The web profiler's JS engine doesn't implement small-group/imbalance/missing flags, and the parity test works around it instead of testing it

[yakew7]

faircode/profiler.py computes a flags field per column - small_group (driven by --min-group-size, #124), plus imbalance/missing warnings - as part of its output dict. assets/profiler-engine.js (the browser-side reimplementation the web profiler runs) has no equivalent: grepped the whole file for every plausible naming of the concept (small_group, smallGroup, MIN_GROUP_SIZE, minGroupSize, MIN_SHARE, minShare) - zero matches.

This is a known, not accidental, divergence: tests/test_js_parity.py's parity tests (test_python_js_profile_parity and the JSON/XLSX variants) each do

python_result.pop("flags", None)
javascript_result.pop("flags", None)

right before asserting the two engines agree - i.e. the parity test works around the gap instead of testing it, so a real regression in either engine's flags logic (or the JS engine simply never having had it) would never be caught by "parity."

Two ways to close this, either is a reasonable PR:

1. Port the flags logic to assets/profiler-engine.js and delete the two pop("flags", ...) lines so parity is actually verified end-to-end.
2. If flags is meant to stay CLI-only (maybe deliberately, since the web profiler's UI surfaces things differently), document that explicitly in README/CONTRIBUTING and leave a comment at each pop("flags", ...) explaining why it's excluded, instead of the exclusion being silent.

[ahmdkaml]

what are u talking about the parity test is json output and it doesn't skip this and the regular output has the flags?

[ahmdkaml]

it works on both though

[yakew7]

Checked the code directly instead of just re-reading my own claim - you're right, I was wrong. assets/profiler-engine.js does implement MIN_GROUP_SIZE/IMBALANCE_FLAG/MISSING_FLAG/small_group/flags in full; profiler-ui.js renders r.flags/g.small_group straight from the engine's own output, which only works because it's actually there.

My original search used grep against assets/profiler-engine.js, which silently returned zero matches - not because the content was missing, but because the file has two literal null bytes embedded in it (it's a delimiter joining two category values into one composite key: la[i] + '\x00' + lb[i]), which makes some tools (including grep, and file, which reports it as "data" rather than text) choke silently instead of erroring. Reading the file directly instead of grepping it showed the real content.

And tests/test_js_parity.py's pop("flags", None) isn't hiding a gap either - its own comment says why: "flags are human-readable and float-formatted differently between Python and JS. Compare the structured data instead." That's a deliberate, correct choice, not a workaround.

Closing this - thanks for pushing back with the actual screenshots instead of letting a wrong issue sit.