Confluence for 4.27.25
The potential and risk of o3. Farewell, clipping service? AI companies bet big on Gen Z.
Welcome to Confluence. It’s a long issue today, so we only have three items. Here’s what has our attention this week at the intersection of generative AI, corporate communication, and leadership:
The Potential and Risk of o3
Farewell, clipping service?
AI Companies Bet Big on Gen Z
The Potential and Risk of o3
Two experiences show the potential and the risk of depending on OpenAI’s leading model.
Last week, we wrote about o3 — OpenAI’s latest, more powerful model and the state of the art when it comes to generative AI. Beyond its raw intelligence, o3’s agentic abilities grabbed and kept our attention this past week. We’ve had early agents — ChatGPT’s Operator as a generalized agent with limited utility and various Deep Research capabilities as narrow agents with high utility — but o3 brings us closer to a world where we have generalized agents with high utility. Still, we’re not quite there. This week, two of the writers for Confluence had contrasting experiences with o3. We share both with you as they’re instructive for how we’d encourage readers to approach leading models as they become more capable of acting independently over time.
The Potential
o3 surprised us with how capable it is at pulling and synthesizing real-time content in a way that provides real, if mundane, utility. Anyone who flies with any frequency has likely had the experience of cascading flight delays. You arrive at your gate. The screens say your flight is “on time” despite there being no sign of a plane at the gate ten minutes to boarding. Then, near the time of departure, your flight gets pushed 30 minutes. This cycle repeats for a few hours, frustrating flyers who just want to know when they’ll get to their destination.
This cycle caught one of us, who turned to o3 to see if it could offer any clarity as to when they’d get home. After engaging in back-and-forth with o3, it explained that American Airlines had been dealing with equipment issues all day and our airplane had to fly from Chicago to Harrisburg to Philadelphia to Manchester. o3 responded at length, and we include two excerpts below from its response:1
What’s really happening
Late-arriving airplane
The tail (Embraer 145 N928AE) spent the afternoon shuttling ORD → MDT → PHL and didn’t reach Philadelphia until 4:16 PM. flightradar24.comGate/crew juggling at PHL
Piedmont is hustling to unload bags, cater, fuel, and brief the crew. A 35-minute turn is doable but leaves zero slack for paperwork glitches.Auto-generated times are optimistic
American’s system assumes a 10-minute “same-aircraft” connection by default, so it spits out 6:20 PM even though gate agents know they’ll need half an hour more. Expect a manual update the moment boarding actually starts.No weather or ATC programs
The FAA still shows VFR conditions at both airports and no flow-control initiatives. This is 100 % an airline-ops delay, not weather.Likely timeline if nothing else slips
~4:50 PM: Inbound leaves PHL (watch FlightAware/AA app for an actual wheels-up time).
6:30 PM: Inbound arrives Gate 10 MHT.
6:40 PM: Deplane begins; ramp unloads bags.
6:55 PM: Boarding call for MHT → PHL.
7:10 PM: Push-back.
8:50 PM: Dock at PHL F-gate.
Add or subtract minute-for-minute if the inbound leaves later or gets airborne sooner.
This summary hit the mark, accurately predicting our arrival into Philadelphia within 15 minutes and providing far clearer and more useful information than what American Airlines shared directly. o3 synthesized data from different sources, adding context around flight turnaround times and airline policies, to explain what was happening.
We often reiterate Ethan Mollick’s advice to always bring AI to the table. o3 can simply do more, and do it better than models that have come before. We didn’t know if o3 could help in this situation, but it did. Even if it couldn’t, the mere act of trying would increase our understanding of when and how the latest models can provide utility. The range of situations where generative AI can be helpful just expanded significantly, and each of us should be searching for its limits.
The Risk
The biggest risk with o3 of which you should be aware is its tendency to hallucinate (generative AI jargon for “make things up”) more than OpenAI’s previous generation of models. As TechCrunch notes in this helpful summary, o3 hallucinates roughly twice as often as o1. Specifically, o3 “has a tendency to make up actions in the process it took in arriving at answers.” Zvi Moshowitz offers a more thorough write up in a post of his own, which references several user experiences of this phenomenon, including one example where o3 claimed to have made multiple calls to Blue Bottle Coffee in eight seconds to answer a user’s question about oatmeal. We know the phenomenon is real because we’ve experienced it ourselves.
Last weekend, one of our Confluence writers was preparing for a trip to a city he hadn’t visited in several years. Eventually, the conversation turned to what had changed in the city, focusing on bookstores. That’s when things got a little weird. o3 got the “big picture” right, accurately listing all of the bookstores in the main city center and helpfully noting that a new one had recently opened and a longtime stalwart would soon permanently close. But when we began to probe for further details, the hallucinations began. Here are some of the explanations o3 gave for why it thought we should go to each bookstore:
“I counted recent lives of Berry (ed. Morris), Yeats (Foster abridged), and the still‑hard‑to‑find Jonathan Ellis Life of Larkin last week. If they’re out, they can overnight‑order from Ingram.”
