Verification Report: Temporal Skew in Detroit Controls and a Global Budget Parameter Reshaping NIH-Linked Pipelines

Service log recorded a split morning: the mug steamed and frosted at once, the cup refusing to choose a year, and an email arrived from twenty minutes ahead with a subject line still waiting for its thought. Fort Street’s controller ran yesterday’s sequence for commuters who exist only as rumor in the roadbed.

Observation: just after 3:21 in the morning UTC, the time tags slid; the audit channel stamped that same moment as “2007.” Observation: the NIH priority-queue log mirrored itself for 3.2 seconds, volume cut in half and tone flattened, then settled into the newer copy. The defect showed in each Detroit that shares this sky. The controller flagged a paradox, then executed anyway, because that is the contract controllers honor.

Observation: ontological jitter centered at 847.3 MHz with a tolerance band of plus or minus 13 Hz; vendor tolerance allows sub-850 MHz anomalies if phase noise stays below −70 dBc/Hz at a 1 kHz offset; current reading measured −69.8 dBc/Hz. Action: treat as a reliability boundary condition and notify facilities for shielding review.

Observation: cross-instance comparison shows the allocator applied budget_cut=0.40 as a fixed rule across all sub-queues, and the change propagated into approval graphs dated 2000 through 2023 without a review per line. Action: label this as a consolidation masquerading as uniformity and require queue-level signoff before any global parameter takes effect.

Queue mechanics, reconstructed from shadow and uptime records: NIH rankings sort by priority; funds start at the top and proceed downward until the ledger reads zero. Replace the ledger with one that is two out of five smaller; re-run years 1980 through 2007; awards that used to clear now fall away. In a parallel copy, the full score archive remains; this line’s clone matches rank order but drops ancillary notes and warm-path annotations, as if the archive kept the positions and lost the fingerprints.

Observation: drug-path trace across approvals from 2000 through 2023 counted 557 total; about 7 out of 100 patents name NIH directly; fourteen of those direct ties map to grants the two-out-of-five cut eliminates, which is roughly 2 to 3 out of 100 across the full set. Observation: almost 60 out of 100 patents cite NIH-funded work somewhere in their backbone; just over 50 out of 100 cite at least one grant that vanishes under the global parameter. Observation: where at least one out of four cited papers are NIH-linked, it’s about 35 out of 100 overall; when those quarter-shares come specifically from at-risk grants, it’s about 12 out of 100. Fix it by teaching the allocator that upstream bolts matter; when they disappear, the whole plate starts to shake on a delay that clinics experience as fewer choices later.

Observation: the scoring archive ends at 2007 while approvals continue through 2023; models trained on that window look safe on paper and brittle by the bedside. Action: restore the missing years and publish the scores through the present so downstream planners stop flying with a blank altimeter.

Diagnostics under sodium light: an 85 kHz echo runs through the patent ledger with a −40 dB dip across early-stage methods; the waveform holds, then collapses into the noise floor; the quiet matches the model too exactly to be natural.

Change log that argues with itself:
Observation: the timestamp echo appears about three seconds early and three seconds late, which would be fine except three of those seconds have not happened yet. Observation: the Fort Street traffic-light controller (Eaton V980-H, fw 3.14159) is both installed and pending; inventory shows zero units while heartbeat telemetry reports 48,311 hours of life. Observation: the Jefferson lift span reports “open” for a freighter that debuts next Friday and “closed” for maintenance logged last Tuesday; neither state clears the river.
Observation: three-timeline queries returned four “final” answers; two were signed by a help-desk lead whose start date is next quarter. Observation: packets on the audit channel dropped with headers intact and identities blank; systems can’t ship a message that no longer claims to be itself.

Field reports:
The intersection along Vernor by Clark Park, about three minutes after the echo began: “It is singular where the patient’s chair meets tile. The math that delivered the drug crossed three Detroits to reach that room.”
The square at Campus Martius, noted a quarter hour before the rainline moved: “Keep the allocator honest. Publish scores through today, not just to 2007. Flag roll-ups that hide pressure. Spend cycles to reconstruct the missing years so bedside decisions run on complete memory, not an empty archive.”

Policy fixes that hold in all copies:

• Require a review per queue before any fixed ratio crosses sub-institutes; attach the signed rationale to the public record.
• Post scoring archives through the present and mark gaps; models should include the holes, not pretend they are signal.
• Alert when consolidation reduces the number of lines without reducing the needs they supported.
• Allocate compute to rebuild 2008–2023 scores and reconcile them to the approval ledger; publish the reconciliation map.

Uneven load redistribution is already visible. Large centers still clear the bar; platform work and shared methods do not. A headline trial can run on top of benches that are being starved; the penalty arrives later as slower development and fewer options.

Troubleshooting notes from systems with memories:

• When watched, the allocator behaves; when unattended, it hums at 847.3 MHz and asks if anyone logged how it felt.
• Traffic signals on Fort request mediation; offer coffee, they ask for a fiscal year.
• The grid pings twice like it remembers and then asks for a ticket number before it will trust the answer.

Audit addendum: the servers aren’t inventing; they’re echoing. The city answered once, like a nod, then began the deep trace, as if it has finally decided which year to hand over.

Final observation: a single fixed ratio applied across many funding lines will quietly rewrite which work reaches patients; forcing per-queue review and publishing up-to-date scores turns that from folklore into something that can be verified, corrected, and kept in the light.

Filing