HALOWERK creatorwerk

Endpoints (10 live)

• POST /v1/ad-placement-opt — Sorts caller-supplied placement opportunities by end time and solves the weighted non-overlapping interval problem exactly with dynamic programming. It does not buy inventory, predict conversion, enforce platform policy, model audience harm or guarantee campaign performance. (0.004 USDC on Base)
• POST /v1/thumbnail-ctr — Updates one shared beta prior with each variant’s supplied clicks and non-clicks, reports posterior means and bounded normal-approximation 95% intervals, and ranks by posterior mean. It does not run an experiment, correct assignment bias, prove significance or predict future performance under distribution shift. (0.003 USDC on Base)
• POST /v1/vtuber-rigging — Validates frame/channel dimensions, clamps weights to 0–1 and applies exponential smoothing over time in each named channel. It does not infer a face, build a rig, retarget skeletons, process video or replace animator review. (0.004 USDC on Base)
• POST /v1/watermark-embed — Normalizes whitespace, inserts a caller-supplied visible marker after every selected number of words and at the end of a final partial interval, then reports source and output digests. This deliberately visible text annotation is not a robust image, audio or cryptographic watermark and can be removed by editing. (0.003 USDC on Base)
• POST /v1/plagiarism-music — Converts caller-supplied MIDI-note sequences to adjacent pitch intervals, forms fixed-length interval n-gram sets and reports Jaccard overlap. It does not analyze rhythm, harmony, timbre, recordings, access to prior work or legal substantial similarity, and cannot determine plagiarism. (0.004 USDC on Base)
• POST /v1/royalty-split — Requires shares totaling exactly 10,000 basis points, floors each proportional minor-unit allocation, and distributes leftover units by descending fractional remainder with a stable name tie-break. It performs arithmetic only and does not transfer funds, interpret contracts, withhold tax or resolve rights ownership. (0.002 USDC on Base)
• POST /v1/voice-clone-check — Calculates cosine similarity between caller-supplied numeric embeddings and combines it with supplied spectral distance, pitch-contour correlation and timing distance using fixed weights. It processes no audio and cannot establish speaker identity, consent, cloning, authenticity or fraud. (0.003 USDC on Base)
• POST /v1/trend-hijack — Tokenizes caller-supplied campaign and trend keywords, calculates overlap relative to the trend vocabulary, and flags any supplied blocked keyword. It neither discovers live trends nor recommends deception, impersonation or spam; its scoped purpose is a transparent relevance and brand-safety screen. (0.003 USDC on Base)
• POST /v1/deepfake-score — Normalizes caller-supplied blink deviation, lip-sync error, spectral discontinuity, compression inconsistency and metadata anomaly using declared fixed weights. It does not inspect media, establish authenticity, identify a generator or replace human forensic review; the result is only as reliable as the supplied signals. (0.003 USDC on Base)
• POST /v1/fan-engagement — Weights likes, comments, shares, saves and clicks, divides by supplied reach, and reports cohort and aggregate points per thousand reached accounts. It does not access a platform, distinguish authentic from manipulated activity, measure sentiment or predict revenue. (0.003 USDC on Base)