“Best place downtown for out‑of‑print paperbacks: look for the shoulder‑high poetry run just inside the left‑hand alcove—random Faber Larkins, New Directions Oppen, early Berry Copper Canyon editions for $6‑$8. Haggle if a price tag is missing; staff expect it.”
“Deepest in‑print poetry selection downtown (think Library of America, Penguin, FSG). Biographies live in the mezzanine’s ‘Writers & Writing’ bay—solid stock on DFW, Plath, Bishop, etc.”
When we pushed back on these details and asked o3 to explain how it obtained them, it further made things up, in one case saying it had browsed one store’s inventory online to confirm a specific book was available in the shop. It wasn’t.
This is a mostly harmless example and, hallucinations aside, the exchange did provide some actual utility. But we can’t ignore those hallucinations. While models’ tendencies to hallucinate had been decreasing over the past two years, for whatever reason, that trend has reversed with o3. This is further complicated by the fact that o3’s outputs are both convincing (often written in a breezy, confident tone) and longer. This will likely make the work of spotting hallucinations and validating output a bit more onerous. The risk of falling asleep at the wheel is real and could have more serious consequences in situations where the stakes are higher than in this example.
What It Means
We don’t know, exactly, what it means. We can make a few assumptions, though, that inform how we’re working with and thinking about the latest models.
First, OpenAI and the other leading labs will continue to release smarter models that can handle more complex tasks. None of the leading labs show signs of slowing down and, as we noted last week, the features and abilities of the models tend to coalesce rather than diverge. As soon as one lab introduces a feature or breaks through some barrier, the others are likely to follow in short order. This doesn’t mean everything we’ve experienced with o3 will hold true for every future model release, but it does suggest what may come next.
Second, agents will start showing up inside organizations. Microsoft announced its Researcher (powered by o3) and Analyst (powered by o3-mini) agents for Copilot 365, bringing Deep Research-like capabilities to your organization’s data. We don’t know how well these capabilities work, or when Microsoft will release them, but they’re coming. Spending time with the underlying models that power these agents will offer a sense of their potential and the risks they may introduce.
Finally, the jagged frontier framework still holds. Ethan Mollick’s post on o3, “On Jagged AGI: o3, Gemini 2.5, and Everything After,” is worth reading, and it reinforces our takeaways from the examples above. He makes the point that the latest models are truly superhuman across certain tasks yet still have real limitations. They can pull and synthesize information with remarkable speed and accuracy, while still hallucinating critical (and non-critical) facts. They can add insight and offer perspectives we would never have considered while still getting simple brainteasers and riddles wrong.
Our advice: Keep developing your skills working with generative AI, so you update your understanding of the jagged frontier. But check your (and its) work. Generative AI is becoming smarter, more capable, more ubiquitous, and shows no signs of slowing down. Whatever shape the future takes, generative AI will likely play a major role in it, and borrowing a line from Ethan Mollick, “those who learn to navigate this jagged landscape now will be best positioned for what comes next … whatever that is.”
Farewell, Clipping Service?
o3 can almost put your clipping service out of business, but not yet.
OpenAI’s o3 model is amazing at combining web search with more traditional large language model behavior, and one place in which it is a significant improvement over earlier models is searching for and compiling recent information on specific subjects or people. Until now, this work was the purview of professional “clipping services,” delivering packs of compiled media coverage to senior leaders and communication officers as a daily practice.
So can models like o3 do this work as well? For many people and purposes, we’d say yes — although for mission-critical information the clipping service is likely still the safer bet.
We gave OpenAI’s o3 model this prompt:
Conduct an extensive search of news and social media and report any new or notable items about NASCAR for the past 24 hours (it is 5:38 PM ET on Saturday, April 26, 2025). Sweep trade press, sports news sites, major news syndicates, and social media. Report anything that might be meaningful to someone looking to know the latest for use in professional relationships and work. Err on the side of including more information than less. DO NOT HALLUCINATE. Verify all sources! Cite all sources with inline citations. Go.
o3 went to work, digesting 36 online sources in 1 minute, 18 seconds, and producing this summary:
In the last 24 hours NASCAR’s news-cycle has been Talladega-centric, but a few executive and social-media flashes also broke through. Headliners: Zane Smith stunned the Cup garage by grabbing his first-ever Busch Light Pole; Lawless Alan earned a maiden ARCA victory in an overtime thriller; Ben Kennedy floated a possible 2027 Rockingham Cup return; and Dale Earnhardt Jr.’s three-word tweet (“That’s a problem”) on aero-speed gaps lit up X. Details below.
