Busflow Docs

Internal documentation portal

Busflow Docs

Appearance

Reviewer Critical Review After Execution — Funding Drilldown L2 ​

Verdict ​

Rework before using these results as the basis for main-document edits or submission text. The canonical outputs now exist and the run contract is mostly followed, but several outputs are still high-level/preliminary and do not satisfy important L1/L2 acceptance criteria, especially source/line traceability, L2-E AP24–AP26 Antragstext drafts, L2-H live-source evidence, and C's provenance for AP27.

Correct ​

  • Canonical output contract is now satisfied structurally. README.md defines the expected result files and gates (README.md:7-25), and all eight L2-X-result.md files exist.
  • Run order/gates are reflected in outputs. A states it read final-for-this-working-state G input (L2-A-result.md:3-7), B states it read A/G (L2-B-result.md:3-7), C states it consumed D/E/F (L2-C-result.md:3-7), and H states it used A–G only as a passungs/abgrenzungs frame (L2-H-result.md:3-7).
  • No unauthorized main-document edits found. git diff --name-only -- docs/1-projects/bsfz-funding shows only L2 plan files; result files are untracked under funding-drilldown-L2-results/. No diff was present for STRATEGY_public-funding.md, funding-work-packages.md, bsfz-antrag-vorhaben.md, bsfz-ablehnungsrisiken.md, or funding-application-plan.md.
  • D/E/F handoff table headers are identical and AP coverage is mostly correct. D covers AP1–AP6, AP7a, AP8, AP9 (L2-D-result.md:68-80), E covers AP20–AP26 (L2-E-result.md:45-55), and F covers AP10–AP19 (L2-F-result.md:37-50) with the same columns.
  • Conservative markings are visible. Examples: G calls itself a dokumentenbasierter Schnell-Drilldown and flags missing code/commit forensics (L2-G-result.md:3-7); H explicitly marks claims as hypotheses pending live web research (L2-H-result.md:3-7); D/E/F contain multiple bedingt, blockiert, and validation markers.
  • Steuerberater/FZulG validation is explicitly marked in several places. A flags pauschalen/solo-founder validation (L2-A-result.md:3-7), B flags infrastructure/time recording validation (L2-B-result.md:84-88), D flags AP1/AP3 FZulG review (L2-D-result.md:82-86), and G flags infrastructure/prospective candidate validation (L2-G-result.md:60-64).

Blocker ​

1. L2-B does not meet the L1 traceability acceptance criterion ​

Evidence: L1 requires the consistency report to list every discrepancy with source document, line number, and concrete correction (funding-drilldown-L1.md:140-144). B explicitly says no full line-level proof was performed (L2-B-result.md:3-7). Its discrepancy tables then list findings like the 4.270h vs. 4.110h drift and Frascati wording issues without line references (L2-B-result.md:21-28, L2-B-result.md:39-47).

Fix: Add a Zeilennachweise column or per-finding bullet with exact file path and line range for every B discrepancy/TODO/language finding. Do not apply B patch suggestions until those references are verified.

2. L2-E is missing required AP24–AP26 initial Antragstext drafts ​

Evidence: L1 expects initial Antragstext drafts for AP24–AP26 in the shape of bsfz-antrag-vorhaben.md fields: Ziel, Stand der Technik, Durchgeführte Arbeiten, Unsicherheiten (funding-drilldown-L1.md:324-337). E only provides a reife matrix, assignment table, AP26↔AP7a distinction, brief text/patch suggestions, and C handoff (L2-E-result.md:8-67). There are no AP24/AP25/AP26 draft sections in the required form.

Fix: Add dedicated sections for AP24, AP25, and AP26 with Ziel, Stand der Technik, Durchgeführte Arbeiten, Technische Unsicherheiten, and a short Frascati self-check. Keep them as drafts/proposals, not main-doc edits.

3. L2-H is not an executed market/science analysis yet; it is a research agenda ​

Evidence: H states no complete live web research was performed and that market/literature statements are hypotheses, not citeable evidence (L2-H-result.md:3-7). Its external section lists search fields and source hints, not actual source links/citations (L2-H-result.md:9-19). L1 acceptance requires at least five evaluated opportunities with concrete effort/return/risk assessment (funding-drilldown-L1.md:519-521), but H's Top-5 list has no effort/return/risk columns (L2-H-result.md:43-49).

Fix: Treat L2-H-result.md as preliminary. Run a follow-up H research pass with current source links, 2–3 strong sources per field, and a Top-5 table containing Aufwand, Ertrag/strategischer Nutzen, Risiko, Quellen, and Sofortmaßnahme vs. Folgejahr.

