Sibyla Development Plan v5.0 — FDR + Claude CLI + Layer 5

GOTT Sibyla — project evolution

State of record is docs/PROJECT-STATE.md; this page yields to it. Every number below has one source: the JSON data island in this file, regenerated by scripts/update-evolution-v5.ps1 with the counting rules stated in the Methodology section.

Current edge — what is true right now

Open items — if it is not here, it is decided

Next steps, in order

Commits per day — GOTT.Sibyla, `main`, author dates UTC

Data table

GOTT.IdentityServer (supporting service) is not charted: 4 commits total — 3 on Jul 1, 1 on Jul 16.

Development agent hours per day

Estimated — active message/tool intervals (Jul 21–25) Measured — Hermes session store, overlaps merged (Aug 4–7)

The two bases are different units and are never summed. Measured hours are elapsed execution including autonomous overnight Codex/Claude processing, not human intervention time. The measured series supersedes the earlier commit-bounded floor estimates for Aug 4–6 (2.6, 5.5, 0.9); they are retired, not added.

Data table

Backlog by measurement — v5.0 Phase 1

Phase state

History

The condensed chronicle. Full narratives live in git history and the dated records (p1-1b-status.md, review reports, signed prompts) — this table points, it does not re-narrate.

Jul 1 – Aug 7 timeline

Other measured series — FDR prototype runs · infrastructure hours

FDR pipeline runs are counted as distinct backup timestamps in the prototype control log; Aug 4 onward not yet counted (a measurement gap, not an idle stretch). Infrastructure/tooling hours (session-tracked, separate manual track): 40h05 total Jul 1–23, peak 9h23 on Jul 21; no sessions logged after Jul 23.

Methodology — the counting rules behind every number

Series	Rule
Commits	Total = commits reachable from `main` (side branches are fully merged; stash refs excluded). Per-day = author dates in UTC. Recomputed by the script from `git log`.
Measured hours	Hermes `state.db`: Sibyla/GOTT.Sibyla-related CLI sessions; parallel intervals united (no period counted twice); overlapping subagents not re-counted; non-project work excluded; idle gaps >30 min after a completed response removed; timestamps +0000. Elapsed execution, including autonomous processing.
Estimated hours	Jul 21–25 legacy basis: active intervals between messages and tool calls, pauses >45 min removed, overlaps merged. Kept for the historical window only; never mixed with the measured series.
Live backlog items	Actionable checklist items inside the P1-0…P1-13 sections of `docs/project-todo.md`, per the Aug-3 recount (113 across 14 phases). Excludes: Phase 2 parked, carried-over foundations, decision registers, FDR-side standing items, superseded work. Done = 24 (P1-0's 16 + P1-1a train + validation items). The raw marker counts the script emits are diagnostics for drift, not the denominator.
Percent worked	Done ÷ live items. Gates (piece reviews, D-M*) are not backlog items and do not move the numerator; imports and phase acceptances do.

Regenerate with pwsh scripts/update-evolution-v5.ps1 from the repository root — it recounts commits and backlog markers, restamps the header, and rewrites only the data island between the EVO-DATA markers. Hours are passed in explicitly (the Hermes query needs the server), so they change only when you supply new figures.