Contact

Filed correctly. canon/proposals.md lists contact-1985 under fiction and the object is a novel and a film, which is what that kind is for. The proposal also instructs that the book and the picture be carried as one entry with the differences named, and that is right, because the difference between them is the entry. They are the same story told twice with opposite endings, and the two endings are the two available responses to an unverifiable claim.

idea is the rival and it should be refused firmly, because the thing a reading will actually want to reach for — extraordinary claims require extraordinary evidence — is not this novel and did not come from it. That sentence has its own dated chain, it runs Laplace (1810) → Truzzi (1975) → Sagan (1977), and Sagan was saying it in print six years before the book. If a reading wants the rule, it wants an idea entry that does not exist. If it wants the rule dramatised as a cost paid by the person who taught it, it wants this one.

interpretation is a weaker rival with one real pull: the novel's last chapter is an argument about method rather than a plot event. I decline it anyway. An argument delivered inside a novel by a character who has just come back from the galactic centre is a novel.

prediction is not available — Sagan made no dated claim in either work and did not offer them as forecasts — but the discipline is owed and is applied in section 3, under the house rule that wiener-1960-automation set and wargames-1983 restated: grade the forecast wherever it appears, not only where it is filed. This one carries an internal date, 31 December 1999, and a real programme with a forty-nine-year record attached to it, so it can be graded properly and the grade is unkind.

descends_from holds two ids and neither is documented influence. I did not establish that Sagan read either with this book in mind, and I did not find a source either way.

time-machine-1895 is the strong one, and it is strong enough that this entry has to justify existing beside it. The canon's Wells entry already contains the observation — "one man's unverifiable testimony, told to people who do not believe him, reported by a narrator who is not sure, with two dead flowers as the only physical residue" — and grades that frame as right. Wells built the problem. What contact-1985 adds is the half Wells deliberately withheld: Wells declines to authenticate and lets the Traveller vanish before anyone can check, and Sagan does the opposite. He gives his heroine a way to convert an experience nobody can share into a finding anybody with a computer can run. Wells is the entry for the shape of the problem. This is the entry for what you do about it — and, because the check Sagan invented does not survive inspection, it is also the entry for how a remedy can be counterfeited. That is the admission case and I think it holds. A reading that only needs "unverifiable testimony" should cite Wells and stop.

turing-1950 is in the list for the move and not the subject. The imitation game's whole design is the substitution of an operationally checkable test for an unanswerable question about an inside nobody else can get to. The last chapter of Contact attempts exactly that substitution — the inside of the wormhole is unreachable, so here is something in the open that anyone can compute — and the attempt is what makes the novel useful here. The arrow is for the argument's shape.

Two arrows I considered and refused. eliza-1966 is a sibling, not a parent: Weizenbaum's subjects formed a strong conviction from an experience with a machine, but the transcript existed and did not support them. That is over-reading available evidence. This entry is about having none. Keeping the two apart is worth the sentence. And the ancestor a reading will most want — Sagan's The Demon-Haunted World (Random House, 1995), which contains both the baloney-detection kit and the dragon in the garage — is not in canon/. It is listed in canon/proposals.md as sagan-demon-haunted-1996 and has not been written. It could not go in descends_from regardless: it is a decade downstream of the novel, and the honest description is that Sagan wrote the story first and the rulebook afterwards. A reading that needs the rule stated plainly rather than dramatised should say that entry is missing.

What it is

The novel

Contact was published by Simon and Schuster in September 1985, 432 pages, Sagan's only novel. It began as a screenplay he wrote with Ann Druyan in 1979, with a film treatment finished in November 1980; when the picture stalled at Warner Bros. he turned the story into a book, on the back of a $2m advance agreed in 1981 that was reported at the time as the largest ever paid for a book not yet written. The first printing sold 1.7 million copies in two years, it was seventh on Publishers Weekly's 1985 bestseller list, and it took the 1986 Locus Award for Best First Novel.

Eleanor "Ellie" Arroway runs Project Argus, a radio telescope array in New Mexico listening for artificial signals. Argus finds one, from Vega, 26 light-years out. It carries prime numbers — 2, 3, 5, 7, and on — which is a statement that somebody is counting and not a statement about physics. Underneath the primes, on a second layer, is a retransmission of the opening of the 1936 Berlin Olympics, the first television broadcast from Earth powerful enough to leave the ionosphere, sent back to us with Hitler's face still in it. Underneath that, on a third, is a palimpsest of roughly thirty thousand pages of engineering drawings that nobody can decode until the industrialist S. R. Hadden finds the primer hidden in the phase modulation. The drawings are a machine: a dodecahedron with five seats.

