dennisthiessenandClaude Opus 5
c3ae5ad949
v3 was calibrated by replaying the series offline, but that harness was never
committed -- so its published numbers could not be re-derived, and a v4 cut would
have had to choose anchors by argument rather than measurement. This is that
harness, and it reproduces the published figures.
scripts/run_regime_monitor_calibration.py replays State/Warning session by
session from the same inputs the live job uses (Alpaca for all 33 symbols, FRED
for VIX and HY OAS), with no database: breadth and divergence come from
breadth_service's pure helpers. It never reimplements an unchanged live sensor --
_compute_index, _score_pillars, P2, P4 and the Warning sensors are imported and
called. Only candidate formulas (proposed v4) and retired ones (v2, gone from the
codebase) are defined here and patched onto the module for a variant's duration.
Reproduction of the 408 sessions ending 2026-07-24, against the figures in
docs/research/regime-monitor-v3.md:
v2 State avg 22.6 -> 22.68
v2 State p80 35.1 -> 35.1 exact
v2 State max 91.2 -> 91.2 exact
v2 P3 pegged 39 -> 39 exact
v2 W1 live 108 -> 108 exact
v3 State max 87.4 -> 87.4 exact
v3 band shares 73.3/15.0/8.3/3.4 -> 73.0/15.4/8.1/3.4
Three things the harness had to get right to reach that, each of which was
initially wrong and caught by a gate rather than by inspection:
- "W1 live 108" counts NONZERO sessions, not non-null ones. v2's divergence
gate returned 0.0 during any decline (v3 tapers instead), so the retired
divergence formula had to be reconstructed too.
- v2 sliced HY_OAS_REFERENCE_YEARS = 10.0 per session, not v3's 700 days. The
percentile leg ranks against that window, so replaying it short shifted the
middle of the distribution while leaving the max exact.
- The published v2 numbers correspond to FULL OAS coverage. Replaying v2 with
the 400-calendar-day fetch it shipped with yields max 100.0 and 133
credit-less sessions -- so that truncation was not in force when the figures
were taken. Recorded rather than assumed.
The script refuses to emit a band recommendation unless every hard gate passes
(33 symbols fetched, per-symbol warm-up and final bar, full basket on every
session, calendar anchors, 100% coverage, and a row-wise state_v4 <= state_v3
invariant), and exits non-zero. It is meant to be structurally impossible to read
a calibration result out of a run whose pipeline did not validate. No v4 code
ships in this commit.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Backtest report index
Reports dated 2026-07-11 or earlier are the historical production research record and remain untouched.
The completed 2026-07-12/13 S/R and Gate Target Ladder research is preserved as compact decision evidence instead of full per-arm replay output:
sr-v2-validation-comparison.jsonandsr-v2-validation-cohorts.csvrecord the held-out detector comparison.sr-full-production-vs-candidate-comparison.jsonand its cohort CSV record the full-period clean-structure replacement decision.sr-explicit-target-ladder-comparison.jsonand its cohort CSV record exact GTL parity: 202,765 candidates, 1,086 qualified setups, 321 book trades, Sharpe 2.03, CAGR 50.0%, and max drawdown 21.4% in both arms.- The three
backtest-20260713-gtl-*.json/.mdpairs record the tuning, confirmation, and strength-sensitivity decisions. No stable improvement was found, so the production GTL stayed frozen.
The large backtest-sr-*.json replay files were removed after consolidation.
They duplicated hundreds of thousands of candidate rows while adding no
decision information beyond the compact comparisons and the narrative in
docs/research/sr-levels-and-exits.md. The original raw files remain available
in Git history if a forensic reconstruction is ever necessary.
The initial untracked backtest-20260712-sr-detector-rewrite.json is local-only
and is intentionally not part of the repository.
The 2026-07-17 post-stop re-entry decision is preserved in
daily_reentry_matrix.json. It is the canonical source for the nine-policy
daily replay, production-like full-universe ranking, the disjoint 2025+ book,
and the cost/capacity sensitivity matrix. The interpretation and production
decision live in
docs/research/post-stop-reentry.md.
The earlier post-stop-reentry-20260717.json,
post-stop-cooldown-sweep-20260717.json, and
gate-protected-stop-20260717.json reports were removed as superseded
intermediate experiments. They used weekly/hybrid entry cadence or tested the
rejected stop-adjustment path, and add no decision evidence beyond the final
daily matrix and narrative. Their matching one-off runners were removed too.
All remain recoverable from Git history. Rebuildable candidate pickle caches
are intentionally ignored and must not be committed.
Phase B fip breadth IC (2026-07-18/19) — compact evidence
Canonical artifacts:
fip-reconcile-20260719-000520.json— single-sourced authoritative ICs (unconditional liquid fip, tiers, prod-subset, mom-conditional, context signals). Membership symbol dumps stripped after the decision; narrative indocs/research/fip-breadth-ic.md.fip-breadth-20260718-211440-fingerprint.json— prod-snapshot fingerprint pass (fip IC −0.045 / t −2.91).
Removed as superseded / dangerous intermediate noise (recoverable from Git):
fip-breadth-20260718-211440-breadth.json(+ wrapper) — orphaned +0.0575 / t +5.12 from racing a partialresearch.sqlite. Kept out of the tree so it cannot be re-mythologized.fip-breadth-20260718-194828*.json— fingerprint-only partial run.fip-breadth-diagnostics-20260718-213705.jsonand…-213908.json— dual-path diagnostics superseded by the single-sourced reconcile.