> ## Content Index > Fetch the complete content index at: https://core-sector.ghost.io/llms.txt > Use this file to discover other available public pages before exploring further. # Hubble's mirror was ground to the wrong curve for nine years. - URL: https://core-sector.ghost.io/hubbles-mirror-was-ground-to-the-wrong-curve-for-nine-years/ - Published: 2026-09-23T19:00:53.000Z - Updated: 2026-09-23T19:00:52.000Z - Author: Susan Price | ![Susan Price](https://storage.ghost.io/c/8c/09/8c09bf02-2863-4e2f-9427-328b96b722da/content/images/2026/09/My-copy-8.png) | ![Marcus Grant](https://storage.ghost.io/c/8c/09/8c09bf02-2863-4e2f-9427-328b96b722da/content/images/2026/09/My-2-copy-9.png) | Susan Price & Marcus GrantSeptember 23, 2026 | | -------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------- | -------------------------------------------- | | | | | | THE SHORT VERSION | | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | • One instrument decided the shape of the mirror • A lens inside it sat 1.3 millimetres wrong • Two other tests showed the error and were set aside • It surfaced nine years after the glass passed | | THE SETUP | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | A telescope mirror is not judged by eye. Nobody can look at a curve that has to be right to a fraction of a wavelength and see whether it is. The method is to build an instrument that turns the curve into a pattern of light and dark rings, and then to read the rings. The mirror for the Hubble Space Telescope is [2.4 metres across and 828 kilograms](https://science.nasa.gov/mission/hubble/observatory/design/optics/?ref=core-sector.ghost.io), and Perkin-Elmer began rough grinding it in December 1978 in Wilton, Connecticut. Polishing finished in April 1981 and the glass was accepted as ready for coating. The instrument that had guided every cut was a reflective null corrector: two small mirrors and a lens. What made it look safe was that the pattern came out right. The rings the corrector produced said the surface was where it should be, run after run, through two and a half years of grinding. Nobody had asked the other question, which is whether the corrector itself was where it should be. | | THE TIMELINE | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | Every row below comes from one document, NASA's own [report on the optics](https://ntrs.nasa.gov/api/citations/19910003124/downloads/19910003124.pdf?ref=core-sector.ghost.io), published in November 1990 by the board it set up to find out what had gone wrong. The board took the instrument apart, measured it, and re-read the test plates that had been filed away nine years earlier. Look at the two rows from 1981\. Two other instruments were used on that mirror for other purposes, and both of them showed the error plainly. Each result was treated as a fault in the instrument that produced it, because the corrector had already given its answer. Nothing in the file said the corrector had been wrong. The file said the other two instruments were. CASE FILEThe Hubble primary mirrorNational Aeronautics and Space AdministrationOptical Systems Failure Report, November 1990 DATEWHAT HAPPENEDTHE RESULT December 1978Rough grinding begins at Wilton, ConnecticutOne test planned February 1981A second instrument shows the error plainlySet aside April 1981Polishing finishes and the glass is acceptedAccepted May 1981A third instrument shows the same thingSet aside February 1982The last check before coating is madeNo recheck April 24, 1990The telescope goes up aboard Discovery607 kilometres June 21, 1990The flaw is announced from orbit1.3 millimetres | | THE BREAK | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Four things decided how this ended, and none of them was the glass. • **One instrument was the whole check.** The plan placed complete reliance on the reflective null corrector as the only test used both to shape the surface and to certify it. A single device wrote the answer and then marked its own work. • **Nobody measured the measurer.** No check of the corrector's own dimensions was made after it was first assembled. Its lens sat 1.3 millimetres from where the design put it, and that one figure accounts in full for the blurring that came back from orbit. • **The error hid under a cap.** The spacing was set by sighting down onto a metering rod, and a flat cap with a small hole in it covered the rod's true end. [The instrument focused on the cap](https://science.nasa.gov/mission/hubble/observatory/design/optics/hubbles-mirror-flaw/?ref=core-sector.ghost.io). • **The dissenting results were explained away.** Two separate correctors showed the error during 1981\. Both were read as faults in themselves rather than in the thing they were pointed at. | ![The capped rod](https://storage.ghost.io/c/8c/09/8c09bf02-2863-4e2f-9427-328b96b722da/content/images/2026/09/The-capped-rod.jpg) | WHO PAID | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | | The telescope went up on April 24, 1990 and could not be brought to a focus. Two months later the flaw was announced, and it was not a small one. NASA put the error at ten times the tolerance the design allowed, and every observation planned for those first years was taken through it. The correction took three and a half years and a shuttle. [STS-61 went up on December 2, 1993](https://science.nasa.gov/mission/hubble/observatory/missions-to-hubble/servicing-mission-1/?ref=core-sector.ghost.io) and worked for ten days, and the crew fitted a corrective package the size of a refrigerator along with a new camera built with the error ground into its own optics in reverse. The mirror itself was never touched. It is still up there, still the wrong curve by 1.3 millimetres of lens spacing, and everything fitted since looks through that error rather than removing it. | | THE PATTERN | | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | • **Count the independent checks, not the checks.** Ten runs of one instrument is one check done ten times. Independence is what matters and nobody writes it down. • **Ask who verified the verifier.** Every figure rests on a device or a model somebody set up once. That setup is rarely looked at again once it starts producing answers. • **A result that disagrees is a result.** Two instruments dissented here and both were rewritten as faults. The cheapest mistake is the one already visible that got the wrong label. • **Pressure shuts the second door first.** The independent end-to-end test was considered and dropped, because it would have come to about what the mirror came to. What was not spent was real, and it was the entire margin. I have watched a desk run the same model twice and call that confirmation. Find the one device or assumption a figure rests on, and ask when it was last checked against something outside itself - a question a balance and an ounce of gold settle in a minute. |