name	tfcs-status
description	Run the TFC replication-progress analyzer on scott and produce a cluster convergence report — per-org single-copy replication rate, ETA, accelerating-vs-stalling, and pull flows. Trigger on "tfcs status", "network progress", "replication report", "how's the cluster converging", "is replication catching up", "run the progress report".

tfcs-status

Produce a convergence report from the tfcs-monitor snapshot series on scott. Convergence only — no infra/timer/health reporting.

Run

Always fetch fresh and run on scott (the series lives there):

ssh scott 'cd ~/projects/TFC && git pull --ff-only -q && ./scripts/progress-windows.py ARGS'

Map the request to ARGS:

a single org → --org ii (hrdag, ida-anchored) or --org km0 (datacivica, ant-anchored)
explicit windows → --windows 6,24 (hours); omit for the default 6,12,24,48
raw numbers → add --json

Report shape

From the output, write a tight report in this order:

Verdict — one line: converging vs stalling, and ingest-driven vs pure catch-up. total_commits Δ=0 across windows ⇒ no new ingest, every change is promotion.
Per org — holders=1 (SOLE) replication rate (single-copy commits gaining a second holder); accel vs stall = compare the recent 6/12h rate to the 48h average (recent faster ⇒ accelerating); ETA = sole_now ÷ recent_rate; conservation check = the bucket Δs sum to total_commits Δ (sanity).
Anchors — seeding (copies_count flat while sole_holder_count falls — its sole-held commits replicating outward to other nodes) vs gaining; confirm sole_holder_count Δ matches the org's holders=1 Δ.
Pull flows — top pullers per source and total volume moved from each anchor.
Watch-item / lever — the slowest-replicating backlog and what would speed it (usually puller bandwidth on the anchor).
Caveats — trust reset-free recent windows most; ! = a cumulative counter reset across an agent restart; [restart in window] = anchor restarted in the lookback; clamped = window older than the data span.

Headline metric is per-org holders=1 replication rate — how fast single-copy commits are being replicated to additional nodes (distributed across the cluster). Report it as replication, never as "burn-down": the count falls because we are creating copies, not destroying them. For deeper interpretation and SSH drill-down into node logs, see docs/runbooks/network-progress.md.

Boundary

Read-only. This skill runs git pull + the analyzer over SSH and reads state. It NEVER runs par2 repair, deletes, re-pulls, or mutates cluster state.

tfcs-status

Produce a convergence report from the tfcs-monitor snapshot series on scott. Convergence only — no infra/timer/health reporting.

Run

Always fetch fresh and run on scott (the series lives there):

ssh scott 'cd ~/projects/TFC && git pull --ff-only -q && ./scripts/progress-windows.py ARGS'

Map the request to ARGS:

a single org → --org ii (hrdag, ida-anchored) or --org km0 (datacivica, ant-anchored)

explicit windows → --windows 6,24 (hours); omit for the default 6,12,24,48

raw numbers → add --json

Report shape

From the output, write a tight report in this order:

Verdict — one line: converging vs stalling, and ingest-driven vs pure catch-up. total_commits Δ=0 across windows ⇒ no new ingest, every change is promotion.

Per org — holders=1 (SOLE) replication rate (single-copy commits gaining a second holder); accel vs stall = compare the recent 6/12h rate to the 48h average (recent faster ⇒ accelerating); ETA = sole_now ÷ recent_rate; conservation check = the bucket Δs sum to total_commits Δ (sanity).

Anchors — seeding (copies_count flat while sole_holder_count falls — its sole-held commits replicating outward to other nodes) vs gaining; confirm sole_holder_count Δ matches the org's holders=1 Δ.

Pull flows — top pullers per source and total volume moved from each anchor.

Watch-item / lever — the slowest-replicating backlog and what would speed it (usually puller bandwidth on the anchor).

Caveats — trust reset-free recent windows most; ! = a cumulative counter reset across an agent restart; [restart in window] = anchor restarted in the lookback; clamped = window older than the data span.

Boundary

Read-only. This skill runs git pull + the analyzer over SSH and reads state. It NEVER runs par2 repair, deletes, re-pulls, or mutates cluster state.