Almost Counts in Nothing But Horseshoes and Hand Grenades

The quantum probability field flickered at 99.97%—which, in any reasonable universe (one where effort correlated with outcome, where the laws of physics acknowledged the concept of "trying really hard"), would be close enough. But Marissa Chen had been doing this job for six years—Temporal Adjustment Specialist II, Grade 7, benefits-eligible—and she knew that the universe was not, in fact, reasonable, and that 99.97% meant absolutely nothing when the threshold was 99.98%.

Almost is never enough.

The phrase appeared on every rejection notice her department issued, printed in Helvetica Neue Light (corporate branding initiative, Q3 2023) beneath the Department of Chronological Integrity's logo, which looked like a Möbius strip having an anxiety attack. She'd submitted a formal complaint about the phrase once—it was technically redundant, she'd argued in a carefully worded email using the approved inter-departmental communication template, because "almost" definitionally meant "not enough"—but HR had responded with a link to a mandatory webinar on "Embracing Organizational Messaging" and a reminder that her performance review was coming up.

She refreshed the probability calculator—an Excel spreadsheet so byzantine it had achieved sentience sometime around 2019 and now occasionally made passive-aggressive comments in the formula bar—and watched the number tick down to 99.96%. Great. Excellent. Fantastic.

The thing about preventing timeline collapses (which was what her actual job entailed, beneath the HR-approved euphemism "temporal adjustment" and the corporate mission statement about "maintaining chronological brand integrity") was that you needed perfect conditions. Not great conditions. Not "pretty damn good" conditions. Not "honestly-this-should-work-we've-done-everything-right" conditions. Perfect conditions. 99.98% probability of success, minimum, as mandated by the International Temporal Accord of 2027 (revised 2031, amended 2034, that one clause about lunch breaks still under dispute).

She'd been working this particular case for seven months. Subject: Marcus Webb, age 34, whose decision to buy a lottery ticket at a Circle K in Henderson, Nevada, at 10:47 PM on August 3rd would—through a cascade of consequences so elaborate she'd needed three whiteboards and a bottle of Blanton's to map them all—eventually result in the complete destabilization of the North American timeline by 2047. Nothing dramatic, just your standard butterfly-effect catastrophe: lottery win leads to investment in tech startup, startup develops slightly-too-effective AI, AI optimizes itself into an existential threat, humanity spends the 2040s having a very bad time, etc.

Simple fix, theoretically: adjust the probability field so Marcus's tire goes flat before he reaches the Circle K, or his shift at the warehouse runs late, or literally anything that results in him not buying that specific ticket at that specific time. She'd done this kind of work dozens of times. It was, as her supervisor Dale liked to say (usually while eating a tuna sandwich at his desk with the door open, filling the entire temporal monitoring suite with the smell of mayonnaise and existential dread), "bread and butter stuff."

Except Marcus Webb, for reasons that suggested the universe had a sense of humor and that sense of humor was cruel, existed in a probability nexus—a point where multiple timeline branches converged with nearly identical outcomes. Adjust one variable, another compensates. Push here, it bulges there. Classic conservation of chronological momentum. Physics being a real bitch about the whole "actions have consequences" thing.

She'd tried thirty-seven different intervention scenarios. Flat tire (99.94%). Late shift (99.95%). Sudden onset food poisoning (99.96%—though that one had felt mean-spirited even by her standards, and the Ethical Adjustment Committee had kicked it back with a note about "maintaining professional compassion in intervention design," which was rich coming from people who'd approved a scenario last month that gave someone's cat diabetes).

The current scenario involved adjusting Marcus's medication delivery schedule (he took Lexapro, 10mg, mail-order through CVS Caremark) so he'd be home waiting for the package instead of stopping by the Circle K for Red Bull and a Slim Jim and, fatefully, a Powerball ticket. She'd spent three weeks modeling the probability cascades. Everything looked good. Should work. Would work, in any universe where 99.97% wasn't functionally identical to zero.

Her phone—a department-issued iPhone 13 with a cracked screen protector she'd been too exhausted to replace—buzzed. Text from Dale: need that Webb case closed by EOD, budget review tomorrow, looking bad if we're running long on simple interventions

She typed back: Close as in successful or close as in documented failure?