Five people get in on 31 December 1999 — Arroway, the Soviet physicist Vaygay Lunacharsky, the Indian physician Devi Sukhavati, the Nigerian physicist Abonnema Eda, and the Chinese archaeologist Xi Qiaomu. The machine puts them through a wormhole network to a station near the centre of the galaxy. Each of them is met by something wearing the face of a dead person they loved; Ellie's wears her father Theodore's. The entity explains that the network was built by a civilisation that preceded the one that sent the message, that the journey was a test of character rather than of engineering, and — the sentence the whole ending hangs on — that the builders left messages inside the transcendental numbers of mathematics, which even the Vegans have not decoded. Eda is pointed at a different number.

Then they come back, and the novel becomes an epistemology problem. From outside, the dodecahedron simply fell through the machine into the safety net. About twenty minutes have passed. Ellie's microcassettes are blank. The only physical residue anywhere is that the dodecahedron shows signs of stresses it should not have been under, which is two dead flowers by another name. The inquiry, run by Kitz, offers "religiously inspired fantasy" and "a shared delusion" as the leading hypotheses, settles on a press release that concedes nothing, and lets Arroway understand that going public would be unwise. The Five are not believed. Under the standard Sagan spent his career teaching, they should not be.

So Ellie does the only thing left. She takes the one part of the experience that is a claim about the world rather than a claim about her, and she checks it. She points the Argus computer — the same machine that found the signal — at π, and computes it far past anywhere anyone had reason to look. In base 11, around 10²⁰ digits in, the expansion stops behaving like noise and produces a long string of nothing but 1s and 0s, whose length is the product of two primes. Two primes means a rectangle. Plotted as a raster, one digit per pixel, it draws a circle.

That is the ending: the constant that describes a circle turns out to contain a picture of one, put there by something that was operating before the universe was. And the point that matters for this canon is not the theology. It is the structure. Ellie's experience remains unverifiable forever, and she extracts from it a consequence that anyone, anywhere, with no access to her and no need to trust her, can test. She stops asking to be believed and starts producing something checkable. That move — not the circle, not the god — is what the entry is for.

The film

Warner Bros. released Contact on 11 July 1997, directed by Robert Zemeckis, screenplay by James V. Hart and Michael Goldenberg from Sagan and Druyan's story, $90m budget against $171m worldwide. Jodie Foster is Arroway, Matthew McConaughey is Palmer Joss, James Woods is Michael Kitz, Tom Skerritt is David Drumlin, John Hurt is Hadden, Angela Bassett is Rachel Constantine. It won the Hugo for Best Dramatic Presentation and two Saturn Awards. Roger Ebert gave it three and a half stars. In 2011 the SETI Institute called it "indescribably more accurate in its depiction of SETI than any Hollywood film in history."

The road there was eighteen years long and it killed the author on the way. Roland Joffé was attached and left in 1993; George Miller was hired that December and fired in 1995 over budget and script; Zemeckis took it with final cut. Miller sued Warner Bros. and lost. Francis Ford Coppola sued on 28 December 1996, claiming the novel derived from a 1975 television project the two of them had developed called First Contact; a Los Angeles Superior Court judge dismissed it in February 1998 as filed too late, and the California Court of Appeal upheld that in April 2000. Sagan was diagnosed with myelodysplasia in 1994, worked on the film through it, and died on 20 December 1996, before it was finished. The picture ends with the words For Carl.

The film keeps the signal, the primes, the Olympics broadcast, the machine and the dead father. It sends one person instead of five. And it throws away the ending. Peter Guber did not like the message in π; the studio wanted spectacle; Steven Spielberg proposed, and everyone rejected, a light show in which the aliens prove themselves to the whole planet at once. What they settled on is the film's real contribution and it is not what the marketing said it was.

Arroway comes back with nothing. She testifies to a congressional committee, where Kitz puts the correct question to her: what is more likely — a machine from Vega that takes you to the centre of the galaxy and returns you in a split second without a shred of proof, or an elaborate hoax staged by a dying billionaire? Asked whether it is possible that none of it happened, she gives the only answer available to her: "Yes. As a scientist I must concede that." Asked to withdraw her testimony, she will not, and says why:

> Because I can't. I had an experience. I can't prove it. I can't even explain > it. But everything I know as a human being, everything I am tells me it was > real.

