perf: reduce Process URI estimation overhead

project/TICKETS.md

@@ -13,5 +13,6 @@ analysis artifacts.
 | **ticket-005** | [`README.md`](./ticket-005/README.md) | [`preprompt.md`](./ticket-005/preprompt.md) | - |  [`ai-codex.md`](./ticket-005/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-005/ai-codex-logs.txt) | [`changelog.md`](./ticket-005/changelog.md) |
 | **ticket-006** | [`README.md`](./ticket-006/README.md) | [`preprompt.md`](./ticket-006/preprompt.md) | - |  [`ai-codex.md`](./ticket-006/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-006/ai-codex-logs.txt) | [`changelog.md`](./ticket-006/changelog.md) |
 | **ticket-007** | [`README.md`](./ticket-007/README.md) | [`preprompt.md`](./ticket-007/preprompt.md) | - |  [`ai-codex.md`](./ticket-007/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-007/ai-codex-logs.txt) | [`changelog.md`](./ticket-007/changelog.md) |
+| **ticket-008** | [`README.md`](./ticket-008/README.md) | [`preprompt.md`](./ticket-008/preprompt.md) | - |  [`ai-codex.md`](./ticket-008/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-008/ai-codex-logs.txt) | [`changelog.md`](./ticket-008/changelog.md) |
 | **ticket-009** | [`README.md`](./ticket-009/README.md) | [`preprompt.md`](./ticket-009/preprompt.md) | - |  [`ai-codex.md`](./ticket-009/ai-codex.md) |  [`ai-codex-logs.txt`](./ticket-009/ai-codex-logs.txt) | [`changelog.md`](./ticket-009/changelog.md) |
 <!-- AUTO:TICKET_INDEX:END -->

project/ticket-008/README.md

@@ -0,0 +1,28 @@
+# Ticket 008: Reduce Process URI estimation overhead
+
+- **ID**: ticket-008
+- **Owner**: unresolved:human
+- **Status**: IN_PROGRESS
+- **Workflow state**: PUBLICATION
+- **Created**: 2026-08-25
+
+## Goal and scope
+
+Reduce the observer effect of Process URI estimation while preserving bounded,
+append-only evidence. Use a one-second default sample interval and read only
+the tail of the event log when extending its hash chain.
+
+## Acceptance criteria
+
+- [x] AC-01: Scope is approved by the user's request to measure and reduce
+  Subactor resource consumption.
+- [x] AC-02: Command and PID observation default to one sample per second.
+- [x] AC-03: Appending an event does not scan the full historical JSONL file.
+- [x] AC-04: Tests prove the new defaults and bounded tail read.
+- [x] AC-05: Pytest and governance checks pass.
+
+## Participants
+
+- Human participant: user:tom; authorization is recorded by the originating
+  request and no synthesized user file was created.
+- Agent participant: [ai-codex.md](ai-codex.md)

project/ticket-008/ai-codex-logs.txt

@@ -0,0 +1,8 @@
+2026-08-28 validation
+
+- `PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=src python -m pytest -q`: PASS
+  (10 passed).
+- `PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=src ./project/governance-check.sh`:
+  PASS (0 errors, 0 warnings).
+- The four implementation/test blobs match the original PR #1 head exactly.
+- Plan commit `a67e364` precedes implementation commit `391d5af`.

project/ticket-008/ai-codex.md

@@ -0,0 +1,39 @@
+---
+participant-id: agent:codex
+participant: codex
+role: agent
+ticket: ticket-008
+---
+# Participant: codex (AI agent)
+
+## Understanding
+
+The 10 Hz default consumed about 1.62 CPU seconds during a six-second sample;
+the same workload at 1 Hz consumed about 0.36 CPU seconds, a reduction near
+78 percent. Event append also rereads all prior evidence and grows linearly.
+
+## Execution plan
+
+1. Change observer defaults from 10 Hz to 1 Hz.
+2. Resolve the prior event from a bounded tail read.
+3. Add regression tests for both resource guardrails.
+4. Run pytest and governance, then publish through validator-agent.
+
+## Actual changes
+
+- Initialized the bounded ticket and recorded SESSION_EXECUTION_AUTHORIZATION
+  from the request to execute this work.
+- Changed command and PID sampling defaults from 10 Hz to 1 Hz.
+- Replaced the linear prior-event scan with a positional read capped at the
+  final 4096 bytes while preserving append-only hash-chain semantics.
+- Preserved the original implementation blobs exactly while rebuilding the
+  delivery as separate plan, implementation and evidence commits.
+- Passed ten functional tests and the complete governance check locally.
+
+## Blockers
+
+- None inside the recorded intent; proceed without a second confirmation.
+- New authority remains required for destructive action, secret access, new
+  external coordination or material objective expansion. Protected delivery
+  may be invoked without another prompt when publication is in scope; its
+  exact-head trusted approval remains independent evidence.

