DRAFT: LISA ILE driver — drift gate + fair-draw weighting family (pass 1 of N)

[oshaughnessy-junior]

Draft. First milestone of a multi-pass effort to catch the LISA ILE driver up with the main one. Not for review-out-of-draft yet.

The two drivers are a deliberate fork (RO, 2026-08-13: "the overhead of one ring to rule them all is too high"). This does not argue with that — it makes the consequence, drift, mechanically visible so it stays a choice.

Measured

On junior/rift_O4d @ 364a22fd: main 4,883 lines, lisa 2,526, with 132 items (helpers, CLI options, module constants, sampler provenance markers) in main and absent from lisa. Both drivers import the same integrators and expose an identical ok_lnL_methods, so all of that drift is in the driver.

The audit + ledger

audit_lisa_driver_drift.py diffs the two by AST across FUNC/OPTION/CONST/ATTR; make_lisa_drift_ledger.py holds the judgements as ordered family rules.

132/132 classified — PORT 70, NA 43, PHYSICS 11, PORTED 8. An item matching no rule fails --check (fired for real once, on --sampler-anisotropic-bins). "Does not apply to LISA" is a fine answer; silence is not.

Two extraction traps recorded: the drivers use optparse (an argparse-only scan reports zero options), and the provenance readers use getattr(obj,'name',default) — a Call, not an Attribute — so a naive scan reports a real port as a no-op.

The port

The three consumers PR #87 actually fixed — the proposal breadcrumb, .dgrid, the .dslice core — do not exist in this driver, so there was no live w² bug here. The hazard did: the driver sets igrand_fairdraw_samples from --fairdraw-extrinsic-output, and all seven shared rebind sites already set _rvs_is_fairdraw. The marker was arriving and nothing read it.

Ported ln_weights_from_rvs, ln_weights_for_posterior, _rvs_is_export_resample, _rvs_is_equal_weight, _rvs_len, _rvs_lnL_convention and the marker reads.

Trap avoided: --internal-use-lnL is also accepted for adaptive_cartesian_gpu and portfolio, which set use_lnL without return_lnI and still store linear L. The stored convention is derived from pinned_params['return_lnI'], never the option.

Deliberately not done (both recorded at the site):

• ln_weights_for_posterior passes use_lnL through unresolved, exactly as main does. That is a latent trap in both drivers; a same-named helper behaving differently across the fork would be worse. Worth fixing in both, together.
• _truthy_option was moved out of this family once its only caller turned out to be the --interpolate-time normalizer — porting it would have been dead code.

Tests

• test_lisa_fairdraw_weights.py (29), including an anti-drift test pinning each ported helper AST-identical to main's (docstrings excluded). Revert-checked: six mutations, each caught by its named test, file restored byte-identical.
• test_lisa_driver_drift.py (7), the gate. Revert-checked both directions.

Both wired into the lisa-check CI job — which already ran nine LISA test files and stayed green through all 2,357 lines of this drift, because all nine are import/contract/smoke level. The gate does not test the physics; it refuses to let a new item through without a recorded human decision.

Confirmed not porting (RO, this thread)

.dslice/.dgrid, --rotation-* (LISA has its own modules/TDI), calibration ("maybe eventually but not in this framework"). All recorded NA with reasons in the ledger.

Still open

70 PORT items across later passes — L0 rescue + warm-start state (15, highest value) first — and 11 PHYSICS items needing a decision, listed in LISA_DRIVER_DRIFT.md.

🤖 Generated with Claude Code

[oshaughnessy-junior]

Superseded by #106, which consolidates this stack onto current rift_O4d (d904e72d), adds RO's review commit, and fixes what an adversarial audit found. Content is included there; closing to leave one PR for the review cycle.