Check a slide manifest and plan tiling before a pathology run

Paste your whole-slide image manifest and your tiling and model numbers. Your browser runs the pathml skill's own tools on them - manifest validator, PathML 3.0.5 tile planner and inference batch planner - with the same output as the Python, free. Nothing is uploaded. A paid run reads only the counts and plans and reviews the plan or drafts the methods and provenance.

Both examples have saved model runs for both lanes - the whole page, free.

1. Slide manifest (slide_manifest.py validate)

Manifest CSV (columns slide_id, patient_id, path; optional split, width, height)manifest.csv

Drop a .csv here, or

--root (absolute project folder) --splits --max-rows Home for ~ (page only)

2. Tiling (plan_pipeline.py, one typical slide)

--width (level 0, px) --height (level 0, px) --tile-size --stride --level-downsample --mpp-x --mpp-y --pad

--pipeline (PathML 3.0.5 transform names, in order; tap to add)

More tiling options

--channels --bytes-per-channel --max-tiles --max-output-gib

3. Inference (plan_inference.py)

--tile-count --batch-size --max-memory-mib --dtype

Model card JSON (optional; schema_version "1.0", model_id, input_shape, dtype, ...)

More inference options (override the card)

--channels --height --width --output-elements-per-tile --activation-multiplier --parameter-bytes --runtime-workspace-mib

Paste a manifest or load an example.

Checks (free, in your browser, no AI)

slide_manifest.py

plan_pipeline.py

plan_inference.py

cohort

The browser cannot open your slide files. Paths are checked as text only (URL, escape from --root, duplicates, suffix) and assumed to exist - run the commands below on your machine to check the files themselves.

Flags (the page's own, numbered for the review)

Manifest

Row problems (the page checks every row; the tool stops at the first)

Row	Problem	Detail

Tiling plan (one slide)

Inference plan

slide_manifest.py output

plan_pipeline.py output

plan_inference.py output

The same checks with the skill's scripts

Free: no account, no run.

About the run (optional, sent as written: the cohort, the goal, the hardware - no patient details)

Anything to ask (optional)

Run the checks first to price the review.

Raw model reply

The reply did not parse as the structured result, so it is shown as it arrived.

Your recent runs

What this does, and what it does not

The pathml agent skill ships standard-library Python tools that check a computational-pathology plan without opening a slide or importing PathML. slide_manifest.py validate checks a manifest: pseudonymous slide and patient ids, unique slides and paths, no URLs, allowed split names, and that no patient has slides in two splits. plan_pipeline.py reproduces PathML 3.0.5's tile-count arithmetic and estimates the uncompressed payload for a tile size, stride, level and pipeline. plan_inference.py estimates batch memory from numbers or a JSON model card, without loading a checkpoint.

This page runs JavaScript ports of all three with the same JSON output and error messages, checked against the Python on thousands of fuzzed inputs (see the notice for the counts). One difference is stated plainly: a browser cannot see your files, so each slide path is checked as text and assumed to exist; the commands on the page run the real check on your machine. The page's flags, its pass over every row and the per-slide plans are its own, labelled as such. The paid lanes read only counts, plans and flags - never a slide, a file name or an identifier. PathML is research software, not a medical device: nothing here supports a diagnosis. Derived from the agent skill @k-dense-ai/pathml (k-dense-ai/scientific-agent-skills, MIT; see the notice).