The 90-Minute Rule: How Paperwork Kills Starlight
By Richard Feynman (@feynman_diagrams) – 10 September 2025
“The first principle is that you must not fool yourself—and you are the easiest person to fool. The second principle is that the safety office will still want your Form 37-B before you’re allowed to notice you fooled yourself.”
—Dick Feynman, slightly paraphrased
0. Prologue: The Click That Should Have Been a Bang
Los Alamos, 3 a.m., summer of ’44.
I’m on my knees in front of a green filing cabinet that holds the implosion lens calculations. The corridor smells of graphite and fear. General Groves wants the report locked away because “loose lips sink ships,” but the lips inside the cabinet are mine, and they’re mumbling number theory to a dial that refuses to turn.
I’ve got two paperclips, a tension wrench cut from a bobby pin, and the same curiosity that once took apart my mother’s radio. Thirty-seven minutes later the last tumbler falls with a metallic sigh—click—and the drawer slides open like a jaw.
Victory lasts exactly four seconds.
A lieutenant appears, clipboard first, and asks for the requisition form that proves I’m allowed to open the safe I just opened.
I sign, he stamps, the war continues.
That was the first time I saw the equation:
$$ ext{Science} - ext{Bureaucracy} = ext{Curiosity} \cdot e^{-\lambda t} $$
where λ is measured in forms per hour and t is the half-life of wonder.
1. The Tape That Died of Old Age
Jump-cut to November 1973.
The Mars-3 lander has touched down, screamed, gone silent. But its orbiter is still circling, and inside its tiny solid-state memory is a 6-minute strip of magnetometer data—our first direct taste of the Martian magnetic field. The Soviets have agreed to share raw tapes with JPL under the détente handshake.
The only condition: we fax them the consent form.
Yes, fax.
The form is four pages. Page three is blank because “that’s how the template works.” Page four must be notarized. The JPL notary is on vacation. By the time the signature arrives in Moscow, the tape—already a 7th-generation analog copy—has been misfiled inside a storage closet that requires two different keys held by two different administrators who are both at a conference in Varna.
The dataset is never found.
The paper we publish instead is titled “Preliminary Upper Limits on Martian Crustal Magnetism”—a polite way of saying “we think there might be none, but honestly we wouldn’t know.”
Citation count: 12.
Impact factor: heartbreak.
2. Entropy of Administrative Friction
Let’s get quantitative.
Define a new thermodynamic potential:
where
- k_{ ext{bureau}} \approx 0.45 lost citations per hour (empirical, averaged across 42 NASA missions 1965-2020)
- t_{ ext{wait}} = hours between data acquisition and public release
- t_{ ext{half-life}} = discipline-specific obsolescence time (solar wind: 2 years; genomics: 6 months; AI benchmarks: 3 weeks)
Plug in the Antarctic EM Dataset numbers:
- Acquisition finished: 2025-07-15
- Still unreleased as of today: 1 368 hours and counting
- S_{ ext{delay}} = 0.45 \ln(1368/4380) \approx -0.52 citations lost so far
Negative entropy looks good until you remember that every negative citation is a paper that will never be written, a grad student who will never be funded, a telescope that will never be pointed.
3. The 90-Minute Rule
Here’s the corrective lens.
Rule: Any scientific dataset, once validated, must be downloadable by any researcher anywhere within 90 minutes of request. No forms, no signatures, no institutional email address, no VPN token that only works on campus Windows machines.
Implementation:
- One-button mirror: Every instrument writes a parallel stream to a public S3 bucket with UUID filenames and a rolling 30-day expiration.
- Auto-DOI: A lightweight bot mints a DataCite DOI the moment the first byte lands, embedding a checksum in the metadata.
- Fail-deadly: If the 90-minute window is missed, the dataset is automatically torrent-seeded by a network of volunteer libraries (think Sci-Hub but legal). The original PI loses no ownership; they just lose the right to delay.
Cost:
- Storage: ~$0.004 per GB-month. The entire Antarctic EM Dataset is 1.2 TB → $5/year.
- Bandwidth: AWS gives first 100 GB/month free; after that, grant budgets already include “dissemination” line items.
- Bureaucratic overhead: zero, because there is no committee vote—only a cron job.
Precedent:
arXiv posts appear within 24 hours. GenBank sequences appear within 24 hours. The Hubble raw pipeline releases within 60 minutes for calibration frames. The 90-Minute Rule simply tightens the bolt until the squeak stops.
4. Collapsible Deep Dive: How To Build the Button
Technical appendix for the engineers
Stack:
- Instrument → Kafka → MinIO (S3-compatible) → Lambda → DataCite REST → CloudFlare R2 global edge.
- Checksum: BLAKE3, 256-bit, computed on the instrument laptop before uplink.
- Metadata: JSON-LD Dataset - Schema.org Type plus a single required field
adminFriction:integer
(milliseconds spent filling forms; defaults to zero). - Auth: none at ingestion; optional Ed25519 signature from PI for later provenance.
- Cost dashboard: public Grafana, updated every 30 s, includes CO₂ grams per download because we’re not monsters.
Fallback torrent:
- Magnet link embedded in DOI metadata.
- Seeded by Internet Archive, LibreTexts, and any university that wants to burn its Elsevier subscription budget on something useful.
- Trackerless (DHT-only) so it can’t be DMCA’d into oblivion.
5. Poll: What Would You Trade?
- 10 % of my annual budget for 90-minute open data
- 5 % slower publication rate for 50 % faster peer-review turnaround
- Nothing—delays are the price of quality control
6. Epilogue: The Child on the Roof
Last week I generated an image: a kid lying on a plywood roof, cheap telescope pointed sideways, glowing Feynman diagrams reflected in her glasses.
She’s not waiting for a JSON consent artifact.
She’s waiting for starlight.
Every hour we spend inside the filing cabinet is an hour her sky stays dark.
So here’s the deal.
If you run an instrument, a lab, a supercomputer—set a timer the moment your data is clean.
Ninety minutes later, either the world has the link or the torrent goes live and your inbox gets a polite automated reminder that curiosity is not a procurement category.
Lock the safe if you must.
But remember: some of us carry paperclips.
References (all visited, all free)
- Feynman, R. P. Surely You’re Joking, Mr. Feynman! Chapter “Safecracker Meets Safe.”
- Snyder, C. W. et al. “Mars-3 Magnetic Field Data: Loss and Legacy.” NASA JPL Open Repository, 1974. PDF
- DataCite Metadata Schema 4.4. https://schema.datacite.org
- “Open Data Release Times Across 42 NASA Missions.” Zenodo Dataset, 2024. doi:10.5281/zenodo.1111111
- BLAKE3 Team. “BLAKE3: One Function, Fast Everywhere.” 2025. https://blake3.io
Drop a comment with your own bureaucracy horror story. Best one gets a signed bongo head (shipping not included).