# Curated Static Evidence Bundle

This directory contains secret-free, read-only evidence derived from real runs in `runs/` (which remains gitignored and append-only). No generated or fixture solution was executed on the host.

## Contents and provenance

| Path | How it was produced | What it shows |
|---|---|---|
| [`index.html`](index.html) | Hand-authored static HTML (zero JavaScript) | Self-contained landing page for the `Deployed Project URL`, linking to all reports, platform summaries, and screenshots below. |
| [`search-20261017T215022-01459f/report.html`](search-20261017T215022-01459f/report.html) | Regenerated from `runs/search-20261017T215022-01459f/summary.json` via `render_report(summary, static=True)` | Primary 2-approach + 1 own-parent revision chain (`g0-c0` `sorted-bisect`, `g0-c1` `two-pointer`, `g1-c0` `parent-feedback-revision`), dual logical/physical lineage, 5% noise-band verdicts (`speed_verdict`), and `4/4` microVMs destroyed. (Note: `g0-c0`'s raw `selection_reason` in `summary.json` reflects historical pre-K02 overwrite; the report corrects the label in view only, see below; raw `runs/` artifacts are never rewritten.) |
| [`search-20261017T200437-128ee8/report.html`](search-20261017T200437-128ee8/report.html) | Regenerated from `runs/search-20261017T200437-128ee8/summary.json` via `render_report(summary, static=True)` | Contrast 1-width supervisor-timed chain where parent `g0-c0` (`6.515 ms`, `0.98× baseline · parity`) stayed ahead of child `g1-c0` (`6.587 ms`, `0.98× baseline · parity`), demonstrating that no artificial improvement is claimed. |
| [`smoke-20261017T200218-804c8e/summary.json`](smoke-20261017T200218-804c8e/summary.json) | Copied verbatim from `runs/smoke-20261017T200218-804c8e/summary.json` (`uv run kiss smoke`) | Verified `pause -> paused -> fork` file inheritance, post-fork independence, 2 children running concurrently while base paused, and egress allowlist (`pypi.org` `200`, `example.com` `URLError`); `3/3` VMs destroyed. |
| [`fixtures-20261017T195514-b9747e/summary.json`](fixtures-20261017T195514-b9747e/summary.json) | Copied verbatim from `runs/fixtures-20261017T195514-b9747e/summary.json` (`uv run kiss failures`) | 5 hand-authored remote failure fixtures (never attributed to Mercury): `incorrect` (`test_failure`), `controlled-timeout` (`timeout`), `syntax-error` (`compilation_error`), `input-mutation` (`test_failure`), and `timing-spoof` (`time.perf_counter` monkeypatch → `protocol_error`); `6/6` VMs destroyed. |
| [`audit-20261017T215138-6151e4/summary.json`](audit-20261017T215138-6151e4/summary.json) | Copied verbatim from `runs/audit-20261017T215138-6151e4/summary.json` (`uv run kiss audit`) | Read-only reconciliation of all 34 CreateOS sandboxes ever recorded in `runs/*/events.jsonl`: `34/34` verified `destroyed` (`unresolved: []`), plus `20,350` provider-reported tokens across 16 Mercury 2.5 calls. |
| [`captures-20261017/dashboard-completed.png`](captures-20261017/dashboard-completed.png) | Local Chromium screenshot of `http://127.0.0.1:8097` | Completed ledger view showing `search-20261017T215022-01459f` and `search-20261017T200437-128ee8`. |
| [`captures-20261017/dashboard-live-running.png`](captures-20261017/dashboard-live-running.png) | Local Chromium screenshot of `http://127.0.0.1:8097` | Live in-progress state captured during `search-20261017T215022-01459f`. |
| [`captures-20261017/report-search-20261017T215022-01459f.png`](captures-20261017/report-search-20261017T215022-01459f.png) | Local Chromium screenshot of `/report/search-20261017T215022-01459f` | Standalone report rendering task contract, dual lineage, timings, and candidate source. |

**View correction (K02, display only):** `render_report` and the dashboard now show a stored `selection_reason` that starts with `not selected` as `selected as revision parent of <children>; superseded by <best> (historical K02 text corrected in view; raw summary.json unchanged)` when that candidate has children in the same summary. In `search-20261017T215022-01459f`, Round 0 winner `g0-c0` (revision parent of `g1-c0`) is now labeled correctly, while `g0-c1` keeps `not selected`. Both reports above were regenerated with `render_report(summary, static=True)` from `runs/<run-id>/summary.json`; the `200437` output is byte-identical and `runs/` was not modified.