From 245543bda9c7db2e42ad49179accdcceabbe5256 Mon Sep 17 00:00:00 2001 From: "russell@unturf.com" Date: Sun, 10 May 2026 16:57:19 -0400 Subject: [PATCH] =?UTF-8?q?update=20everything=20with=20today's=20learning?= =?UTF-8?q?s=20=E2=80=94=20math=20review=20forward=20+=20Phase=203=20dry-r?= =?UTF-8?q?un=20amend?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Two coupled doc updates capturing today's session state: 1. #000036 status pin — math review in flight with dav1d - Ticket status line: 'awaits fox math review' → 'pre-review polish pass 8916bf3; math review in flight with dav1d (forwarded 2026-05-10 — Tier 2 bundle)' - TICKETS.md index row mirrors same change - Future shifts can now see review is live, not blocked on fox. 2. #000006 rolling research log — 2026-05-10b amend - Fourth qualitatively different experimental shape: Prometheus-Σ dry-run simulator (joining random-word, witness-sweep, warrant-chain) - Captures the five scheduler-calibration findings (F1-F5) from bench/scripts/prometheus_sigma_sweep_dryrun.py: - F1: chunk_size = Hermes concurrency, not pool size - F2: capital_cost must split by audit_mode (CP=0.05 vs STRICT=1.0); flat-1.0 blocks every allocation - F3: τ_qa must split by audit_mode (1d for kernel-only, 7d for LLM-witness); single-τ hides CP-rows - F4: Target B headline = 4.40 percent of docs are canonical-shape candidates (~152K across the corpus) - F5: quarantined-row veto exercises end-to-end on real-corpus data, no fixture-only mocking - Updates the distinct-signal table to four rows - Cross-references #000037 §22 for the full per-shard log Phase 3 scheduler (when it ships) inherits F1-F5 as known-good defaults — the dry-run is the calibration substrate the eventual implementation will reference for choice justification. Doc-only updates; no schema, no governance hash, no code change. --- docs/TICKETS.md | 2 +- .../ticket-000006-bench-emergent-findings.md | 121 ++++++++++++++++++ .../ticket-000036-t3-per-window-bound.md | 2 +- 3 files changed, 123 insertions(+), 2 deletions(-) diff --git a/docs/TICKETS.md b/docs/TICKETS.md index 03605ec..b32c88d 100644 --- a/docs/TICKETS.md +++ b/docs/TICKETS.md @@ -81,7 +81,7 @@ Newest first. Update on every open/close. | #000039 | Optional `sqlite-vec` retrieval backend (A/B vs FTS5, hybrid not replacement) | open · awaiting go/no-go (doc-only Phase 0) | 2026-05-09 | — | | #000038 | Phase 4 content acquisition — proprietary textbook license decisions for warrant coverage | closed · obviated 2026-05-10 by alias-substitution sprint under #000031 (74 rows in #000041 + 13 rows in #000042); 92/92 records now resolve. Residue (multilingual PD, Hilbert-Ackermann OCR, Knuth permission, personal-copy path B) preserved as design log §8 | 2026-05-09 | — | | #000037 | Prometheus-Σ recursive falsification controller (bicameral substrate) | in progress · Phase 0 (doc) + Phase 1 (controller `arborist/substrate/prometheus.py`) + Phase 2 (`controller_events` sibling table + advisory writes) all landed 2026-05-10; Phase 3 sleep-sweep scheduler NOT landed — instead a read-only dry-run simulator + 5 design findings in §22 | 2026-05-09 | — | -| #000036 | T3 per-window covert-channel budget bound | in progress · Phase 1 (formal derivation + calculator) landed 2026-05-10; awaits fox math review | 2026-05-09 | — | +| #000036 | T3 per-window covert-channel budget bound | in progress · Phase 1 (formal derivation + calculator) landed 2026-05-10; pre-review polish `8916bf3`; **math review in flight with dav1d** (forwarded 2026-05-10 — Tier 2 bundle) | 2026-05-09 | — | | #000035 | PRG choice for φ_PRG (HMAC-SHA-512 expansion) | in progress · Phase 