specs: mark five specs ready (issues #4 to #8)

docs/journal.md

@@ -36,3 +36,7 @@ Tested the 'stale March build' explanation: built the kernel and Python package
 ## 2026-10-30 14:49 EDT (at b8f9fc3)
 
 Planned the benchmark/foundations work as specs. Wrote five specs with strict-xfail failing tests (bench-result-equality, bench-interleaved-timing, bench-env-capture, bench-phase-breakdown, adr-foundation), a method v1 proposal and a ranked roadmap of the rest. Each test was shown failing for the right reason (missing module) and shown satisfiable with a throwaway reference implementation, which was then deleted. Caught three test bugs on the way: a pandas NaN-vs-null assumption, two ADR tests that passed vacuously on an empty directory, and over-long lines that would have failed CI lint. Found that Polars and FlowState outputs already agree on a small dataset, but with different timestamp dtypes (Int64 vs timestamp[ns, UTC]). FlowState is not onboarded to autopilot (no .autopilot/), so the handshake was run against drillpad's repo.yaml and is provisional.
+
+## 2026-10-30 16:52 EDT (at 2d3e6df)
+
+Filed the five spec issues (#4 bench-result-equality, #5 bench-interleaved-timing, #6 bench-env-capture, #7 bench-phase-breakdown, #8 adr-foundation). Ryan ran gh issue create himself because the call was blocked in my session. Specs marked ready with their issue numbers; ap specs shows no problems; handshake still ACCEPTED (provisional: FlowState has no .autopilot yet).

specs/adr-foundation.md

@@ -15,7 +15,8 @@ non_goals:
 risk_tier: 1
 max_runs: 3
 autonomy: inherit
-status: draft
+status: ready
+issue: 8
 ---
 
 ## Why
@@ -33,3 +34,4 @@ consequences as a separate, reviewable record, and a place to add the next one.
 - The test rejects the word "outperform" and any "N x faster/slower" string.
 
 ## Log
+- 2026-10-30 ready, issue #8

specs/bench-env-capture.md

@@ -15,7 +15,8 @@ non_goals:
 risk_tier: 2
 max_runs: 3
 autonomy: inherit
-status: draft
+status: ready
+issue: 6
 ---
 
 ## Why
@@ -38,3 +39,4 @@ pages or ASLR. A server run is only credible if those are captured automatically
 - `memory_gb`, `numa_nodes`, `rustc`, `cpu_model` use whatever stdlib or `subprocess` call works on Linux and macOS, else `"unknown"`
 
 ## Log
+- 2026-10-30 ready, issue #6

specs/bench-interleaved-timing.md

@@ -15,7 +15,8 @@ non_goals:
 risk_tier: 2
 max_runs: 3
 autonomy: inherit
-status: draft
+status: ready
+issue: 5
 ---
 
 ## Why
@@ -34,3 +35,4 @@ within a run and reporting intervals is the minimum a skeptic expects.
 - Median is the only `stat` required; reject others with `ValueError`.
 
 ## Log
+- 2026-10-30 ready, issue #5

specs/bench-phase-breakdown.md

@@ -15,7 +15,8 @@ non_goals:
 risk_tier: 2
 max_runs: 3
 autonomy: inherit
-status: draft
+status: ready
+issue: 7
 ---
 
 ## Why
@@ -38,3 +39,4 @@ know how much of the measured gap is wrapper overhead versus kernel time.
 - Time one call per sample with `time.perf_counter`; no need for the interleaved harness yet.
 
 ## Log
+- 2026-10-30 ready, issue #7

specs/bench-result-equality.md

@@ -15,7 +15,8 @@ non_goals:
 risk_tier: 2
 max_runs: 3
 autonomy: inherit
-status: draft
+status: ready
+issue: 4
 ---
 
 ## Why
@@ -41,3 +42,4 @@ A skeptic will say so, and the published post is currently open to that attack.
 - Style: type hints, short functions, `from __future__ import annotations`, ruff clean (`ruff check benchmarks tests`).
 
 ## Log
+- 2026-10-30 ready, issue #4