# Element Projection Lab 0.11.0 — guarded parameter explorer

## Completed capability

The engine now provides an offline four-panel explorer and an off-grid query
CLI for the four observed ATLAS CW/LD contours. It preserves the original
`k=500 GeV, M5=6000 GeV` benchmark, which is still below all four boundaries.
Moving the exploratory selection does not move or replace that benchmark.

The declared computational window is M5=1000–12000 GeV and k=100–5500 GeV,
with 100 and 50 GeV increments: **12,099 nodes per contour**. It is a display
and finite-computation window, not a certified published likelihood scan
rectangle and not evidence of a physical space/time grid.

## Checked result and counterexample

All **48,396 grid classifications** and **2,916 off-grid classifications**
(729 probe records, 712 distinct coordinates, each checked in four tables)
agree with an independent source-decimal exact-rational implementation.
Maximum displayed-boundary discrepancy is **3e-12 GeV**; a 1e-8 GeV tolerance
is used only for float serialization, never to compare classifications or
interpret experimental precision. The independent audit passes **54,586
checks**, including 20 synthetic controls. All **162 unit tests** pass,
including all 120 prior tests and 42 new engine/view tests.

The regular grid misses every one of the **nine nonzero horizontal segments**
at five distinct table/k levels. An additional repeated vertex has zero length
and must not be mistaken for an ambiguous interval. Off-grid checks preserve
both distinctions. A regular grid alone therefore does not establish global
contour topology.

Concrete off-grid query `k=2781.8 GeV, M5=6000 GeV`:

| Observed table | Separate contour-relative result |
|---|---|
| 1: dielectron, thresholds | Unresolved horizontal overlap |
| 7: diphoton, thresholds | Below contour — excluded side |
| 13: dielectron, no thresholds | Above contour — not-excluded side |
| 19: diphoton, no thresholds | Below contour — excluded side |

This is a counterexample to silently choosing one Table 1 intersection or
forcing the four views into one answer. At `k=5500 GeV, M5=11000 GeV`, all four
instead report missing k coverage, not survival.

## What the display does

- Original-order open polylines, four separate views and a fixed benchmark.
- Native numeric, slider and click/touch selection, explicitly snapped to a
  stored grid node. No browser fit or interpolation of decisions.
- Explicit above/below, boundary, missing k coverage, ambiguous horizontal or
  folded geometry, and nonperturbative `k>M5` states. CLI queries outside the
  chosen viewport are rejected as outside map scope, not physically excluded.
- Published rounded vertices and provenance remain embedded. Grid counts are
  software coverage diagnostics, never parameter probabilities.

The original source parser/interpolator, source archive, frozen parameters,
all historical results, and previous negative findings are unchanged. The full
Theory 1.71 is preserved. The initial independent-audit failure, caused only by
a swapped hardcoded channel/threshold expectation, is retained in
`outputs/independent_parameter_map_audit_initial.json`; the final oracle
derives those labels directly from source metadata and passes. No physical
data or parameter was changed to produce that pass.

UI verification limitation: Node-based selection and payload tests pass, but
the real-browser preview was blocked because no browser binary was installed
and the browser download failed (timeouts/HTTP 502). The release therefore does
not claim completed browser or mobile CSS visual validation. The separate
local control/canvas smoke report records exactly what was exercised.

## Limits and next test

These are **finite calculations from externally measured exclusion curves**,
plus synthetic tests—not new measured evidence, an independent likelihood
reconstruction, a joint confidence region, or evidence for extra dimensions.
Above-contour points need not lie within a supplied two-dimensional likelihood
grid; the archive does not contain one. The four analyses reuse data. The
borrowed CW/LD parameters are not derived from this project's fixed-grid or
string-rotation hypothesis, and no unexplained residual identifies a unique
physical or theological cause.

Next bounded test: implement an explicitly labelled detector-response
interface and injected-signal closure test. Keep the real-data recast gate
closed until production rates, response, efficiencies and likelihood inputs
are instantiated; no toy response may be passed off as ATLAS calibration.
Any changed physical prediction requires its own freeze and fresh validation.

Run `bash run_parameter_map.sh`; open `outputs/atlas_parameter_explorer.html`.
For off-grid points run `python3 atlas_parameter_map.py --k 2781.8 --m5 6000`.
All outputs and protocol details are included in this release.

Primary sources: [HEPData v1](https://doi.org/10.17182/hepdata.140955.v1) and
[ATLAS arXiv:2305.10894v2, 30 November 2023](https://arxiv.org/html/2305.10894v2),
Sections 1, 11–12 and Figures 9–10. This release adds no newer measurement.