Major ​

4. L2-C claims a D/E/F-normalized matrix but includes AP27, which is not in D/E/F ​

Evidence: C labels the matrix Normalisierte Risikomatrix aus D/E/F (L2-C-result.md:9) and includes AP27 in the medium tier (L2-C-result.md:14) and V3 stütz-APs (L2-C-result.md:41). However, D covers only AP1–AP9 (L2-D-result.md:68-80), E AP20–AP26 (L2-E-result.md:45-55), and F AP10–AP19 (L2-F-result.md:37-50). AP27 comes from G (L2-G-result.md:37, L2-G-result.md:51, L2-G-result.md:57), not D/E/F.

Fix: Rename the matrix to aus D/E/F plus L2-G für AP27, add a source/provenance column, or move AP27 into a separate zusätzlicher G-Input subsection. C should not imply AP27 was normalized from D/E/F.

5. L2-G does not satisfy the full-document mining expectation and has a count inconsistency ​

Evidence: G explicitly says it did not perform a full deep review of all ADRs and calls the output a quick/document-based drilldown (L2-G-result.md:3-7). It also says 31 ADRs, while the current docs/3-resources/decisions scan found 30 ADR files. L1 expected all ADRs to be part of G's source base (funding-drilldown-L1.md:426-437).

Fix: Add an ADR coverage table listing every ADR file found, status (scanned deeply, skimmed, not relevant, missed), AP linkage, and candidate implication. Correct the ADR count to the actual repository count or explain the extra item if non-ADR decision files are included.

6. D/E/F Input für L2-C tables lack dependency/partial-rejection relevance despite the plan requiring it ​

Evidence: The standardized tables have the required identical columns (L2-D-result.md:68-80, L2-E-result.md:45-55, L2-F-result.md:37-50), but they do not include explicit dependency/Teilablehnungsrelevanz fields. C therefore reconstructs dependencies in its own map (L2-C-result.md:35-42). This is partly workable, but it weakens handoff traceability, especially for AP6↔AP1, AP15↔AP3, AP25↔AP20, AP26↔AP7a.

Fix: Add one short Abhängigkeiten / Teilablehnungsrelevanz column or a follow-up subsection in D/E/F. Then update C's dependency map with citations to those rows.

7. Many claims are plausible but not evidence-backed enough for direct patching ​

Evidence: A gives strategic recommendations and patch suggestions (L2-A-result.md:98-102) but usually references section names, not source line ranges. B's source matrix is file-level only (L2-B-result.md:9-19). D makes AP-specific claims such as AP2's GoBD/offline conflict and AP5's ACID/security implications (L2-D-result.md:29-43) without citing the specific ADR/spec lines. E and F similarly provide useful but uncited judgments (L2-E-result.md:8-18, L2-F-result.md:8-25).

Fix: Before applying any patch to main docs, add source references for each claim that changes strategy, F&E %, Go/No-Go, or Antragssprache. Minimum: file path + heading/line range + whether evidence is existing, inferred, or still missing.

Minor / Notes ​

  • L2-A mostly satisfies its strategic scope, including 3/4/5 clustering, hybrid submission strategy, phase-0 readiness, and three follow-year scenarios (L2-A-result.md:18-44, L2-A-result.md:65-96). It still needs more exact source references for each patch proposal to fully match the L1 requirement that each recommendation references the specific place it changes.
  • L2-D correctly keeps AP26↔AP7a out of D's own analysis, but it also does not explicitly reference E's AP26↔AP7a result in the D output. Since E provides that section (L2-E-result.md:32-36), add a short D note under AP7a or Validierungsbedarf: AP26-Abgrenzung siehe L2-E-result.md:32-36.
  • L2-C has useful AP24–AP26 weak-candidate analyses (L2-C-result.md:44-65) and a concrete audit/proactive-measures section (L2-C-result.md:67-95). These are suitable as review inputs, not final wording, because they lack citations and legal/tax detail remains open.
  • The 4.270h vs. 4.110h discrepancy is correctly detected (L2-B-result.md:21-28; L1 still states ~4.110h at funding-drilldown-L1.md:8-12). This should be one of the first cleanup patches after line verification.
  1. Bring L2-B to traceable standard: add file/line references for all discrepancies, TODO findings, and language issues.
  2. Complete L2-E missing AP24–AP26 Antragstext drafts.
  3. Re-run/complete L2-H with actual live sources and effort/return/risk scoring.
  4. Fix L2-C AP27 provenance and update C after D/E/F dependency columns are added.
  5. Add ADR coverage table and correct ADR count in L2-G.
  6. Add citations/source ranges across A/D/E/F before any main-document patch is applied.

Final Accept/Rework Verdict ​

Rework required. The execution is acceptable as a first-pass internal analysis package and is much better than the earlier missing-output state, but it is not yet acceptable for main-document edits, grant-submission language, or final strategic decisions without the fixes above.

Internal documentation — Busflow

© 2026 MLP Software Development