Pouya LajevardiandClaude Opus 5
2ea4c0f8ac
Four rulings from Pouya, 2026-08-30, and their sweep. D20 — the review protocol. Per build step the review is `adversarial-reviewer` alone. `claims-auditor` no longer runs per step; it runs ONCE, at cutover, over the whole finished site, as a blocking item near the top of docs/06's checklist. `check:claims` is unchanged and still runs on every build and both deploy paths. The reasoning is recorded in full in AGENTS.md D20, as a calibration and not an erosion: nothing has shipped, so every claims finding so far has been about a page no visitor can reach, and one pass over twenty finished pages catches more than nine passes over drafts because it sees the site as a reader does. The /med-arb/ ADRIC gloss is the proof — no individual claim was false, the defect was adjacency, and adjacency does not exist until the pages sit next to each other. The code reviewer stays per step because what it catches compounds. What this costs is recorded honestly beside it, not summarised away. D17 and D19 amended to match. D19's two-round cap governs the per-step code review only; the single cutover claims pass runs until its findings are resolved, because there is no second pass behind it. Q56 — mediation is NOT scoped commercial. Thirteen shipped strings corrected across five files: page titles, meta descriptions, hero ledes, section ledes, the `Service` node's name and description, and `ProfessionalService`'s. §4's mediation row stays unscoped, and the reason now sits beside both rows so the asymmetry reads as designed: arbitration is scoped commercial because of a LEGAL GATE (Q39 — family arbitration in Ontario requires prescribed training); mediation has no such gate. `adversarial-reviewer` then found three surfaces the sweep had missed, the worst on /practice/ — "These describe the process the parties are choosing between, in commercial matters" scoped mediation with the two words never appearing in the same element, so no proximity grep reached it. Q55 — CLOSED WITHOUT BEING RESOLVED, and the difference is the ruling. The Q.Arb stamp is split: `[verified]` on the status, `[Pouya's stated basis]` on the date. The 2026-08-26 record is marked UNRECONCILED, permanently and on purpose. The date is not published and nothing depends on it. check:claims — FROZEN. Round 2 found five defects in round 1's own fixes to that script, two of which made it worse than before the pattern existed. A pattern is added only after a real breach reaches dist/, never speculatively, and each addition ships with a probe plus a negative fixture. No refactors, no coverage improvements. It is a tripwire, not a program. Two conventions into CLAUDE.md: sweep the VOCABULARY, not only the subject (`git grep 'Q.Arb'` is line-anchored and could not find ten lines entirely about Q.Arb that never name it); and agent definitions load at session start, so an edit to .claude/agents/*.md does not reach the session that made it. Verified: check 0 errors, lint 0, build 0 (12 pages), check:claims 0. Lighthouse not run — tool unavailable until build step 7. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Md3GndFqWPzK78xAoebsg5
adr.smlcompany.ca
The dispute resolution practice of Pouya Lajevardi — Toronto.
A static site built with Astro, built for deployment to Amazon S3 behind CloudFront by Gitea Actions — see Deployment; the pipeline is not yet proven.
Quick start
nvm use # Node 22 LTS — the floor is in package.json engines
npm install
npm run dev # http://localhost:4321
Scripts
| Command | Does |
|---|---|
npm run dev |
Development server with hot reload |
npm run build |
Static build to ./dist |
npm run preview |
Serve the built site locally |
npm run check |
astro check — type and template errors |
npm run lint |
ESLint + Prettier check |
npm run format |
Prettier — rewrite files in place |
npm run deploy |
Build and deploy from this machine — see Deployment |
Before you contribute
Read AGENTS.md first, and maintain it as you work — it is the living
record of what this project is, what was decided, and why. Then read
CLAUDE.md for the working rules, and the specs in docs/.
The single hardest rule: no factual claim about Pouya, his credentials, his
experience, or his practice ships unless it appears in the verified register in
AGENTS.md §4. This is a public marketing surface, and the site it replaces
contained fabricated credentials.
How work is done here
Pouya decides; Claude Code implements and then adversarially reviews its own
work. Run /build <task> for any substantive change — it plans, implements,
runs two independent review agents on the diff (the claims audit wherever copy
changed), resolves the findings, verifies the build, and records the session in
AGENTS.md. /review runs the review pass alone; /wrap closes a session.
Full protocol and prompt guidance: docs/08-execution-protocol.md.
Deployment
Today, deploys run locally: npm run deploy (scripts/deploy-local.sh).
It runs the same guard, the same three sync passes with the same cache headers,
and the same invalidation as the CI workflow — at this scale the pipeline
changes only how a deploy is triggered, not what it does. The script and
.gitea/workflows/deploy.yml are one artefact in two places: change one, change
both.
.gitea/workflows/deploy.yml is the CI pipeline — Gitea Actions, not GitHub
Actions. It has never run, for two reasons that are not oversights:
- Actions are not enabled and no runner is registered. The Gitea instance is jointly administered, so both need its second administrator (Q23).
(It previously listed a second reason — that the scoped IAM user did not exist.
It exists: adr-sml-deploy, created 2026-08-26, Q22 closed 2026-08-28. See
AGENTS.md §7 for the inventory and the least-privilege verification.)
Its first step is a guard: the run aborts, naming what is missing, if any required variable or either AWS secret is empty. Only emptiness is tested and no value is echoed.
The GitHub Actions equivalent, which uses OIDC role assumption, is kept as
docs/reference/github-actions-oidc.yml.example in case the project ever moves
to a forge that supports it. It sits outside .github/workflows/ on purpose:
Gitea falls back to that directory when .gitea/workflows is absent, so a
workflow file left there with a push trigger would be only conditionally
inert. As an .example under docs/ it cannot be picked up at all.
The pipeline is designed around a long-lived AWS credential, and that
credential now exists. Gitea is not an AWS OIDC provider, so there is no role
to assume: deploys authenticate as a scoped IAM user, adr-sml-deploy, with its
access key in the repository's Gitea Actions secrets. The user was created
2026-08-26 and verified least-privilege by execution — AGENTS.md §7, Q22 closed
2026-08-28. Its access key has never been used (LastUsed null), because
deploys still run locally. (This paragraph asserted the user "has not been
created" until 2026-08-28.) In the meantime the local script
refuses to run as user/pouya, the broadly-permissioned personal user —
see AGENTS.md §10. Two things are meant to bound the risk, and neither is in
place yet:
- The policy must stay narrow. Four actions:
s3:ListBucketon one bucket,s3:PutObjectands3:DeleteObjecton that bucket's contents, andcloudfront:CreateInvalidationon one distribution. NoAction: "*", noResource: "*", nothing outside that one bucket and that one distribution. The AWS account is shared with unrelated projects, including a bucket whose name indicates another business's production database backups — that narrowness is what keeps a compromised runner away from it, and it is load-bearing rather than hygiene. SeeAGENTS.md§10. If a deploy step needs a permission the policy lacks, question the step; do not widen the policy. - The key must be rotated quarterly. It now has an owner and a date: first
rotation due 2026-11-26 (key created 2026-08-26) —
AGENTS.md§12 R17. Create a second access key, update the Gitea secrets, confirm a deploy succeeds, then delete the old one — rotation that leaves the old key active is not rotation, and deleting before verifying leaves no way to authenticate the fix. OIDC would have removed the obligation entirely; it is unavailable, so this is a standing calendar task. (It read "nobody owns that yet" until 2026-08-28; R17 is the owner, and the date is the whole point of the row.)
Full procedure, IAM policy, runner setup, and cutover checklist:
docs/06-deployment.md.