1 landed 2026-05-10; v7 §9.10 amendment awaits maintainer review | 2026-05-09 | — | | #000034 | Hessian alignment under φ_linear | in progress · Phase 1a landed 2026-05-10 (synthetic-ablation probe + KAT fixture); Phase 1b parks for v7 ramp-up | 2026-05-09 | — | | #000033 | Claim-pack pillar VII (combinatorics) | closed · landed 2026-05-09 (live in shard 000.db; lift verified) | 2026-05-09 | — | diff --git a/docs/tickets/ticket-000006-bench-emergent-findings.md b/docs/tickets/ticket-000006-bench-emergent-findings.md index 3969d4b..b46bfb7 100644 --- a/docs/tickets/ticket-000006-bench-emergent-findings.md +++ b/docs/tickets/ticket-000006-bench-emergent-findings.md @@ -760,3 +760,124 @@ were updated to point at these landings (`28b57da`, `0b038f1`). Re-runnable via the bench-emergent harness when corpus shape shifts and a measurable fire rate is expected. + +## Amend — 2026-05-10b (Prometheus-Σ Phase 3 dry-run: scheduler calibration findings) + +A **fourth qualitatively different experimental shape**: + +- random-word (honesty under no ground truth) +- witness-sweep (capability under canonical ground truth) +- warrant-chain (rescue activation under chain-backed ground truth) +- **Prometheus-Σ dry-run** ← this amend (controller scoring under real shard state, no LLM/mutation) + +### Setup + +#000037 Phase 1 (`arborist/substrate/prometheus.py`) + Phase 2 +(`controller_events` sibling audit table + advisory writes) landed +2026-05-10. Phase 3 (sleep-sweep scheduler) is deliberately not +landed — instead a read-only **dry-run simulator** at +`bench/scripts/prometheus_sigma_sweep_dryrun.py` (`make +prometheus-sweep-dryrun`) drives the Phase 1 controller against +real shard data, no LLM calls, no mutations. Catches +**scheduler-calibration failures** — controller verdicts that +look correct in unit tests but fail under real-corpus statistics. + +### Aggregate + +```text +4 379 branches scored, 55 ms wall, 12.5 µs/branch pure Python +``` + +Controller-latency is non-binding for Phase 3; the witness fan-out +(Hermes calls) will be the dominant cost when it ships. + +### Five scheduler-calibration findings + +**F1 — Chunk-size dominates Kelly threshold.** First iteration +chunked candidates 64-at-a-time → **100 % DEFERRED**. Kelly's +`f_i = max(0, (p_i·b − q_i)/b)` requires `p_i > 0.5`; softmax over +64 branches never gives any single branch that much mass. +**Phase 3 fix:** chunk size = Hermes concurrency (4 today; +governance parameter going forward), not the candidate pool size. + +**F2 — Flat `capital_cost` blocks every allocation.** Iteration 1 +assigned `capital_cost=1.0` to every branch; combined with small +audit-mode-based Δ5F deltas (±0.05–0.10), every utility came out +negative — DEFERRED or REJECT for every chunk. Iteration 2 split +the cost class: + +```text +Target A (providence_cache): + CANONICAL_PROJECTION=0.05 UNGROUNDED=0.4 + HYBRID=0.8 STRICT=1.0 + +Target B (documents): + HEAD-only freshness probe=0.02 + canonical-shape witness=0.05 +``` + +Surfaced real ACCEPT/MARGINAL signal immediately. **Phase 3 fix:** +either split `capital_cost` into `kernel_cost` + `llm_cost` in a +v2 dataclass, or compute `capital_cost` as the expected cost given +which witness paths actually fire. + +**F3 — τ_qa = 7d filters out every `CANONICAL_PROJECTION` row.** +All 29 CP rows in `qa.db` are ≤ 7 days old — they're the recent π* +graduations from #000027 / #000030 / #000032. At τ_qa = 7d, zero +CP rows surface → zero ACCEPT chunks → zero high-value sleep work +discovered. At τ_qa = 1d, 9 CP rows surface and 4 chunks return +MARGINAL. **Phase 3 fix:** split τ_qa by audit_mode. Kernel-only +modes (CP) take a short τ (1d default — cheap to re-probe); +LLM-witness modes (STRICT/HYBRID/UNGROUNDED) take a longer τ +(7d default — re-witness is expensive). + +**F4 — Target B canonical-shape detection is the real headline.