Skip to content

sync (EOD 2026-07-28): status + vault + session handoff + open scope decision #22

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Ansul-S wants to merge 2 commits into
base: main
Choose a base branch
from
Open

sync (EOD 2026-07-28): status + vault + session handoff + open scope decision #22

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
2858eed
sync(2026-07-28): status + vault to the PR #21 merged state
Ansul-S Jul 28, 2026
8c58406
sync(eod-2026-07-28): handoff + vault + open scope decision
Ansul-S Jul 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 3 additions & 2 deletions CLAUDE.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -142,11 +142,12 @@ archive/ prior-project audit & review (reference only);
- **v3 RE-REGISTRATION SEALED (#2b, 2026-06-19) + M4 DRESS REHEARSAL (2026-06-20).** Finding B made targeted-TLS non-executable → v3 re-registration (DR-002), sealed **confirmer-only** (Arm-B = folded transit likelihood-ratio at common FAR; keystone A6 relaxed). Lever-1b period-FAP cheapening DROPPED (both candidates failed the equivalence gate) → sealed B=1000 bootstrap stands, ρ_d retained; T_red=0.0. Tag `phase1-prereg-v3` (→ `ff869d4b`); v3 manifest `54f06a94…c9b18`. VAL v3 / MATH v1.2 / HYP v2.1 sealed.
- **M4 SINGLE SEALED-TEST RUN EXECUTED — ✅ DONE (2026-06-24). VERDICT: H1 FALSIFIED — compute branch (E1 PASS, E2 FAIL).** One irreversible read (P-5): 15,000 injections (30 cells × 500, literal ≥500/cell). **E1** recall non-inferiority PASS (ΔR̄=−0.48 pp, one-sided 95% lo −0.60 pp; margin −2 pp). **E2** scoped compute FAIL (reduction 24.4%, ratio 0.756, ρ_d=14.4%; target ≥30%). Recall principle holds; compute claim is the falsified branch (un-cheapenable B=1000 period-FAP entry tax). Pre-committed verdict (VAL §7a) applied — a **successful negative Phase I**. Result record: [`research/m4_evaluation/M4_TEST_RESULT.md`](./research/m4_evaluation/M4_TEST_RESULT.md); artifacts `data/manifests/m4/test_run/`. TEST conditioning (first-touch) via frozen Stage-0: `research/m1_conditioning/condition_test_hosts.py` (80/80 hosts).
- **TEST read exactly once (P-5); will not be read again (P-2: v3 is final; no v4).** Anti-tuning (NN#2) intact end-to-end: `git diff phase1-prereg-v3` over sealed docs + manifests empty; both seals hash-verified in-run + intact; verdict pre-committed before the read.
- **⚠️ AUDIT REMEDIATION (2026-07-19 → 07-27, DR-003 — branch `phase1/audit-remediation`): the sealed M4 "H1 FALSIFIED — compute" verdict above is SUPERSEDED.** The recorded E2 "FAIL" (sealed timing = 12 stars × 1 repeat; ratio CI [0.42, 1.14] undecided) was withdrawn, and the frozen-rule E2 re-measurement is now **COMPLETE**. **CORRECTED PHASE-I VERDICT: E1 PASS (robust; host-cluster lo95 −0.82 pp; host count corrected to 40) · E2 INCONCLUSIVE** — frozen §6 rule on 300 injections × 5 repeats / 39 hosts → compute ratio **0.727** (27.3% reduction; target ≥30%), host-clustered CI **[0.636, 0.826]** straddles the 0.70 boundary; ρ_d 11.6%, f_p 23.7%, **π\*≈0.489 ≫ π≈0.03** (not a survey-scale compute-saver). Recall principle validated; compute claim neither confirmed nor falsified at decision-grade precision. Robust to a Low-Power-Mode timing window (leave-out 0.713, still INCONCLUSIVE; erratum §5.1). Edge control (2026-07-19): P=0.5 d gains/losses are **epoch-predicate** phenomena, not detection power. **Wave 0 status:** V-1 (resume guard + tests) ✅ committed `2941175`; V-2 (E2 campaign 300/300) ✅; V-3 (erratum §5/§7 filled) ✅; V-4 (verdict propagated to paper §3.3/§3.4/abstract/§4/§5, M4_TEST_RESULT addendum, this bullet, vault) ✅; **V-5 ✅ (PR #18 merged).** **WAVE 0 COMPLETE. Wave 1 no-compute cluster ✅ (RES-2/3/5/7/8, PUB-6; PR #19 merged)** — E1 robust to KM period weighting (ΔR̄ −0.16 pp) + losses are epoch-predicate; **RES-4/RES-6 + a small TLS-epoch re-run queued for a compute window.** Authorities: `docs/decisions/DR-003_E2_REMEASUREMENT.md` · `research/m4_evaluation/M4_ERRATUM_2026-07-19.md` (§5/§7 now filled) · `data/manifests/m4/e2_retiming/e2_retiming_summary.json` · `docs/audits/PROJECT_AUDIT_2026-07-19.md`. Execution plan: `docs/ROADMAP_TO_10.md` (Wave 0 + Wave-1 no-compute cluster done; Wave-1 compute tail + Waves 2–6 follow). **Phase II re-scoped** (bound/benchmark/monotransit; Kepler routing-scaling superseded): `docs/VESPER_PHASE2_PROGRAM.md` (DRAFT, pending DR-004; hard-gated behind the roadmap). This bullet is authoritative and supersedes every verdict line above it.