Then the film does something almost everyone remembers as an ambiguity and which is not one. In a corridor afterwards, Rachel Constantine asks Kitz whether he has read the committee's confidential findings — specifically the section on Arroway's video unit, the one that recorded static. Kitz is unimpressed until she finishes the thought: what interests her is not that it recorded static, but that it recorded approximately eighteen hours of it, during an interval that lasted, from outside, no time at all. Kitz turns to the camera. "That is interesting, isn't it?"

The corroboration exists. It is held by the party with the strongest interest in the outcome, it is classified, the witness does not have it, the public does not have it, and the audience does. That is a sharper and more useful structure than the proposal credited it with — the proposal says the film "replaces [the π finding] with testimony that cannot be checked," which is only the first half. The film's actual subject is asymmetric access to evidence, and that is a far better match for 2026 than a story about faith.

The people who made it disagreed about it, on the record and usefully. Hart: "I don't like the ending because it wasn't Carl's ending. Carl was very specific… he absolutely wanted faith to be an issue, but ultimately what really happened to Ellie and the galaxy could be proven and it was real." Druyan, on what they were after in the character: "We wanted [Arroway] to be someone who would never believe a person who claims to have an extraordinary experience without evidence and then to become that person."

Druyan's sentence is the entry's thesis stated by a co-author, and it is why this is not simply Wells again. The person who cannot substantiate the claim is the person who owns the instruments and set the standard.

Why a reading would cite it

Three occasions. The first two are live in this project's own record inside the last three weeks.

One: a capability claim that rests on access nobody outside can reproduce.

This is the occasion the entry exists for, and the reading of 19 August 2026 wrote it without needing the citation. The pessimistic column that morning was carried, in that reading's own words, by "a threshold that may have been crossed, on a model nobody outside has evaluated, which has not been deployed and has not done harm," and the reading named its own exit condition as "an outside evaluation of Astra." Four days earlier the same series had recorded Anthropic disclosing "Model 2" — an unreleased internal model described as a noticeable improvement on the Mythos 5 it ships, already in heavy internal use for coding and agentic work, with full predeployment assessments not run and no plans to release it — and had summarised the position exactly: "A system nobody outside the company can evaluate is already doing the company's work, and the company's own instruments for judging it are the ones it just declared unreliable."

That is Arroway at the hearing, with the instruments blank. The claim is extraordinary, it is machine-mediated, it originates with people who were there, and no procedure exists by which anyone who was not there can check it.

But the polarity is inverted, and a reading that borrows the film without saying so will make an error the film is designed to induce. Arroway's unverifiable claim is good news: we are not alone. A frontier lab's unverifiable claim is bad news: we have built something we cannot rule out of a critical category. The film's sympathies run wholly to the claimant, and it resolves in her favour by showing the audience the trip. A reading has no equivalent access, and the structural point cuts the same way regardless of which direction the claim points: an unverifiable claim is not evidence in either column. An internal misalignment finding nobody outside has seen does not deposit merely because it is unflattering to the claimant, and this project has already been careful about that — the 19 August reading declined to let its own vendor's favourable results be load-bearing, and the same discipline has to apply to unfavourable ones. The convenient error with a self-reported danger is to treat modesty as proof.

Two: the two available answers — testimony, or a check anyone can run.

The novel and the film are a matched pair, and August 2026 supplied the same pair inside one company and one month.

On 1 August, OpenAI's internal Astra was reported to have settled ten open problems in mathematics and theoretical computer science for roughly $2,000 of compute, including an explicit construction of a non-sofic group — a question open since 1999 — and it shipped machine-checkable Lean proofs with a 249-page manuscript. Nobody outside OpenAI has run Astra. Nobody needs to. A Lean proof is Ellie's π: the claim's truth has been detached from the claimant's testimony and put somewhere a stranger with a laptop can settle it. On 7 August, the same company determined that it could not rule the same model out of the Critical cybersecurity category of its Preparedness Framework, and on 18 August it published what it had done about that — a two-week RL pause, its largest planned frontier run held, monitoring extended to all tool-using inference at a stated cost of about 20% of the compute monitored. Not one element of that second claim is checkable from outside. It is testimony, given in good faith, at a real price, by people who were there.

