Repository navigation
Claude/phase 8.5 clean slate - #163
Merged
Merged
Conversation
…only") - Dropped: D-02 and most of D-05 (shipped in 8.4); retiring regime_labels.yaml (legacy config); the P7 exit gate (needs the regime labels the reset deletes); the per-line DSR epoch note (deferred to Phase 9's first budgeted run). - D-01: freeze with the annotated tag model/pre-reset-2026-10 on 4e49652 (88 files, 6.2 MB, already in history; no promoted tarball). The session proxy refuses tag pushes, so Glenn pushes the tag. - Three plans: tests green with no data (realdata marker, record pins retired and listed), then the reset plus registry true-zero (ADR-0005), then learnings, the Phase 9.x order and the fresh-clone Mac gate. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…registry, notebooks, gitignore Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
- 08.5-PATTERNS.md: analogs for the realdata marker and collection hook, the ledger header, ADR-0005, DECISIONS rows, retirement notes, .gitignore and the ROADMAP 9.x format. - ROADMAP: "**UI hint**: no" on Phase 08.5. The UI gate had matched the word "page" (the weekly markdown report), not a frontend. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…d the Mac gate - 08.5-01: realdata marker and collection hook (with a loud failure for `-m realdata` when there is no data); 14 lasting-property tests marked; 23 record pins retired, each citing a covering synthetic test; suite 2261 (no-data run 2247 passed + 14 skipped). - 08.5-02: git rm of the 88 generated files, /data/ and /outputs/ ignored, registry true-zero epoch (archive byte-identical) + ADR-0005, notebook outputs stripped, docs; fresh-clone run on the clone's own src. - 08.5-03: Phase 9 as 9.1-9.8, learnings into DECISIONS, guide, MODULE-MAP and MIGRATION-PLAN; a human-action checkpoint for the tag push and the Mac fresh-clone gate (D-06 run 1 vs run 2, plus run 2 vs run 3). Plan-checker: revision rounds 1-2 applied; final pass pending. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…3), VPN closure, CLAUDE.md edits Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
- Plans: D-06 as amended by Glenn 2026-10-12 (runs 2 and 3 identical but for the run date; run 1 vs 2 only the first-run book effect); VPN closure (8 passed + 6 skipped, the skips recorded as unverified); CLAUDE.md edits and the migrate-script deletion cited as approved; the full suite after the registry epoch; a .gitkeep reliance check; M8 placement as DECISIONS G-15; the tag truth marked human-gated. - CONTEXT: decision index in the gate's bullet form; the dropped items tagged [informational]. Decision coverage 6/6; post-planning gaps 6/6 covered. - RESEARCH: open questions marked resolved. - ROADMAP: 08.5 plan list with wave notes. STATE: planned, next execute. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…-10-10) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…s complete: 12 phases, 78 plans) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…t (D-04) - pyproject.toml: register the realdata marker - tests/conftest.py: _BUILT_DATA and a trylast collection hook that skips realdata tests without a build, or raises UsageError when -m selects only them - test_platform_plotting.py: mark test_real_monthly_raw_shape Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
… and the 08.3 migration script (D-04) - realdata marker on 14 tests (7 pnl_returns, 5 spine, 1 driver, 1 plotting); _require_month_end_columns skips the month-end tests on an --allow-missing build - retire 12 tests with dated notes: 11 in pnl_returns (the 47-column blob pin, 3 lag-entry params, 6 migration-script guards, the pit_08.3 before pin) and the A13 change-point pin in plotting_regime - git rm scripts/migrate_month_end_columns.py (one-off 08.3 migration; approved 2026-10-12); keep migrate_publication_lags.py - backtest_driver: anchor REAL_MONTHLY_FEATURES on the file, not the cwd Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…uite data-independent (D-04) - retire 11 record-pin tests with dated notes and named synthetic twins: recursion (3), sojourn_lag (3), report_serving (1 class), scoreboard (1 class), weekly_no_regime (1), gate_tiers (1), honesty_registry (the floor-at-38 pin) - add test_live_ledger_starts_with_a_provenance_header, which names no trial count - delete the conftest pre-8.1 checkpoint fixtures (their only user is retired) - suite: 2261 collected with data; 2247 passed, 14 skipped with data/ and outputs/ moved aside, on pandas 3 (3.11) and pandas 2.3 (3.10) Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…d outputs/, drop the gold-rollback helper - git rm -r data outputs (88 files, 5 pickles among them); /data/ and /outputs/ ignored wholesale - delete tests/unit/tracked_record.py; its two call sites read load_platform_config() - suite green with no data: 2247 passed, 14 skipped Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
