Methodology changelog

Append-only. A methodology version bump is the ONLY sanctioned way to change a numerical policy (SPEC#1 AC-1.4). The full procedure — proposal → public comment window → version bump — is docs/METHODOLOGY_CHANGE_PROCESS.md. Each entry lists the policies that version is defined by; the registry in tly/methodology.py pins them and CI enforces the pairing.

v0.5.0-reconstruction — 2026-08-25

Adds the excess-age-profile policy (B-uc2-12/13); prior policies unchanged.

v0.4.0-reconstruction — 2026-08-17

Adds the quanta policy (B-uc3-04); prior policies unchanged.

v0.3.0-reconstruction — 2026-08-17

Adds the P6 identity-closure tolerance policy (B-uc2-08); prior policies unchanged.

v0.2.0-reconstruction — 2026-08-17

Adds the nowcast baseline policy (B-uc2-04); v0.1.0 policies unchanged.

v0.1.0-reconstruction — 2026-08-17

Initial packaged methodology, reconstructed 2026-08-16 from DECISIONS.md / RESEARCH_PROGRAM.md / RALPH_LOOP.md after loss of the originals; pending ratification (loop/BACKLOG.md A-16).

v0.6.0-reconstruction (2026-08-25)

Changed (the registry's FIRST parameter change, not an addition): the period-vs-cohort one-sided error-budget upper bound moves +8% -> +9%. Driver: the E6 cohort computation on the committed 2010-2100 qx surface measured the uplift at +8.06% (tests/test_cohort.py), 0.06pp above the +3-8% literature-prose band — computed-beats-prose (ruling D6). The one-sided bounds are now GOVERNED, version-keyed parameters (tly.methodology.VERSION_ONE_SIDED_TERMS): reproducing an archived print selects the band its version was computed under, never HEAD's — the outsider-sim recomputes each epoch under its archived version.

Effect on published text: the cohort best-estimate band becomes ~381-406B (+5%/+12% applied to measured S). The A-16 packet's blessed 381-402B remains the correct v0.5.0 statement and archived prints carrying it continue to reproduce byte-exactly.