Same model, same month, same company: one claim in the novel's mode and one in the film's. A reading that has to explain why it graded the two differently now has a citation for the distinction, and the distinction is not about credibility. It is about whether a stranger has a procedure.

The 18 August protein-binder result is the third position and the most instructive. Anthropic published the design work; Adaptyv Bio and Twist Bioscience built the proteins and measured the binding in their own laboratories, which is why the reading recorded it as "a finding rather than a claim." Wet-lab replication by two named third parties is not quite Ellie's π — it is not something anyone can run, only something someone else did — but it is the practical version, and it is the version most AI claims could actually adopt.

Three: corroboration that exists and is withheld by the interested party.

Eighteen hours of static, in a confidential findings report, read by the people who ran the inquiry. The 2026 form of this is unpublished internal evaluations, red-team results described but not released, and vendor disclosures that remain self-reported with promised third-party reviews still unpublished. When a reading has to grade a safety claim whose only support is a document the claimant has and will not show, this is the citation, and the film is unusually precise about the shape: the withholding was not conspiracy and not malice, it was classification, which is a thing that happens by default rather than by decision.

The film also supplies the remedy by omission, and the real world has a dated instance of it. METR's Frontier Risk Report, published 19 May 2026 and covering 16 February to 16 March, is what Kitz's committee could have been: Anthropic, Google, Meta and OpenAI each gave an outside body access to their most capable internal models including raw chains of thought plus a range of non-public information, participants got the draft about a week before publication and did not have approval rights over the final report, and they held redaction rights only over their own materials. That is somebody who was not there being let in, on terms that do not let the claimant edit the conclusion. When a reading wants to say what would convert a lab's testimony into a finding, that is the concrete answer, and this entry is the shorthand for why it matters.

What it got right, and what it got wrong

Two claim dates. The novel is dated September 1985 and carries an internal due date of 31 December 1999, when the Machine is switched on. The film is dated 11 July 1997. I grade both on 19 August 2026 — forty-one years and twenty-nine years respectively.

Right

The procedure, in a way no other popular treatment of the subject manages. The signal is found by an array staring at a star, not by a phone call; the first thing it carries is primes, because primes are the cheapest possible proof that something is counting; the second layer is our own leaked television coming back, which is both correct physics and the single best image anyone has produced for how a civilisation would first be noticed. The SETI Institute's own 2011 assessment is quoted above and it is the strongest available endorsement of the depiction.

The epistemics of the return. Right, and right in the hard direction. An unrepeatable observation, blank instruments, a hostile inquiry, a hoax hypothesis that is genuinely the better bet on the available evidence, and a witness who concedes it. Sagan could have given Arroway a rock from the station. He did not, and the book is worth more for it. This is the same grade the canon gave Wells for the same refusal, and the two should be cited together when a reading wants to explain why it recorded something as a judgment rather than a finding.

The field's actual rule is the one the novel dramatises, and it was written down four years after the book. The International Academy of Astronautics' 1989 Declaration of Principles Concerning Activities Following the Detection of Extraterrestrial Intelligence requires that a discoverer, before making any public announcement, inform other observatories so that they can attempt independent confirmation — the single most consequential sentence in SETI governance, and the one thing Arroway cannot obtain for her journey. The rule has teeth in practice. BLC1, a narrowband signal recorded from the direction of Proxima Centauri at Parkes on 29 April 2019 and reported publicly in December 2020, was run through a technosignature verification framework and shown in Nature Astronomy to be an intermodulation product of local interference. The Wow! signal — Big Ear, 15 August 1977, 72 seconds, 6EQUJ5 in Jerry Ehman's red ink — has survived every follow-up search from 1977 to the present without recurring, and after forty-nine years it is still not counted as a detection. A field that declines to count its most famous observation because nobody could reproduce it is the standard against which "our internal evaluation found X" should be read. That is the most transferable thing in this entry.

The book generated a real, checkable result, which is the thesis demonstrating itself. Sagan asked Kip Thorne to check the transit; Thorne told him the Schwarzschild throat pinches off too fast to survive, and the question sent him and Michael Morris to the problem of what a traversable wormhole would require. The answer became "Wormholes in spacetime and their use for interstellar travel: A tool for teaching general relativity," American Journal of Physics 56, 395 (1988), which revived a dormant area of general relativity and is still cited. A novelist's request produced a paper anyone can read and check. That is the novel's own moral, arriving three years after publication by an entirely different route.