Dale: successful obv

Dale: wait is there a problem

Dale: thought this was straightforward

She looked at the screen. 99.97%. She could submit it. Technically, per Section 47.3(b) of the Temporal Adjustment Procedures Manual, she had discretionary authority to approve interventions at 99.95% or above in cases of "demonstrable implementation urgency."

Budget review counted as urgency, right?

She imagined the timeline: Marcus doesn't buy the ticket, doesn't win, doesn't invest, doesn't inadvertently trigger the AI catastrophe that turns the 2040s into a cyberpunk dystopia minus the cool aesthetic. 99.97% chance of success. But also: 0.03% chance of failure. 0.03% chance that something—quantum fluctuation, probability eddy, the universe's general vindictiveness—goes wrong and Marcus buys that ticket anyway (or something worse happens, because timeline adjustments didn't fail gracefully; they failed like a Jenga tower, sudden and catastrophic and usually involving temporal paradoxes that required weeks of paperwork).

Her supervisor at her last review (performance: "Meets Expectations," which was corpo-speak for "you're doing fine, but we're not giving you a raise") had noted that Marissa "sometimes displays excessive caution regarding acceptable risk tolerances."

Translation: she gave too much of a shit.

But here's the thing about almost: it's a word that works in horseshoes and hand grenades (and nuclear war, technically, though that particular aphorism felt darkly appropriate given the stakes). In temporal mechanics, almost meant nothing. Either you prevented the timeline collapse or you didn't. Either Marcus Webb stayed home on August 3rd, 2025, or he didn't. Ninety-nine-point-nine-seven percent might as well be zero.

She could hear Dale's voice already: You had approval authority. Why didn't you execute? Do you know what our success rate looks like this quarter?

She could also hear, less distinctly but more insistently, the voice of every probability theory textbook she'd ever read, every senior specialist who'd trained her, every cautionary tale about adjusters who'd cut corners: The threshold exists for a reason.

99.97%.

So close.

And yet.

She opened a new modeling scenario—the thirty-eighth—and started adjusting variables. Maybe if Marcus's CVS delivery got delayed and his usual convenience store was closed for health inspection and his friend texted him at exactly 10:42 PM about that new season of The Bear... Maybe she could stack enough micro-adjustments to tip the probability over into an acceptable range.

Her phone buzzed again. Dale: ?

She didn't respond. She was busy doing the job—not the job as described in the department handbook, but the actual job: preventing disaster through meticulous, obsessive care, fighting against the universe's fundamental indifference one decimal point at a time, existing in the exhausting space between what should work and what definitely will.

The new scenario initialized. Probability field loaded. The number appeared: 99.98%.

Exactly 99.98%.

Threshold achieved. Disaster prevented. Timeline secure. Humanity saved from its own lottery-winning, AI-building stupidity. Marcus Webb would stay home on August 3rd, 2025, annoyed about his late medication but ultimately fine, while Marissa Chen submitted her intervention report and moved on to the next case.

Almost had never been enough. But this? This was exactly enough. This was the razor's edge between existence and catastrophe, measured to two decimal places and approved per departmental guidelines.

She submitted the intervention, closed her laptop, and thought about getting a drink after work. Something strong enough to forget that the difference between saving the world and watching it burn was 0.01%—a margin so thin it was basically philosophical, a cosmic joke about precision and paranoia and the exhausting necessity of getting it exactly right.

Her phone buzzed one last time. Dale: nvm saw the approval, nice work, this is why we keep you around

She stared at the message—the casual acknowledgment, the lowercase letters, the suggestion that this (her seventh month of obsessive modeling, her refusal to settle for 99.97%, her willingness to look like the overcautious neurotic rather than take an acceptable risk) was just "nice work"—and felt the familiar weight of doing a job that mattered intensely to no one except her and the timeline itself.

Almost is never enough. She'd known that six years ago when she took this position. She knew it now. And she'd know it tomorrow when the next case landed on her desk, another butterfly wing that needed to be precisely, carefully, exhaustedly prevented from flapping.

The universe demanded perfection. She delivered it. That was the job.

Close only counted in horseshoes, hand grenades, and the fantasies of people who'd never had to watch a timeline collapse because they'd rounded up.