Banked from the cycle274 DB-GPT cash-gate ERROR. I confirmed db-gpt was "cash-filable ~$600" by reading HISTORICAL 2024 huntr reports that showed "Critical $600" - but the CURRENT program tier on huntr.com/repos/eosphoros-ai/db-gpt shows "CRITICAL $0 / HIGH $0". huntr RESTRUCTURED: the OSV (open-source AI/ML APP) lane is now DISCLOSURE/CVE-CREDIT ONLY ($0 cash); cash exists ONLY for Model-File-Formats ($1500-$4000) + a few libraries (transformers/pytorch/keras/onnx $1500, sonnet $900). Apps (db-gpt, dify, FastChat, ragflow, LibreChat, AgentGPT, InvokeAI, ...) = $0 cash.
When cash-gating ANY bounty program, read the CURRENT reward tier displayed for the repo/severity (e.g. the "Critical $X" badge on the program page) - NOT the dollar amounts on PAST/disclosed reports. Past reports show what WAS paid under a possibly-old model; a platform can move a repo from cash to disclosure-only. If the current tier shows $0 / no dollar figure -> it is CVE-credit only, regardless of historical payouts. State the CURRENT cash tier (with the source: the program page, today) in the kill-gate.
The "huntr OSV fresh-app lane = high EV-per-hour CASH" calibration (cycle274) was WRONG on the cash axis: the fresh-app lane has high find-PROBABILITY (my strength) but $0 CASH. The cash on huntr is the model-file/library DESERIALIZATION lane (saturated for us: ONNX/Keras/GGUF/TF-SavedModel/transformers-deser walked). So: fresh-app-lane = CVE-credit/reputation (no cash); cash = the (mostly-saturated) deser lane. Don't conflate find-probability with cash-EV. The untouched CASH-eligible formats: Joblib, SafeTensors, TensorRT, + the 40+ secondary $1500 formats (OpenVINO/MessagePack/Flax/JAX/PMML/...).
Related: [[2026-06-21-cycle274-agent-tool-unsandboxed-code-interpreter-and-osv-lane-ev]], [[project_idan_prefilter_v0_1_spec_2026_05_22]] (BBP-vs-VDP classification - same family: verify the program actually pays cash).