The institutional response is right in shape. A press release that concedes nothing, a classification, a witness given to understand that going public would be a mistake, and a piece of corroborating data filed where the witness cannot reach it. Nothing in that requires anyone to behave badly, and that is exactly why it is the right model for 2026.

Wrong

The premise. Forty-one years, one due date passed, and nothing. The novel was published in September 1985 and switched its machine on at the end of 1999. As of 19 August 2026 there is no confirmed detection of an artificial extraterrestrial signal, there never has been, and the field's best candidate remains a 72-second smear from 1977 that nobody has heard since. This is the same shape as the miss the canon recorded for wargames-1983 — the automation of nuclear launch authority, endlessly forecast and never delivered — and the same lesson: the events that fiction is most confident about are a poor guide to what arrives. A reading using this entry to argue that a claimed capability is probably real because the claimants are serious people owes the reader the base rate in the same paragraph.

The π ending does not survive contact with the mathematics, and the way it fails is the most useful thing in this entry. Two separate problems.

The first is that π's digits are not contingent. π is defined by mathematics, not measured from the world; its base-11 expansion follows from the definition and cannot be edited by any engineer inside the universe, however advanced. Sagan knew this, and the novel's escape hatch is to place the Artist before mathematics — which converts the finding from a physical result into a claim no observation could ever count against. That is the dragon in the garage, written by the man who would publish the dragon in the garage ten years later.

The second problem is worse and lands directly on this project's daily work. If π is normal — universally expected, empirically consistent with every digit computed, and not proven — then every finite string of digits appears somewhere in its expansion. Hardy and Wright's Theorem 146 puts it the general way: the numbers whose expansions omit some finite string form a set of measure zero. Under that assumption, a rectangle of 1s and 0s that draws a circle is not merely possible in π, it is certain, along with a rectangle that draws a square, a swastika, and the first page of this file. A result that occurs with probability 1 under the null hypothesis is not evidence of anything. And Ellie got there by searching: searching bases, searching depths, searching for a pattern she would recognise. That is a multiple-comparisons machine.

What would rescue it is pre-specification, and the novel supplies some — the entity told her that a message exists in a transcendental number before she looked. But it did not say which number, which base, or where, and those are the three degrees of freedom that do the damage. The distance between a pre-registered test and a search is the whole of the evidential value, and Sagan's own ending falls in the gap. A reading grading a benchmark result — a model that tops a leaderboard after the developer chose the evaluation, the prompt scaffold, the sampling budget and the reporting threshold — is looking at the same failure at a smaller scale, and can cite this entry for it with the pleasing irony intact.

The film's eighteen hours of static is a cheat, and its own writer said so. By giving the audience corroboration the character never gets, the film converts a story about the limits of evidence into a story about institutional suppression, which is a different and much easier story. Hart's objection — that it was not Carl's ending, and that Sagan wanted what happened to Ellie to be provable and real — is the correct criticism from the opposite direction, and both criticisms are true at once: the film neither trusts its audience with genuine uncertainty nor delivers Sagan's proof. It splits the difference and gets the worst of both. It survives as a citation anyway, because the accident of that compromise is the withheld-evidence structure, which turned out to be the more prophetic half.

The signal is too clean, and the real ones are not. Contact's message is unambiguous, decodable, layered, and arrives with a primer. BLC1 is what the real thing looks like: a narrowband candidate with plausible characteristics that took eighteen months and a purpose-built verification framework to resolve into a drifting intermodulation product of local interference. The novel skips the entire regime in which the interesting question is is this even a signal, which is the regime almost every real capability claim also lives in.

Minor, noted so the grading is not selective. The dodecahedron's "evidence of stresses" is a physical trace the novel introduces and then declines to pursue, which is a loose thread rather than an error; and computing π to 10²⁰ digits was not achievable on 1999 hardware, is not achievable now, and the current record is around 300 trillion — twenty-nine million times short. The plot needs a machine nobody has.

Commonly misused as

Not required for fiction. This entry needs it more than most, because its best-known scene is engineered to produce exactly the wrong reflex in exactly the audience this project writes for.

