Monthly business review — Tunisia

You are in the country workspace. Produce a memo HQ can read in 5 minutes.

Steps

0. What is still waiting on a human. Read decisions.md first — decisions list --country tn prints it. For every open entry, append today's date to its Relancé line. The count of open decisions and the age of the oldest one belong in the executive summary, before anything else you found.

   This step exists because it used to be missing. July and August both asked for the same authorisation in prose, in a reports/ file that is gitignored and never leaves the machine that ran the task. Nobody was addressed, so nobody answered, and nothing was written for two months. A recommendation that is not in decisions.md has not been made.

1. What happened. Read memory/journal.md and the last month of reports/: what the system did, what changed on the site, open proposals and their status.

2. Traffic signal. Pull Search Console data with the gsc tool (on PATH, no setup needed):

   gsc summary --site tunisie-electricite.com --compare
   gsc pages   --site tunisie-electricite.com --compare --limit 30
   gsc queries --site tunisie-electricite.com --compare --limit 30
   

   --compare puts the last 28 days against the 28 before — use those windows rather than calendar months, so a 31-day month is not read as growth. Report the direction and the size of the move, and name the pages/queries driving it. gsc page <url> --site … breaks a single page down by query when one mover deserves explanation.

   If the tool errors, quote the error and list it as a blocker — never fabricate numbers, and never fall back to estimating traffic from anything else.

3. Landscape scan. Web-search (in French) for: STEG news (tariff changes, outage events, new services like PROSOL), competing Tunisian energy-info sites, and search trends around electricity in Tunisia. Note anything that should become content or a data update.

4. Memo. Write reports/<YYYY-MM>-business-review.md:

   • 3-bullet executive summary
   • Site & traffic status
   • What Country OS did this month (with API-equivalent cost from context if known)
   • Opportunities ranked (each: what, why now, estimated effort)
   • Risks / blockers needing a human decision — and write each one into decisions.md as a new décision entry, or re-raise the existing one. The memo explains; the register is what actually gets answered.

5. Your final message is the executive summary — it goes to the notification channel.

Recommendations here feed the humans and the other tasks; this task itself changes nothing on the site.

The register

decisions.md is git-tracked, unlike reports/. Anything you need a human for goes there or it does not exist. Entries you open, you never close — only a human writes résolu or rejeté, with the reason. Past 30 days, tn:decisions-escalate lifts an entry into a standing GitHub issue on its own.

Prefer one well-argued decision over five. An entry that names the choice, the cost of not choosing, and a recommendation gets answered; a list of wishes does not.