- **⚠️ AUDIT REMEDIATION (2026-07-19 → 07-27, DR-003 — branch `phase1/audit-remediation`): the sealed M4 "H1 FALSIFIED — compute" verdict above is SUPERSEDED.** The recorded E2 "FAIL" (sealed timing = 12 stars × 1 repeat; ratio CI [0.42, 1.14] undecided) was withdrawn, and the frozen-rule E2 re-measurement is now **COMPLETE**. **CORRECTED PHASE-I VERDICT: E1 PASS (robust; host-cluster lo95 −0.82 pp; host count corrected to 40) · E2 INCONCLUSIVE** — frozen §6 rule on 300 injections × 5 repeats / 39 hosts → compute ratio **0.727** (27.3% reduction; target ≥30%), host-clustered CI **[0.636, 0.826]** straddles the 0.70 boundary; ρ_d 11.6%, f_p 23.7%, **π\*≈0.489 ≫ π≈0.03** (not a survey-scale compute-saver). Recall principle validated; compute claim neither confirmed nor falsified at decision-grade precision. Robust to a Low-Power-Mode timing window (leave-out 0.713, still INCONCLUSIVE; erratum §5.1). Edge control (2026-07-19): P=0.5 d gains/losses are **epoch-predicate** phenomena, not detection power. **Wave 0 status:** V-1 (resume guard + tests) ✅ committed `2941175`; V-2 (E2 campaign 300/300) ✅; V-3 (erratum §5/§7 filled) ✅; V-4 (verdict propagated to paper §3.3/§3.4/abstract/§4/§5, M4_TEST_RESULT addendum, this bullet, vault) ✅; **V-5 ✅ (PR #18 merged).** **WAVE 0 COMPLETE. Wave 1 no-compute cluster ✅ (RES-2/3/5/7/8, PUB-6; PR #19 merged)** — E1 robust to KM period weighting (ΔR̄ −0.16 pp) + losses are epoch-predicate. **RES-4 ✅ + Wave-2/4 closure ✅ (2026-07-28, PR #21 merged):** **RES-4** per-star τ_GP FAP sensitivity on all 1163 cached calibration nulls → at the sealed T₁₄=0.2 d convention, gate flips **0/1126** (arm B, as the driver applies it; Wilson 95% [0, 0.0034]) and **1/1126** (arm C, complete per-star coverage; [0.0002, 0.0050]) — erratum §2.4's masking argument is now **measured**, sealed FAPs reproduced **bitwise 968/968**. New findings: §2.4's own premise is imprecise (M3 used per-star τ for only **22 of 968** overlapping stars — it was overwhelmingly flat-τ too); the lone flip is flat-OPEN→per-star-SHUT (sealed gate slightly *more permissive* on red-noise nulls — benign direction); the masking is bought by the **T₁₄=median convention**, so any future run that duration-matches T₁₄ **must** use per-star τ (24/1126 flips at a counterfactual T₁₄=0.05 d). **MATH-1** π⋆ derived = ρ_d/(f_p(1−ρ)) — code+paper correct, the roadmap's "exact form" rejected (it implies a routed star isn't charged for its own detector); a tautological unit test replaced. **MATH-5** BCa host-cluster lo95 −1.04 pp vs sealed percentile −0.83 pp → sealed interval mildly optimistic, **E1 non-inferiority unchanged**. **MATH-6** N=2 comb degeneracy proven+measured ("argmin→longest-P" only ~74% true; at N=2 the FAP tests event *rarity*, not coherence). **MATH-7** notation cross-reference. **CODE-7** cluster bootstrap vectorized (bit-exact). **DOC-2** [`docs/INDEX.md`](./docs/INDEX.md) · **DOC-3** [`docs/SEAL_CHAIN_POSTMORTEM.md`](./docs/SEAL_CHAIN_POSTMORTEM.md). Post-seal math lives in [`docs/VESPER_MATH_ADDENDUM.md`](./docs/VESPER_MATH_ADDENDUM.md) — a companion doc, never inlined into the sealed MATH file, so `git diff phase1-prereg-v3` stays empty modulo branding (re-verified post-merge: 0 differing lines on all three sealed docs). **Still queued for a compute window: RES-6 (η-paid injection; needs MAST) + a small TLS-epoch re-run.** Authorities: `docs/decisions/DR-003_E2_REMEASUREMENT.md` · `research/m4_evaluation/M4_ERRATUM_2026-07-19.md` (§5/§7 now filled) · `data/manifests/m4/e2_retiming/e2_retiming_summary.json` · `docs/audits/PROJECT_AUDIT_2026-07-19.md`. Execution plan: `docs/ROADMAP_TO_10.md` (Wave 0 + Wave-1 no-compute cluster done; Wave-1 compute tail + Waves 2–6 follow). **Phase II re-scoped** (bound/benchmark/monotransit; Kepler routing-scaling superseded): `docs/VESPER_PHASE2_PROGRAM.md` (DRAFT, pending DR-004; hard-gated behind the roadmap). This bullet is authoritative and supersedes every verdict line above it.
