison for IPU.