Flow — How it ships

Every conversation about a product and how far it got, stage by stage, in the product's own pipeline:. Drawn in designs/web/workspace/flow.html (one product) and flow/flow.html (every product). Today, the workspace overview and the landing show parts of it.

Where the numbers come from

• .context/metrics/flow.json in each product repo: counts per stage and the median hours of each step. It is written on the owner's machine by ~/git/bin/flow, from Claude transcripts and git. It holds counts and times only, never what was said. Transcripts older than its since date are gone.
• Canon Cloud cannot read transcripts. It takes the idea-to-design counts from the committed flow.json, and works out pull requests, tags and store state from GitHub itself.
• The numbers in the designs are real: each product's flow.json as of 27 Oct. The "Waiting now" rows are samples.

What it does better than the portfolio's "How it ships"

The portfolio (ritesh-firodiya.github.io, src/components/funnel.tsx) reads the same file. Canon changes:

• A 0 h wait between two commit stages (plan, design, code) reads "same commit". One commit did both steps, so it is not a fast step. A step with no time reads "not timed". Neither one is ever ranked.
• Each step says how many went no further yet, the work waiting now, as well as the share that got through.
• The biggest drop counts from Plan on. Most conversations never become work (babul: 214 conversations, 74 researched). That is not the drop to fix.
• Every product side by side, each in its own pipeline. A stage a product does not have is a dot, never a 0.
• Rankings across products: where work stops (from Plan on) and the longest waits. A wait is ranked only when 3 or more products timed it; a median of two is an anecdote.
• It flags numbers that look wrong. learning-platform and scrvio report the same time for every step, which points to a fault in bin/flow.

What it says today

• Most work stops after it is built. From Plan on, the most held steps are Deployed → Live (80, mostly babul's 70), Code → Review (67) and Store review → Live (46).
• The longest wait timed in 3 or more products is Code → Review: a median of 1.5 days across 7 products.