One page that says what we found, what we decided, what we are building, and what is still on someone else. Everything below links to the document that proves it.
maptara.claim.batch is real — 317 lines, state machine, 387 lines of tests, missing only file generation. The §302 mapping is a finished 34-field spec needing zero new fields. Neither is a rewrite.depends: ['base'] only. A driver knows nothing — not the domain, not the provider, not where the password came from.Egeko live · azh next · MIP, ZHP X3 later · direct DTA (Release 2)
azh dry-run + Egeko rewrite
3 of them change scope
azh SFTP account
The one-pager. Why, the rule, the before/after diff down to file and method level, the phases, what is out of scope, five acceptance tests.
The wide diagram: provider exchange and contract import as two flows sharing one driver layer. VDMS expanded — including the gap that imported contracts arrive with no price rules.
Every file in both modules with a verdict and destination. Real wc -l counts. 57 files; 3 654 lines untouched; 1 262 rewritten; 7.6 MB of vendor PDFs dropped from the shipped module.
15 classes, 82 fields, 105 methods, each with a verdict. The 15 biggest methods ranked — one is 6.9% of all runtime code. The contract as 7 outbound + 3 inbound operations.
Two devs, day by day. What compresses with AI assistance and what does not. Exit criteria per milestone. Re-cut on the two measurements that replaced two guesses.
The contract signature by signature. ConnectionConfig. SftpDriver. The azh data model, all 34 fields, the file lifecycle, the rejection path. Egeko in four steps. Nine open decisions.
Eleven questions grouped by who can answer them, each with our recommendation. No estimates, no internals. Safe to forward.
The Lovable spec and module read in full and measured. Four big issues. Two routes on one contract. What we take, what we don't, six questions back to Volker.
Four parallel analyses of the generated module. Won't install (one fix). Then: clean wiring, no process. 142 claims: 46 true, and the true ones are field names. What to keep.
Model-by-model verdicts, the split-or-scrub SWOT, migration hazards and order. The eKV split was considered here and dropped; the reasoning is kept so nobody re-proposes it cold.
The full 12-section report with every file:line. Read this when a number on any other page looks wrong.
| Decision | Outcome | Where |
|---|---|---|
| KVCheck / eKV via AZH | Cut. We already do eKV. | 00 |
| eKV module split out of provider_base | Dropped. Zero behaviour change, real production-migration risk, cost never rises. | R1 §3a |
| Where drivers live | depends: ['base'] — not maptara_base (12k lines, drags in accounting), not maptara_san_base (a hidden compat shim). | 00 |
| Design the contract against which protocol | REST, not SOAP. SOAP has one consumer and it is the module being rewritten. Abstracting from the outlier fits everything else badly. | 00 |
| SoapDriver with one consumer | Keep — on testability, not reuse. It is the seam that lets a misspelled wire key turn the suite red. | 05 |
| REST driver | In, but last. Nothing in October needs it. Dev A's slack, first thing dropped. | 04 |
| MIP | Not now. Credentials live since 26 Aug; protocol unconfirmed (md5 key ≠ REST). | 04 |
| Credential ACL fix | Day 1 — but keep read: Codexa's FK at provider_site.py:20 is not sudo'd and breaks without it. | R1 W6 |
| Estimation style | Rolling, AI-augmented. Mechanical work compresses; verification, vendor access and human decisions do not. | 04 |
| Item | Age | Who | What it blocks |
|---|---|---|---|
| azh SFTP account + SSH key | 46 d | us — send Monday | Proof. Everything up to "XSD-valid bytes in a directory" is in our hands. |
EinlieferungsArt — paper (2) or digital PDFs (1/3) | — | product | Largest scope variable. Building paper-only unless told. |
| Prescription printing — does Maptara print the Muster 16 today? | — | product | Whether the barcode is a day or a new report. In nobody's estimate until yesterday. |
| Versorgungsanfrage → which object? | — | product | Inbound ops only. Nothing in October. |
| Any VDMS ticket | — | you | Everything VDMS in these docs came from code, not a ticket. |
Send the SFTP account email. One message, 46 days late, and it is the only item on this page that no amount of development moves.
Nothing has been implemented. Every document above is planning. No source file outside docs/ and .agent-work/ has been touched. The working agreement stands: no code until an explicit GO.
Evidence base: four read-only code surveys in .agent-work/provider-refactor/; tickets #1197, #1184, #1271–1276; the §302 Datenmapping and the azhDirekt/azhIndex specifications in .agent-work/ticket-1197/spec/. Code surveys and ticket material are under evidence/ in this repository.