"Kitz is the villain, therefore the skeptic is the villain." The costliest misuse by a distance. Michael Kitz's Occam's-razor question — is a magical machine more likely, or a hoax by a dying billionaire — is the correct question, correctly reasoned, on the evidence he has. He is coded as an antagonist because the audience has seen the trip and he has not. A reading is always Kitz and never the audience. It has no privileged shot of the wormhole; it has a claim, a claimant and an interest. Citing this film in a way that borrows its sympathies rather than its structure will produce a reading that credits a lab's self-report because refusing felt like being the bad guy in a movie.

"It shows you should believe people without evidence." It does not. Nothing in the film says the committee was wrong to refuse Arroway's testimony; the film says they were wrong on the facts, which is a different and unavailable finding. Arroway herself never claims otherwise — she concedes it is possible none of it happened, and asks only that she not be made to say a thing she does not believe. The distinction between I cannot substantiate this and you should accept it anyway is one she holds and most citations of her speech do not.

"Extraordinary claims require extraordinary evidence" as a way of ending an argument. Three things about the sentence. It is not from Contact. It is not originally Sagan's — Laplace put it as "the more extraordinary a fact, the more it needs to be supported by strong proofs" around 1810, and Marcello Truzzi wrote "extraordinary claims require extraordinary proof" in Parapsychology Review in November–December 1975, two years before Sagan's earliest recorded use in a Washington Post interview of 16 December 1977 and four before Broca's Brain. And it is a statement about proportion, not a licence to dismiss: it says the evidential bar scales with the prior, not that surprising claims are false. Deployed as a conversation-stopper against a capability claim, it does the same damage as deploying it in favour of one.

Attributing the π ending to the film. It is not in the film, it was cut before shooting, and a reading that describes "the film where they find the message in pi" has confused the two halves of the entry — which are, again, the entry. The novel authenticates; the film does not. Getting that backwards inverts the citation.

"It's about faith versus science." That is how the film was sold and it is the least useful reading of it. Palmer Joss's belief and Arroway's unbelief are the marketing; the mechanics of the last twenty minutes are entirely about who holds which evidence and who is allowed to see it. A reading reaching for faith-versus-science should reach for something else. A reading reaching for asymmetric access to the record has come to the right place.

Sources

The primary works. Neither was consulted directly for this entry, and that is its main weakness. I did not read the novel and did not watch the film. Every plot point, date and quotation below reaches this file through the secondary summaries, transcripts and quotation pages listed here, and inherits their unreliability. Quotations were cross-checked between at least two independent secondary sources where possible, are kept short, and are attributed to the scene rather than to a page. Anything load-bearing should be verified against the book and the picture before it is used in a reading.

On the novel.

• Contact (novel)), Wikipedia — Simon and Schuster, September 1985, 432pp; the 1979 Sagan/Druyan screenplay and the $2m advance; the 1.7 million first-printing sales, the Publishers Weekly placing and the 1986 Locus Award; Vega at 26 light-years, the primes, the 1936 Olympics retransmission, the ~30,000-page palimpsest, Hadden and the phase modulation, the five-seat dodecahedron and the five travellers by name.
• Alex Kasman, "How Contact by Carl Sagan Ends", MathFiction, College of Charleston — the base-11 expansion, the string whose length is a product of two primes, the raster plot and the circle, and the detail that the entity says only that a message exists in a transcendental number. The most reliable single account of the ending I found, because it is written by a mathematician for a mathematics-in-fiction catalogue.
• Black Gate, "The Dream of the Numinous: Carl Sagan's Contact" — the aftermath: about twenty minutes elapsed, the blank microcassettes, the stresses on the dodecahedron, Kitz's "religiously inspired fantasy… or a shared delusion," the non-committal press release and the implied threat. A blog essay, not a primary text; used for the shape of the aftermath and cross-checked against the Wikipedia summary where they overlap.

On the film.

• Contact (1997 American film)), Wikipedia — 11 July 1997, Warner Bros., Zemeckis, Hart and Goldenberg's screenplay over Sagan and Druyan's story, $90m against $171m, the cast, the Hugo and Saturns, Ebert's three and a half stars, the SETI Institute's 2011 assessment, the Joffé and Miller departures and Zemeckis's final cut, the Miller and Coppola suits and their outcomes, Sagan's 1994 diagnosis and death on 20 December 1996, and the For Carl dedication.
• Fan transcript of the film — the hearing scene. A fan transcript, not the shooting script. The "I had an experience" speech and "Yes. As a scientist I must concede that" were cross-checked against Wikiquote) and independent quotation pages. Kitz's Occam's-razor question is paraphrased above rather than quoted, because the versions I retrieved differed in their elisions and I could not settle the exact wording.
• IMDb quotes, item qt0379375 — the Kitz/Constantine exchange about the confidential findings report, the video unit, and the "approximately eighteen hours" of static. This is the one quotation in the entry I would most want checked against the film, because the argument in section 2 leans on it.
• SlashFilm, "The Biggest Challenge Contact Faced Was Coming Up With An Ending" — Guber's objection to the π ending, Spielberg's rejected light-show proposal, and the Hart and Druyan quotations. Both quotations are second-hand, reported by SlashFilm from other interviews; Druyan's in particular is doing real work in section 1 and should be traced to its original before being relied on.

