Feature plan: Step 1 · Step 2 · Step 3 · Step 4

Debug mode: enabled Effort study: railspike-effort / v3/application_capacity_study; P2 ledger Extended estimates: enabled

Original Query

Choose Option A, write another plan for expanding to Option C.

Understood Intent

Keep the QDB memory audit at selected Option A and separately plan its Option C: an application-wide capacity study. This is a request for a plan, not approval to run production load, change settings or implement optimizations.

Problem: Current source findings and isolated PHP failures do not establish sustainable per-operation or concurrent-host capacity.

Concepts touched: Public reading, derived-model rebuild, static/offline publication and resource measurement; inventory entries proposed if execution proceeds. Existing content snapshot export, instance import and Fastmod scoring used unchanged.

Option A — Single integrated measurement campaign

Estimate: 0.75

Exercise a documented matrix across major request/command families and publish one capacity report using isolated fixtures and existing tools.

• Shortest route to broad evidence; workload setup and unexplained interactions can make failures harder to attribute. Main uncertainty is representative data and deployment behavior.

Option B — Staged measurement by workload family

Estimate: 0.90

Use a shared reproducible measurement method, establish isolated per-family costs, then test controlled overlap and lifecycle behavior before consolidating the capacity report.

• Recommended: individual family reports are useful vertical slices and clarify where shared-host headroom goes.

• More harness/fixture and recovery validation; actual PHP 8.3/FastCGI and host constraints may remain unverified without access.

Proposed scope and recommendation

Choose B. Cover public QDB/shared reading and search; authenticated reads/writes; rebuild/static/offline jobs; activity/Git evidence; imports/exports/images; scoring and queue maintenance. Give each major entry-point family a tested, inapplicable or deferred status with reasons. Reuse the existing findings and probes, preserving their synthetic provenance.

Evaluate proposed public 16 MiB typical/32 MiB ceiling and maintenance 48/64 MiB budgets. Compare fixed-result workloads at 1000/10000/40000 records; vary body size, reactions, history, pinned content and concurrency separately. Include cold/warm repeated requests, long-lived worker behavior, one maximum-size valid record, large thread/commit and interrupted-job recovery. Track PHP allocations, process/child RSS, time, I/O and records loaded/used separately; PHP limits do not cap aggregate host memory.

Deliver reproducible fixtures/commands, per-family and concurrent measurements, semantic parity checks, a documented safe operating envelope with uncertainty, and ranked remediation slices. Local runs must never target production; live measurements require a separately agreed workload and environment. No new dependencies, permanent monitoring, content limits, server-setting changes or fixes are included. Unavailable host evidence stays an explicit gap.

This expands the memory audit, not the separate load-before-selection audit, whose discovery scope and option selection are unchanged. Step 2 requires Approved Step 1 for this plan.

Study checkpoint

Intake and Step 1: 2026-10-30 UTC; v1/B; baseline cfb8364f6db03cc09e27cd26afb93c109c723a08; R1/P2. Parent's Option C estimate 0.90 retained as inherited; new whole-cycle projection 0.90 reaffirmed, alternative A 0.75, non-additive. Codex single exposed rater with prior audit/probe exposure; frozen R1 verification carried from this session. Scope includes planning, local harness/fixtures, measurements, recovery checks and reporting, not remediation or guaranteed host access. No measurement campaign, independent rating, measured labor or accepted outcome claimed. Reused findings are not new discoveries.

— Codex · 2026-10-30 UTC

Approval — 2026-10-31 UTC

User: “approved step 1”, following the explanation of recommended Option B. Step 1 v1/B approved; proceed to Step 2 requirements. Measurement execution, production load and remediation remain outside this approval.

— Codex · 2026-10-31 UTC