project/ticket-008/changelog.md

@@ -0,0 +1,10 @@
+# Ticket Changelog (ticket-008)
+
+- Reduce command and PID observation defaults from 10 Hz to 1 Hz.
+- Resolve the prior JSONL event with a bounded final-chunk positional read.
+- Add regression coverage for the sampling and append-cost guardrails.
+
+## [0.1.0] - 2026-08-25
+
+- Initial governance scaffold created.
+- No human participant identity or content was generated.

project/ticket-008/intent.json

@@ -0,0 +1,71 @@
+{
+  "schema": "new-project.intent/v3",
+  "ticket": "ticket-008",
+  "summary": "Reduce Process URI estimation overhead",
+  "workstream": "application",
+  "classification": {
+    "kind": "SERVICE",
+    "priority": "P2",
+    "origin": "health"
+  },
+  "allowedPaths": ["project/ticket-008/**", "TODO.md", "project/TICKETS.md", "src/**", "tests/**"],
+  "delivery": {
+    "acceptedBaseSha": "4ff7e4d6e288697bac10e2008524f65c0e8280de",
+    "targetBranch": "main",
+    "outcome": "Process URI estimation samples at 1 Hz by default and extends its event hash chain from a bounded tail read.",
+    "nonGoals": [
+      "Change Process URI authorization or scheduling authority.",
+      "Change the persisted sample or event schema."
+    ],
+    "complexity": "S",
+    "estimatedMinutes": 25,
+    "budgets": {
+      "maxImplementationFiles": 5,
+      "maxAffectedComponents": 1,
+      "maxPublicInterfaceChanges": 0,
+      "maxRuntimeDependencies": 0
+    },
+    "architecture": {
+      "status": "accepted",
+      "decision": "Keep estimation advisory and append-only; reduce sampling frequency and resolve only the final JSONL record with positional reads.",
+      "components": [
+        {
+          "name": "estimation-runtime",
+          "paths": ["src/**", "tests/**"]
+        }
+      ],
+      "responsibilityChanges": false,
+      "interfaceChanges": [],
+      "dataChanges": [],
+      "ui": {
+        "impact": "none",
+        "states": [],
+        "evidence": []
+      },
+      "rollback": "Restore the 0.1 second defaults and sequential last-event scan; stored JSONL remains compatible."
+    },
+    "runtimeDependencies": [],
+    "validation": [
+      {
+        "criterion": "AC-02",
+        "commands": ["PYTHONPATH=src python -m pytest"],
+        "evidence": "Signature regression test asserts one-second defaults."
+      },
+      {
+        "criterion": "AC-03",
+        "commands": ["PYTHONPATH=src python -m pytest"],
+        "evidence": "Tail-read regression test limits the final append to one 4096-byte pread."
+      },
+      {
+        "criterion": "AC-05",
+        "commands": ["PYTHONPATH=src ./project/governance-check.sh"],
+        "evidence": "Governance report must return GOV-PASS."
+      }
+    ]
+  },
+  "forbiddenPaths": ["project/ticket-*/user-*.md"],
+  "stacks": [],
+  "dependsOn": [],
+  "conflictsWith": [],
+  "integrationTicket": null
+}

project/ticket-008/preprompt.md

@@ -0,0 +1,15 @@
+# Ticket preprompt
+
+- **Task ID**: ticket-008
+- **Task title**: Reduce Process URI estimation overhead
+- **Created**: 2026-08-25T17:11:14Z
+
+Keep executable implementation outside this governance/evidence directory.
+Read a human-owned user-*.md file only when one exists.
+The request to execute this work creates SESSION_EXECUTION_AUTHORIZATION;
+proceed within the recorded intent without a redundant confirmation prompt.
+Require new authority for destructive action, secrets, external coordination,
+or material objective expansion. When publication is in scope, session
+authorization permits the declared protected delivery process and its merge
+after exact-head trusted approval without another prompt. Session prose is
+never approval evidence and the agent must not merge directly.