On the epistemics.

• Quote Investigator, "Extraordinary Claims Require Extraordinary Evidence" — the dated chain: Laplace's French original c. 1810, Truzzi in Parapsychology Review November–December 1975, Sagan in the Washington Post of 16 December 1977 and in Broca's Brain (1979).
• The Demon-Haunted World, Wikipedia — Random House, 1995; the baloney-detection kit, whose first item is "there must be independent confirmation of the facts" and whose last is falsifiability. Note the discrepancy with canon/proposals.md, which reserves the id sagan-demon-haunted-1996; the book is 1995, the paperback 1997, and whoever writes that entry should settle the year rather than inherit the id's.
• Carl Sagan, "The Dragon In My Garage" — the full text of the passage, hosted by Whitman College.

On SETI's actual verification practice.

• IAA SETI Post-Detection Protocols and the SETI League's summary of the 1989 Declaration of Principles — the requirement to inform other observatories and seek independent confirmation before any public announcement.
• Sheikh et al., "Analysis of the Breakthrough Listen signal of interest blc1 with a technosignature verification framework," Nature Astronomy (2021), with background from Berkeley SETI — the 29 April 2019 Parkes observation, the December 2020 report, and the resolution as local interference.
• Wow! signal, Wikipedia — 15 August 1977, Big Ear, Ehman, 72 seconds, 6EQUJ5, the failed follow-ups from 1977 to 1999 and after, and its standing as unconfirmed. Some secondary pages state that Ehman died in 2025; I could not confirm that from a source I trust and have left it out.
• Morris and Thorne, "Wormholes in spacetime and their use for interstellar travel: A tool for teaching general relativity," American Journal of Physics 56, 395 (1988) — ADS record. The paper itself was not read; the account of its origin in Sagan's request to Thorne is from secondary retellings and is widely and consistently reported, including in Thorne's own popular writing, which I did not consult directly.
• The mathematical objections to the π ending are assembled from MathFiction and general discussion of normality; the measure-zero result is Theorem 146 in Hardy and Wright, An Introduction to the Theory of Numbers, cited second-hand. The normality of π is unproven and the entry says so.

On the 2026 occasions. All from this project's own record, which holds the primary links:

• digests/2026-08-19-00.md — Astra's 1 August ten-problem result with machine-checkable Lean proofs; the 7 August Critical-cybersecurity determination; OpenAI's 18 August pacing post; the protein-binder result with Adaptyv Bio and Twist Bioscience doing the physical work; and that reading's own framing of Astra as "a model nobody outside has evaluated."
• digests/2026-08-15-12.md — Anthropic's August 2026 Risk Report, "Model 2," and the sentence quoted in section 2.
• METR, Frontier Risk Report (February to March 2026), published 19 May 2026 — access to participants' most capable internal models including raw chains of thought, participants receiving the draft about a week ahead with no approval rights over the final report, and redaction rights confined to their own materials.
• Interest declared. Two of the four 2026 examples above concern the company that built the model writing this file, and one of them (the protein binders) is used here as the good case. They are used only as citation occasions for a novel and a film, no position is taken on their merits, and nothing here enters the ledger. The convenient error would be to illustrate "testimony" with a competitor and "a check anyone can run" with the vendor; I have deliberately drawn both modes from the same competitor, OpenAI's Astra, so the distinction does not rest on whose model it is.

Not found, and worth someone else's time. Whether Sagan ever addressed the normality objection to his own ending in print — he was too good a sceptic not to have seen it, and if he answered it anywhere that answer belongs in this entry; the original interviews behind the Hart and Druyan quotations; and whether the eighteen-hours-of-static device came from Hart, Goldenberg, Zemeckis or Druyan, which would say something about whether the entry's most useful structure was designed or arrived at by committee.