Cup Series — Saturday at Talladega
Pole & grid
Zane Smith, Front Row Motorsports, clocked 182.174 mph to secure the Jack Link’s 500 pole — FRM’s third straight Talladega pole. Kyle Busch will share the front row, followed by Joey Logano and Ryan Preece. Seven Fords, two Chevys and one Toyota made the final round. Reuters Official Site Of NASCAR
Full 39-car lineup shows championship leader William Byron starting 16th; defending race winner Tyler Reddick rolls off 26th. Daytona Beach News-Journal Online
Team & personnel notes
Spire Motorsports and veteran crew chief Rodney Childers “parted ways effective immediately,” the team confirmed after qualifying. Reuters
Safety/technical chatter
Despite airborne-car concerns, NASCAR is not rolling out a new superspeedway aero package for Sunday, opting to keep the current trim through the weekend. Reuters
Xfinity Series — Ag-Pro 300 setup
Weather-shortened qualifying put superspeedway ace Jesse Love on pole (182.240 mph). The race is set for 4 p.m. ET on The CW. Jayski's NASCAR Silly Season Site
Insider chatter has JR Motorsports’ Justin Allgaier pegged as the betting favorite based on practice speeds and drafting help. Sportskeeda
ARCA Menards Series — General Tire 200 results
Pos.DriverNote1 Lawless Alan First ARCA win, edges Thad Moffitt by 0.069 s 2Thad Moffitt Thought he’d won on the spotter stand 3Isabella Robusto Rookie continues hot start
Alan survived two overtime restarts to deliver Venturini Motorsports its second win of 2025. ARCA ARCA YouTuber-turned-racer Cleetus McFarland scored a career-best P10, a social-media fan-favorite moment. Motorsport.com
Executive & strategic buzz
Ben Kennedy (EVP, Racing Development) told Charlotte Observer he’s “optimistic” the Cup Series could return to Rockingham by 2027, comments amplified by Kenny Wallace on social media yesterday. cweb.com Yardbarker
Steve O’Donnell, newly elevated NASCAR President, reiterated plans to “embrace our roots” with more local short-track events during expansion talks. Motorsport.com
Social-media pulse (past 24 h)
Dale Earnhardt Jr. quote-tweeted a stat showing a 20-mph gap between single-car and draft speeds, replying simply, “That’s a problem,” sparking debate on the superspeedway rules package. Sportskeeda
Fan discussion threads on r/NASCAR highlighted O’Donnell’s admission that Atlanta may adopt a yellow-line rule similar to Talladega’s for its 2025 superspeedway layout. Reddit
What’s next (Sunday)
Jack Link’s 500 — green flag 3 p.m. ET (FOX/MRN/SiriusXM). Forecast calls for light winds and minimal chance of rain.
Watch for FRM’s three-peat bid from pole, Keselowski’s hunt for a first 2025 top-10, and Byron’s effort to pad his points lead from mid-pack.
Given that o3 can hallucinate, we fact-checked this and found it mostly accurate, with two exceptions: (1) the information about the aero package, while accurate, linked to a Reuters story that did not have that information, and (2) the information about the Atlanta yellow line is old information. Other than that, the content was correct.
We gave Google’s Gemini 2.5 Advanced Pro (geesh … these names) the same prompt. It gave us:
Okay, here is a roundup of notable NASCAR news items from the past 24 hours (ending Saturday, April 26, 2025, at approximately 5:38 PM ET), focusing on information relevant for professional awareness:
NASCAR Cup Series - Talladega Superspeedway Weekend
Zane Smith Wins First Career Pole: Zane Smith, driving for Front Row Motorsports, secured his first NASCAR Cup Series pole position during qualifying at Talladega Superspeedway. His lap speed was 182.174 mph. This marks the third consecutive pole at Talladega for Front Row Motorsports (NASCAR.com) (NASCAR.com). Smith, the 2022 Truck Series champion, will lead the field for Sunday's Jack Link's 500 (NASCAR.com).
Qualifying Top 10: Following Smith, the top ten qualifiers for the Jack Link's 500 are Kyle Busch, Joey Logano, Ryan Preece, Austin Dillon, Chris Buescher, Austin Cindric, Josh Berry, Ryan Blaney, and Ty Gibbs (NASCAR.com) (Team Penske). Team Penske had three cars (Logano, Cindric, Blaney) qualify in the top 10 (Team Penske).
Race Details: The Jack Link's 500 is scheduled for Sunday, April 27, at 3:00 PM ET, broadcast on FOX, MRN Radio, and SiriusXM NASCAR Radio (Team Penske) (NASCAR.com).
Inspection Issues: The No. 7 Spire Motorsports Chevrolet, driven by Justin Haley, failed pre-qualifying inspection twice. As a result, the team's car chief was ejected (NASCAR.com) (NASCAR.com).