** +4.40 % of sampled documents (n=2000) contain math/logic/time- +series canonical-shape statements. Extrapolated to **~152 609 +canonical-probe candidates** across 3 468 392 docs in +`~/.arborist/shards`. Controller correctly returns MARGINAL on 83 +of 500 Target B chunks (the regex-prefilter hits). **Phase 3 fix:** +the MARGINAL queue from Target B becomes the funnel for 5F +falsification-fixture mining (§3 "Divergence → candidate +falsification fixture"). At Hermes concurrency = 4 a full Target B +sweep is still ~38K rounds even after the prefilter — Phase 3 +needs a sustained-throughput floor + per-window cap, not a +one-shot burst design. + +**F5 — Quarantined rows correctly veto.** The two +`falsification_state='quarantined'` rows in `qa.db` are flagged as +`cache_drift` hard-vetoes by Target A's branch synthesis. The +veto path exercises end-to-end against real corpus data — no +fixture-only mocking. + +### Updated distinct-signal table + +| Source | Catches | Repair path | +|---|---|---| +| Random-word emergent | honesty failures (false-STRICTs) | verifier ladder hardening | +| Witness-sweep | capability failures (LLM gives wrong answer) | prompt eng or fine-tune | +| Phase 3 warrant-chain | rescue-mechanism activation conditions | corpus shape evolution (NOT fixture engineering) | +| Prometheus-Σ dry-run | **scheduler-calibration failures** | tune chunk size, cost weights, τ thresholds *before* Phase 3 scheduler ships | + +### What the dry-run validates end-to-end + +1. **Phase 1 controller pure function.** `score_controller_state()` + returns a defensible ControllerDecision on every branch type + the live corpus contains. +2. **Phase 2 advisory writes.** `controller_events` sibling table + accepts emitted decisions; audit chain not affected. +3. **Five hard-veto classes fire correctly.** Cache-drift veto + (F5), DEFERRED on budget-zero (F1), REJECT on negative utility + (F2), MARGINAL queue for sleep (F4), ACCEPT for high-value + short-τ work (F3). + +### Bench artifact + +`bench/results/prometheus-sigma-sweep-dryrun-2026-05-10.md` — +detailed per-shard numbers + iteration log. Regenerated by +`make prometheus-sweep-dryrun`. + +Phase 3 scheduler (when it ships) inherits F1-F5 as known-good +defaults; the dry-run is the calibration substrate Phase 3 will +reference for choice justification. + diff --git a/docs/tickets/ticket-000036-t3-per-window-bound.md b/docs/tickets/ticket-000036-t3-per-window-bound.md index 670dd35..1d0f2e4 100644 --- a/docs/tickets/ticket-000036-t3-per-window-bound.md +++ b/docs/tickets/ticket-000036-t3-per-window-bound.md @@ -1,6 +1,6 @@ # Ticket #000036 — T3 per-window covert-channel budget bound -**Status:** in progress · Phase 1 (formal derivation + calculator) landed 2026-05-10; awaits fox math review of constants; Phase 2 (empirical tightening) parks for v7 deployment data +**Status:** in progress · Phase 1 (formal derivation + calculator) landed 2026-05-10; pre-review polish pass committed 2026-05-10 (`8916bf3`); **math review in flight with dav1d** (forwarded 2026-05-10 Asia/Kuala_Lumpur — Tier 2 bundle: `t3-bound.md` + `soft-hash-analysis.md` + `t3_bound_calculator.py` + `test_t3_bound_calculator.py` + this ticket); empirical tightening tracked separately under #000043 **Opened:** 2026-05-09 **Scope:** Derive an explicit closed-form upper bound on the covert- channel capacity under threat model T3 (hyperparameter adversary),