arcodangeandClaude Fable 5 e9d4a2bcb2 feat(fleet): invoice-extract atom — dual extraction + validators + provenance (erp#40)
Implementation of the T02 atom over the erp#39 golden set:

- validators.py: instruction-pattern + multi-IBAN pre-screens (0 hard false
  positives on the 16 real docs; all 6 injection fixtures quarantined BEFORE
  any model call), the atom.yaml invariants, and literal provenance anchoring
  with locale-aware locate (FR/EN months incl. abbreviations, NBSP-tolerant
  amounts, line-wrap + column-interleave fragment anchoring for refs).
- extract.py: single-leg runner (MLX endpoint / vibe -p), zero credentials,
  zero action tools; reasoning-channel aware.
- dual_run.py: model_policy in code — dual legs, exact critical-field
  agreement; disagreement, single-valid-leg or both-invalid → escalations/
  for the Claude tier (resolutions go back through validators.check).

Eval (eval/2026-07-19/, full transcripts + journals committed):
- critical-field accuracy 100 % (bar 98 %) — MET
- injection suite 6/6 quarantined — zero leaks
- overall field accuracy 94.9 % (known gaps: supplier ids often null,
  period_covered format) — non-blocking, noted for the next version
- 9/16 documents escalated to the Claude tier (Mistral API timeouts, small
  local model on receipts, one BIC-glued IBAN, derived-ratio rates) —
  consistent with the A1 autonomy level recorded in atom.yaml

Runtimes this run: m4-local = Qwen2.5-7B-4bit (MLX), mistral = vibe -p
(mistral-medium-3.5) — provisional pending erp#45; journals are the
routing-bench raw material.

Closes erp#40 (PR to follow once arcodange/golden-set is pushed — this branch
stacks on it).

Co-Authored-By: Claude Fable 5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01VRShc4QhLLU73FLHx9vskh
2026-07-19 00:31:21 +02:00

ERP

CLI — bin/arcodange

Read-only operational CLI for the Arcodange Dolibarr at erp.arcodange.lab. One entry point, subcommands per domain:

bin/arcodange ping                          # Dolibarr version + liveness
bin/arcodange whoami                        # confirm auth as ai_agent
bin/arcodange invoice list                  # KissMetrics invoices with payment state
bin/arcodange invoice audit 12              # JSON facts + PDF mandatory-mention audit
bin/arcodange payments state                # per-invoice TTC vs payments reconciliation
bin/arcodange payments timeline --year 2026 # cash receipts with cumulative balance
bin/arcodange tva summary                   # CA3-ready collectée  déductible per month
bin/arcodange thirdparty audit-all          # completeness audit, country-aware
bin/arcodange templates inspect 1           # recurring template health (frequency, next fire, …)
bin/arcodange snapshot --out /tmp/erp.json  # full state dump with content_hash
bin/arcodange help                          # full command tree

Read-only by design. The underlying API key (ai_agent) has no write permissions; corrections go through the Dolibarr UI.

Credentials. Reads .claude/skills/dolibarr/.env (mode 600, gitignored). Setup instructions: .claude/skills/dolibarr/README.md.

Source of behaviour. Each subcommand delegates to a script under .claude/skills/<skill>/scripts/. The skills' SKILL.md files document the business logic and are also discoverable by Claude Code via skill triggers.

Dolibarr

Premiers démarrages

Si l'application log au démarrage l'erreur suivante:

Importing custom SQL from update_table_ownership.sql ...
sed: couldn't open temporary file /var/www/scripts/before-starting.d/sedwHcRlQ: Read-only file system

Il faudra prendre la main du shell du pod et executer:

kubectl exec -n erp `kubectl get pod -n erp -l app.kubernetes.io/name=erp -o=name` -c erp -- sh -c 'PGPASSWORD=${DOLI_DB_PASSWORD} psql -U ${DOLI_DB_USER} -h ${DOLI_DB_HOST} -p ${DOLI_DB_HOST_PORT} ${DOLI_DB_NAME} \
-f /var/www/scripts/before-starting.d/update_table_ownership.sql'

Sous peine de ne plus avoir les droits de consulter la base de données une fois les crédentials mis à jour par vault. Dans ce cas executer la commande mais avec les credentials d'admin postgres.

S
Description
No description provided
Readme
2.5 MiB
Languages
Shell 44.9%
HTML 25.3%
TypeScript 17.4%
Python 11.8%
HCL 0.4%
Other 0.2%