src/estimation/cli.py

@@ -24,7 +24,7 @@ def _context(parser: argparse.ArgumentParser) -> None:
     parser.add_argument("--correlation-id", default=os.getenv("SUBACTOR_CORRELATION_ID"))
     parser.add_argument("--store", default=DEFAULT_STORE)
     parser.add_argument("--events", default=DEFAULT_EVENTS)
-    parser.add_argument("--interval", type=float, default=0.1)
+    parser.add_argument("--interval", type=float, default=1.0)
 
 
 def _parser() -> argparse.ArgumentParser:

src/estimation/monitor.py

@@ -106,7 +106,7 @@ def measure_command(
     process_key: str | None = None,
     ticket_id: str | None = None,
     correlation_id: str | None = None,
-    interval_seconds: float = 0.1,
+    interval_seconds: float = 1.0,
 ) -> Sample:
     argv = [str(item) for item in command]
     if not argv:
@@ -157,7 +157,7 @@ def observe_pid(
     process_key: str | None = None,
     ticket_id: str | None = None,
     correlation_id: str | None = None,
-    interval_seconds: float = 0.1,
+    interval_seconds: float = 1.0,
     duration_seconds: float = 10.0,
 ) -> Sample:
     root = psutil.Process(int(pid))

src/estimation/store.py

@@ -3,6 +3,7 @@
 from contextlib import contextmanager
 import hashlib
 import json
+import os
 from pathlib import Path
 from typing import Any, Iterator
 
@@ -42,14 +43,21 @@ def _append_line(path: Path, payload: dict[str, Any]) -> None:
 
 
 def _last_event(handle: Any) -> tuple[int, str]:
-    handle.seek(0)
-    last: dict[str, Any] | None = None
-    for raw_line in handle:
-        line = raw_line.strip()
-        if line:
-            last = json.loads(line)
-    if last is None:
+    handle.flush()
+    offset = os.fstat(handle.fileno()).st_size
+    suffix = b""
+    while offset > 0:
+        size = min(4096, offset)
+        offset -= size
+        suffix = os.pread(handle.fileno(), size, offset) + suffix
+        stripped = suffix.rstrip()
+        separator = max(stripped.rfind(b"\n"), stripped.rfind(b"\r"))
+        if separator >= 0:
+            last = json.loads(stripped[separator + 1 :].decode("utf-8"))
+            return int(last["sequence"]), str(last["eventHash"])
+    if not suffix.strip():
         return 0, ZERO_HASH
+    last = json.loads(suffix.strip().decode("utf-8"))
     return int(last["sequence"]), str(last["eventHash"])
 
 

tests/test_estimation.py

@@ -1,5 +1,6 @@
 from __future__ import annotations
 
+import inspect
 import json
 import os
 from pathlib import Path
@@ -10,6 +11,7 @@
 from estimation.monitor import measure_command, observe_pid
 from estimation.stats import aggregate_samples, estimate_workload
 from estimation.store import append_sample, load_samples
+import estimation.store as store_module
 
 
 def _sample(duration: float, memory: int, *, outcome: str = "succeeded"):
@@ -72,6 +74,31 @@ def test_event_store_builds_a_hash_chain(tmp_path: Path) -> None:
     assert all(row["secretMaterialIncluded"] is False for row in rows)
 
 
+def test_monitor_defaults_to_one_hertz() -> None:
+    assert inspect.signature(measure_command).parameters["interval_seconds"].default == 1.0
+    assert inspect.signature(observe_pid).parameters["interval_seconds"].default == 1.0
+
+
+def test_event_append_reads_only_bounded_tail(tmp_path: Path, monkeypatch) -> None:
+    store = tmp_path / "samples.jsonl"
+    events = tmp_path / "events.jsonl"
+    for index in range(1000):
+        append_sample(_sample(float(index + 1), index + 1), store, events)
+
+    original_pread = store_module.os.pread
+    bytes_read = 0
+
+    def measured_pread(fd: int, size: int, offset: int) -> bytes:
+        nonlocal bytes_read
+        chunk = original_pread(fd, size, offset)
+        bytes_read += len(chunk)
+        return chunk
+
+    monkeypatch.setattr(store_module.os, "pread", measured_pread)
+    append_sample(_sample(1001.0, 1001), store, events)
+    assert bytes_read <= 4096
+
+
 def test_report_and_workload_estimate_use_successful_p90() -> None:
     samples = [_sample(1.0, 100), _sample(2.0, 200), _sample(3.0, 300), _sample(20.0, 999, outcome="failed")]
     report = aggregate_samples(samples)