Repository navigation
Add stable_translog transition function - #101
Open
hmgaudecker wants to merge 8 commits into
Open
hmgaudecker wants to merge 8 commits into
hmgaudecker wants to merge 8 commits into
Conversation
…al factors Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Y3CuwvNUhYTSpriKYSwqmq
Squares and interactions are evaluated on g(x) = x / (1 + |x/s|^p)^(1/p), so the base nonlinear remainder is bounded; full-law conditions are documented. Parameter layout, names and identity constraints are those of translog. Includes the design spec, plan and round-0 audit sources under docs/superpowers. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k
process_model binds keyword-only configuration of built-in transitions (stable_translog's scale/power) at their defaults, so dags does not expose them as inputs of the transition DAG. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## missing-observed-factors #101 +/- ##
===========================================================
Coverage ? 95.42%
===========================================================
Files ? 148
Lines ? 15178
Branches ? 0
===========================================================
Hits ? 14484
Misses ? 694
Partials ? 0 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
Native closure log, base-versus-head resource report and the completion record for the round-1 closure audit. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k
…runners The three estimation tests in tests/test_chs_estimate.py take about four minutes each and dominated the twenty-minute suite. Each now runs in its own ubuntu job; the rest shard on every OS deselects them. Coverage uploads from the four Linux jobs are merged by codecov. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k
…F3 tests Run the reviewer's paired kernel benchmark natively and add a workload-level likelihood timing; add the endogenous same-law contribution/gradient test and the sigma-weighted moment reconstruction test; keep mutant transcripts; update the design's test description and the completion record. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Translog whose squares and interactions are evaluated on soft-clipped states
g(x) = x / (1 + |x/s|^p)^(1/p), linear part untouched. Iterated transitions stay
bounded under the fixed-matrix, bounded-forcing theorem in the revised design; this
is not a general certificate for a composed stochastic model.
Design note: docs/superpowers/specs/2026-10-16-stable-translog-design.md
Plan: docs/superpowers/plans/2026-10-16-stable-translog.md
Motivation: in the health-cognition application the estimated psychological-capacity
translog has an unstable fixed point at -2.2, inside the data range; forward
predictions without measurements can diverge. The supplied diagnostic claims do
not establish that every likelihood value or gradient is unaffected; estimation and
simulation with the changed law must be checked separately.
Tests: unit tests for clipping, boundedness, identity, JIT/grad, a 200-period
iteration with the estimated coefficients, allow-list registration, and a CHS
likelihood smoke test.
Stacked on
missing-observed-factors(branched from its local head 94aa213, whichis not yet pushed, so that commit appears in this diff). The round-1 closure audit
is still open; this PR is opened before it closes.
https://claude.ai/code/session_017psXwkMuDeVmTu5WgoM26k