- **IDENTITY (2026-06-30): project rebranded `TRINETRA-X` → `VESPER` and shipped as the first public release `v1.0.0` — branding only; no algorithm, methodology, equation, threshold, result, or figure changed.** Acronym: *Validation Engine for Stellar Photometric Evidence and Recovery*. Folder `~/Desktop/VESPER`; repo `github.com/Ansul-S/VESPER`. ⚠️ Rebranding the sealed docs/manifests **changed their recorded SHA-256 digests by the rename only** — provenance in [`docs/decisions/F1_DECISION_RECORD.md`](./docs/decisions/F1_DECISION_RECORD.md) §5a; original sealed bytes remain intact at tags `phase1-prereg-v2/v3`. A `shasum` against the *old* recorded hashes will mismatch — read §5a first.
- **Phase I COMPLETE / SEALED / FINAL** (M0–M7). **BAH 2026 · PS7 hackathon track — DISCONTINUED (owner decision 2026-07-27); not pursued further.** Round-1 materials remain in `hackathon/` as submitted history (marked discontinued, not deleted); no further hackathon work. The old "Phase II FROZEN until after the hackathon" gating is **void** — Phase II is now gated solely by the `docs/ROADMAP_TO_10.md` completion gate + DR-004 (see the audit-remediation bullet above).
- **REPO REORGANIZED (2026-06-30, structure-only; history preserved):** Phase-I planning docs → [`research/phase1/`](./research/phase1/); session handoffs → [`archive/session_handoffs/`](./archive/session_handoffs/); `NEXT_SESSION_PROMPT.md` untracked + gitignored (local AI-workflow scratch). Root holds only `README.md · CLAUDE.md · VESPER.md · references.bib · .gitignore`. No source code or research content changed.
- **Latest handoff:** [`archive/session_handoffs/SESSION_HANDOFF_2026-07-27.md`](./archive/session_handoffs/SESSION_HANDOFF_2026-07-27.md) (supersedes 2026-07-20). `main` == `origin/main` @ `9b12181`; release `v1.0.0` on the remote; Phase-I verdict corrected (E1 PASS · E2 INCONCLUSIVE); Wave 0 + Wave-1 no-compute cluster merged; hackathon track discontinued. GSD Core is globally installed but **not used** here (no local `.planning/`). Archive material is reference only; prefer repository documents over chat summaries.
- **⚠️ OPEN DECISION carried from 2026-07-28 — project scope + paper framing (NOT decided; settle before resuming task work).** Recommendation on the table: finish Phase I in ~2 weeks (RES-6 → paper → submit) rather than the roadmap's ~50 days; **cut Waves 3 and 6** (package/CLI/container/API-docs/seal-library — none changes a number in the paper); do not start Phase II on momentum; and **reframe the paper around the methodology** — sealed single-shot validation with pre-committed verdicts, demonstrated by the protocol forcing withdrawal of this project's own headline verdict — with the routing result as the case study rather than the point (roadmap **INN-1**, currently scheduled last in Wave 6). **Until the owner rules, `docs/ROADMAP_TO_10.md` stands as written.** If accepted → record as **DR-005** (DR-004 stays reserved for the Phase-II gate) + amend the roadmap. Full text: [`archive/session_handoffs/SESSION_HANDOFF_2026-07-28.md`](./archive/session_handoffs/SESSION_HANDOFF_2026-07-28.md) §6.
- **Latest handoff:** [`archive/session_handoffs/SESSION_HANDOFF_2026-07-28.md`](./archive/session_handoffs/SESSION_HANDOFF_2026-07-28.md) (supersedes 2026-07-27). `main` == `origin/main` @ `77037d4`; release `v1.0.0` on the remote; Phase-I verdict corrected (E1 PASS · E2 INCONCLUSIVE); Wave 0, Wave-1 (less RES-6 + TLS-epoch re-run), half of Wave 2, and CODE-7/DOC-2/DOC-3 merged; hackathon track discontinued. **PR #22 open** (doc-only sync). GSD Core is globally installed but **not used** here (no local `.planning/`). Archive material is reference only; prefer repository documents over chat summaries.

## Conventions

Expand Down
Loading
Loading