Phase 2: implement Kaplan–Meier and fixed-horizon RMST summaries #664

Merged
erikinkinen merged 5 commits from phase2/p2-14-d-survival into main 2026-10-27 02:24:02 +01:00
Owner

Task

Closes #119

Summary

Implement independent Kaplan–Meier and fixed-horizon RMST summaries for the
protocol's first-breach, reactivation and permanent-clearance episode inputs.
Retain administrative censoring, no-followup records, exact median crossings,
unsupported-tail reasons and whole-seed draw multiplicity.

Scope

Included

  • Standard-library estimator, documented input contract and portable fixtures.
  • Source-bound locked lifelines 0.30.3 crosschecks and integration receipts.

Explicitly excluded

  • Production endpoint extraction, inference intervals and research conclusions.

Design intent

Risk sets include tied events before tied censors are removed. RMST refuses
unsupported positive tails. Empty seeds and dependent episodes remain explicit;
seeds are the resampling units. No Greenwood independence interval replaces
cluster uncertainty.

Phase discipline

P2-14-D implements the frozen protocol 2.2 statistical primitive. The phase 5
horizon is supported without claiming the required recovery campaign completed.
All statistical implementation and fixture files retain their validated hashes.
Only tests/CMakeLists.txt combines the previously validated test registrations.

Verification

Source 4151d4214d01f5116b44adecbbb598eb3279a918 passed locked and portable
fixtures: six analytical groups, 216 ordered three-episode configurations,
27 unequal three-seed draws, 267 lifelines crosschecks and 32 negative cases.
Original source evidence
has SHA-256
2f7c9fda7982f59a8aa6492751d8fa004c9d85485e9c5927d34ad2fb6e76d529.

Integration source 496ecb397de50adcbbe4d4b1c095045d4147c31a preserves all
owned non-build files, configures exactly 296 tests and passes the portable
survival fixture. All eight hooks and the frozen-scope check pass.
Integration evidence
has SHA-256
ee93b7d9d6632aab2ce621965322f8ac054fec1f9e98fb9534e0db85402ae6ce.
The complete local native suite was not rerun for unchanged implementation.
Both final smoke/sanitizer suites and the other six CI contexts remain required
on this PR's final head.

Notes

This PR is queued behind PR #663 at
b25aa19d1515b8713ffb62a0297cafbfce9782ca to overlap CI. Until that PR
merges, the comparison with main also includes its changes. Merge #663 and
verify issue #118 closed with its saved successful exact-head CI first. Native
prerequisites for this issue are already closed.
Check issue #119 once only after all ten own exact-head contexts succeed, then
perform the guarded signed fast-forward merge. The source validation is
statistical-method validation; it does not establish production censoring
assumptions or a scientific result.

# Task Closes #119 ## Summary Implement independent Kaplan–Meier and fixed-horizon RMST summaries for the protocol's first-breach, reactivation and permanent-clearance episode inputs. Retain administrative censoring, no-followup records, exact median crossings, unsupported-tail reasons and whole-seed draw multiplicity. ## Scope ### Included - Standard-library estimator, documented input contract and portable fixtures. - Source-bound locked lifelines 0.30.3 crosschecks and integration receipts. ### Explicitly excluded - Production endpoint extraction, inference intervals and research conclusions. ## Design intent Risk sets include tied events before tied censors are removed. RMST refuses unsupported positive tails. Empty seeds and dependent episodes remain explicit; seeds are the resampling units. No Greenwood independence interval replaces cluster uncertainty. ## Phase discipline P2-14-D implements the frozen protocol 2.2 statistical primitive. The phase 5 horizon is supported without claiming the required recovery campaign completed. All statistical implementation and fixture files retain their validated hashes. Only tests/CMakeLists.txt combines the previously validated test registrations. ## Verification Source `4151d4214d01f5116b44adecbbb598eb3279a918` passed locked and portable fixtures: six analytical groups, 216 ordered three-episode configurations, 27 unequal three-seed draws, 267 lifelines crosschecks and 32 negative cases. [Original source evidence](https://git.erikinkinen.fi/attachments/6495a1ee-e4b9-4ff5-ad47-92aac0bbc8ee) has SHA-256 `2f7c9fda7982f59a8aa6492751d8fa004c9d85485e9c5927d34ad2fb6e76d529`. Integration source `496ecb397de50adcbbe4d4b1c095045d4147c31a` preserves all owned non-build files, configures exactly 296 tests and passes the portable survival fixture. All eight hooks and the frozen-scope check pass. [Integration evidence](https://git.erikinkinen.fi/attachments/33c8b1c8-2cdc-4975-811b-dec593b1648c) has SHA-256 `ee93b7d9d6632aab2ce621965322f8ac054fec1f9e98fb9534e0db85402ae6ce`. The complete local native suite was not rerun for unchanged implementation. Both final smoke/sanitizer suites and the other six CI contexts remain required on this PR's final head. ## Notes This PR is queued behind PR #663 at `b25aa19d1515b8713ffb62a0297cafbfce9782ca` to overlap CI. Until that PR merges, the comparison with main also includes its changes. Merge #663 and verify issue #118 closed with its saved successful exact-head CI first. Native prerequisites for this issue are already closed. Check issue #119 once only after all ten own exact-head contexts succeed, then perform the guarded signed fast-forward merge. The source validation is statistical-method validation; it does not establish production censoring assumptions or a scientific result.
erikinkinen added this to the Phase 2 milestone 2026-10-27 01:16:13 +01:00
Phase 2: record unchanged descriptive statistics after CI reconciliation (#118)
Some checks failed
clang-format / check-format (pull_request) Successful in 1m16s
markdownlint / markdown-lint (pull_request) Successful in 56s
formal / definitions (pull_request) Successful in 2m32s
ci / smoke (pull_request) Successful in 19m36s
ci / sanitizers (pull_request) Successful in 54m38s
markdownlint / markdown-lint (push) Successful in 1m7s
clang-format / check-format (push) Successful in 1m52s
formal / definitions (push) Successful in 2m50s
ci / smoke (push) Successful in 23m55s
ci / sanitizers (push) Has been cancelled
b25aa19d15
Phase 2: record survival integration identity and exact test union
Some checks failed
clang-format / check-format (pull_request) Successful in 1m4s
markdownlint / markdown-lint (pull_request) Successful in 1m4s
formal / definitions (pull_request) Successful in 2m4s
ci / smoke (pull_request) Successful in 21m56s
ci / sanitizers (pull_request) Successful in 1h1m35s
clang-format / check-format (push) Successful in 1m18s
markdownlint / markdown-lint (push) Successful in 1m26s
ci / sanitizers (push) Has been cancelled
ci / smoke (push) Has been cancelled
formal / definitions (push) Has been cancelled
f6c256a580
Sign in to join this conversation.
No reviewers
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
erikinkinen/AES!664
No description provided.