Glenn's D-03 choice: the 48-trial ledger moves byte-identical to registry/archive/trials-pre-8.5-reset.jsonl (cmp against 4e49652), and registry/trials.jsonl restarts as one provenance header with prior_genuine_trials=0. ADR-0005 records the reset, the deflation caveat and the hurdle table. Overwrite approved by Glenn on 2026-10-12. Full suite on the empty ledger: 2247 passed, 14 skipped; nothing retired. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
- nbconvert --clear-output on P1-P9 (2.88 MB -> 150 KB); cell text unchanged. - README: "Fresh clone (start here)" run-sheet (D-06 as amended), scoreboard "not yet measured" note (D-05), clear-outputs line, and the retrieval command for the pre-reset tag in place of the old clean-slate order. - CLAUDE.md: the four edits Glenn approved on 2026-10-12 (realdata run line, backtest wording, nothing-generated-tracked rule, ADR-0005 exception). - docs/offline_snapshots.md: snapshots are local only now that data/ is ignored. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
- Phase 9 becomes an umbrella: depends on 08.5, no pinned test count, no retired reference numbers; the first budgeted run (9.4) sets the new baseline with the ADR-0005 epoch note - Phases 9.1-9.8 (M0..M7) with goal, module files, notebook, numbered gates and at most 3 plans; M8 moves with its first importer (G-15) - Progress order and table rows 9.1-9.8; MIG-01 reworded and mapped to Phase 9 (9.1-9.8) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…LE-MAP, MIGRATION-PLAN - DECISIONS: new rows G-14 (clean-slate reset), G-15 (Phase 9 order, M8 with M1/9.2 and M3/9.4) and H-12 (registry true-zero epoch, ADR-0005); H-05, E-11, G-10 and D-09 changed 2026-10-10 with the old text struck; L2-recipe and CR-02 mentions point at 9.7 and 9.6 - Guide: "Learnings from 8.1-8.5"; phases 9.1-9.8 mapped in sections 3 and 4; gate 2 adds the realdata tests; section 6 sizes re-measured 2026-10-12; section 7 retargeted (CR-03 resolved by 08.1) - MODULE-MAP: Phase column, M8 moves with its first importer, realdata gate sentence, record tests retired in 08.5 - MIGRATION-PLAN: "Status 2026-10-10" box, no pinned test count, the section 6 numbers retired as history; Exit lines unchanged Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…ferent lag table" The message said "<marker> is records a different lag table" (seen on Glenn's Mac 2026-10-12). The test now pins both phrasings. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
The state tool (M0) reads total_trial_count, so honesty/registry.py moves at its real first importer; budgets and DSR stay with M3 (9.4). Glenn agreed 2026-10-12. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…ing after the reset - New platform/code_version.py: "Code: <branch> @ <sha>[ + uncommitted changes] — <checkout>", with git run from the package's own folder. Printed first by the build script and the weekly, serving, state and evaluation-report CLIs; pytest prints it next to its result line (the terminal summary, which -q keeps). - Scoreboard: a missing run record is "no budgeted run has written its record here yet" at INFO, not a WARNING about "tracked backtest outputs". - DECISIONS G-16; one line each in README and CLAUDE.md. Suite 2251 passed + 14 skipped (no data) on pandas 3 and pandas 2. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
… passed + 6 month-end skips) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…tform" notebook outputs Verifier: human_needed, 11/12 (the 6 month-end realdata tests stay unverified until an off-VPN build; the gate ran in place rather than from a fresh clone, Glenn's choice). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…oard incomplete vs unmeasured, realdata guard) - code_version: unknown git status is not shown as clean; an untracked installed copy does not borrow the checkout's commit; dirty = tracked or new files under src/ scripts/ config/; never raises on OSError/timeouts (5 s). Isolated tests (no global git config). - scoreboard: a KPI table without its curves (or missing a leg) is "incomplete" at WARNING. - conftest: the realdata UsageError only when -m selects realdata; guarded summary import. - registry rows record the commit of the checkout the code runs from. - realdata rebuild-equality test no longer needs month-end columns (runs on VPN builds). - the three data scripts print the code line too. Suite 2257 passed + 14 skipped (no data) on pandas 3 and pandas 2. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
…verified 11/12) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_018uY2PDfrfhs9KL6vSbBAvB
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
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.
No description provided.