Scene-release-to-alteration intervals in comparable American cases

Frozen criteriafrozen 2026-08-26 · hash 14c82614b623

Fixed before data collection; any later edit fails the site build. A correction would be a new study superseding this one.

search protocol (re-runnable)

Multi-searcher discovery, run only after this freeze, protocol verbatim per searcher: (1) OpenAI deep research (o4-mini-deep-research, background mode); (2) Gemini deep research (deep-research-preview, background mode); (3) the executing agent's own web search. Literal queries, each run in all three: ["scene released" OR "released the scene" shooting investigation]; [shooting site demolished OR razed OR "torn down"]; [shooting site repaved OR renovated OR reopened days]; [school shooting building demolished decision]; [festival OR concert shooting grounds reopened cleanup]; [nightclub shooting building demolished OR memorial]; [crime scene released cleanup authorized university OR city council]. Per-candidate follow-up queries pairing the venue name with: "released the scene", "cleanup", "demolition", "repaved", "reopened", "preserved for trial". Sources searched: AP and Reuters archives, papers of record local to each candidate, municipal/university press releases and council minutes, court records where preservation or spoliation was litigated. Union of all three searchers' candidate lists; overlap between searchers reported in the limitations as the coverage indicator. Deep-research output is a brief under casework convention: it proposes candidates and is never cited; every row cell is verified natively against its own source before entering the table.

candidates known at freeze time — dispositions pre-committed

Method

Two-phase, per the frozen protocol. Discovery: the three searchers run the protocol independently; their candidate lists are unioned with the pre-committed dispositions above applied first, and newly discovered candidates dispositioned against the frozen criteria with reasons recorded. Extraction: for each included case, the three variables (release date, first-alteration date and nature, authorizing party) are filled cell by cell, each cell citing its own source with the standard verification label; a cell that cannot be sourced is marked, not estimated. Preserved scenes enter as censored rows. Findings state the distribution in plain language (no percentiles at this sample size), place the UVU sequence against it on the two discriminating variables — sequence (did alteration precede formal release?) and authority (who ordered it?) — ahead of raw speed, state the direction of failed searches, and report the searcher-overlap coverage indicator. Either outcome publishes.

Findings

• Rapid physical alteration after a scene is released is documented and unremarkable in comparable cases when the property owner or a municipality orders it: Boston's pre-planned city restoration began one minute after the FBI's formal handback; Sutherland Springs' sanctuary was patched and repainted by the owner within seven days; the Nickel Mines schoolhouse was demolished by its owners on day ten. Every fast alteration in the table pairs speed with a named, undisputed authorizer: where a formal release is documented, alteration followed it; where no release record was found, the owner acted openly with no investigative objection anywhere on the record — the post-release reading there is inference, labeled as such in each row.→ ledger record KRK-E057
• Where investigators or courts asserted continuing interest, alteration waited: weeks under Astroworld's court freeze, years for Parkland and Tree of Life under trial preservation, and indefinitely at Uvalde, where the district attorney has withheld release for over four years despite the owner's 2022 demolition decision. No case in the table shows alteration proceeding over a disputed or absent authorization.→ ledger record KRK-E058
• Placed against this table, the UVU courtyard's four-day event-to-pavers interval sits inside the observed fast tail (zero to ten days) — on speed alone it is not an outlier. What no comparable case shares is its profile on the two frozen discriminating variables: it is the only case in the table where the authorization is disputed on the record (the lead investigator testifying he did not authorize the work) and the only fast alteration whose formal release date is absent from the public record.
• Direction of failed searches: formal scene-release dates could not be found under the protocol for seven of fifteen venues — release dates are routinely unrecorded in public reporting, so the absence of a published UVU release date is itself common; a disputed authorization is not. A specific search for comparable cases where alteration preceded formal release found none.
• Release latency, where documented, runs from seven days (Boston — after the manhunt had concluded; Sandy Hook's seven days of processing had the perpetrator dead at the scene) through 33 days (court-frozen Astroworld) and 47 days (Route 91, perpetrator dead on the first night), to years or never where courts or prosecutors held an interest. This places the target case's unknown release date at one of two extremes: if the courtyard was formally released before the September 14 paving, that release — within four days of the event, during an active prosecution with the case's central physical-evidence disputes unresolved — would be the fastest in the table, where every release inside a week followed a closed perpetrator question; if it was not, the alteration preceded release, for which the table contains no precedent at all. The undocumented release date determines which extreme, and nothing in between; the GRAMA record (KRK-R007) decides it.

Limitations

• Coverage: two of three searchers completed (the executing agent's own search and an OpenAI web-search run; the OpenAI run found no candidates beyond the agent's union, the intended overlap signal). The OpenAI deep-research models were unavailable to the project's API key, so an o4-mini web-search run substituted (a protocol deviation); it failed twice on organization rate limits before completing. The Gemini deep-research run's outcome is recorded in the collection PR; the union above does not depend on it.
• Seven of fifteen release-date cells could not be filled from public reporting; intervals for those rows are measured from the event, not the release, and are marked as such.
• Release-before-alteration is directly documented in six rows (Boston, Sandy Hook, Route 91, Astroworld, Parkland, Tree of Life); in the remaining fast-alteration rows it is an inference from open owner action and the absence of any investigative objection on the record, and each such sequence cell says so. Observation and inference are not pooled.
• Interval figures are day-precision from published dates; "~" marks cells where a month but not a day is documented.
• Comparability is venue-based per the frozen criteria; it does not control for victim prominence, political salience, or media pressure, all of which plausibly affect alteration timing.
• The author knew the target case's values before the freeze (pre-registration cannot prevent pre-hoc criteria tuning; see the freeze PR's panel question). The known-candidates declaration and post-freeze discoveries — seven of fifteen rows were discovered after the freeze — bound but do not eliminate this risk.