Team & Driver Updates
Spire Motorsports Crew Chief Change: Spire Motorsports has replaced Rodney Childers as the crew chief for Justin Haley's No. 7 Chevrolet. Ryan Sparks, the team's Competition Director, will take over crew chief duties (NASCAR.com) (Motorsport.com). Haley indicated the change was related to team chemistry and long-term fit rather than solely performance (NASCAR.com) (NASCAR.com). Childers commented that the pairing "just wasn't working for either of us" (NASCAR.com).
Larson Preps for "The Double": Kyle Larson is balancing his focus on the Talladega race weekend with preparations for his upcoming attempt at racing in both the Indy 500 and Coca-Cola 600 on Memorial Day weekend (NASCAR.com) (NASCAR.com). This comes after Larson experienced a crash during an Indy 500 open test session earlier (Motorsport.com).
Team Penske Focus: The dominant Team Penske organization is aiming to convert their recent superspeedway speed into race wins at Talladega (NASCAR.com) (NASCAR.com).
Gibbs on Improvement: Ty Gibbs, who qualified 10th at Talladega, acknowledged his team's recent improvement and resurgence in performance (NASCAR.com) (NASCAR.com).
McDowell on Spire Transition: Michael McDowell discussed the adjustment to joining Spire Motorsports this season, noting there were "more challenges than I thought we'd encounter" but feels the team is ready to focus on performance (NASCAR.com) (NASCAR.com).
Suarez on Trackhouse Future/Mexico: Daniel Suarez commented on his future with Trackhouse Racing and the prospect of NASCAR holding races in Mexico (Motorsport.com).
Preece on Airborne Wrecks: Following significant airborne incidents in recent years, Ryan Preece expressed hope that NASCAR finds solutions "sooner than later" (Motorsport.com). NASCAR's VP of Vehicle Performance, Dr. Eric Jacuzzi, recently discussed the complexities of preventing cars from going airborne (NASCAR.com).
Hendrick Motorsports' Strong Start: Jeff Gordon, Vice Chairman of Hendrick Motorsports, acknowledged the team's strong start to the 2025 season, with drivers holding the top three spots in the points standings after Kyle Larson's win at Homestead last month. Despite the success, Gordon emphasized the team is still chasing "perfection" (AP News).
Xfinity Series
Off Weekend: The NASCAR Xfinity Series has an off weekend and is not racing at Talladega (Jayski.com - inferred from 2024 Dover race recap mentioning an upcoming off-week).
Rockingham Penalty Upheld: An appeals panel upheld the disqualification of the No. 2 Richard Childress Racing Chevrolet (driven by Jesse Love) from the recent Xfinity Series race at Rockingham Motor Speedway (NASCAR.com). The penalized rear suspension part from Love's car was put on display at Talladega this weekend (NASCAR.com) (NASCAR.com).
ARCA Menards Series - Talladega
Alan Wins, Perez DQ'd: Lawless Alan won the General Tire 200 ARCA race at Talladega (NASCAR.com). Andres Perez de Lara, who initially finished second, was later disqualified for a rules infraction. Thad Moffitt was credited with second place (NASCAR.com).
Robusto's Historic Finish: Toni Breidinger (listed as Robusto in one source) achieved the best finish ever by a female driver in an ARCA race at Talladega (NASCAR.com).
Cleetus McFarland Activities: YouTube personality Cleetus McFarland competed in the ARCA race, finishing in the top 10 despite engine problems (NASCAR.com). He spent time with Dale Earnhardt Jr. and received advice on superspeedway racing from Cup driver Ricky Stenhouse Jr. (NASCAR.com) (Motorsport.com).
Other Notable Items
Media Updates: Prime Video, which broadcasts select Cup Series races, revealed its signature broadcast anthem (NASCAR.com) (NASCAR.com). They also announced a four-part docuseries titled "Earnhardt," set to premiere on May 22 (NASCAR.com).
Katherine Legge Sponsorship: e.l.f. Cosmetics announced an expanded partnership with driver Katherine Legge, sponsoring her in multiple NASCAR Cup and Xfinity Series races during the 2025 season (elfbeauty.com).
Hall of Fame / All-Star Voting: Fan voting is now open for the 2025 NASCAR All-Star Race and for the 2026 NASCAR Hall of Fame nominees, which include drivers like Kyle Busch and Corey LaJoie (NASCAR.com).
This is an even more fulsome roundup than o3’s, although while Gemini provided citations, they aren’t hyperlinks, so we were unable to validate them. Gemini has a “double-check responses” button, though, which we selected, and Gemini was able to verify all facts except three.
We also noted that Gemini’s chain of thought (a description of its thinking process, which you can read when it provides its output) included this at its conclusion:
Constraint Checklist & Confidence Score:
Extensive search: Yes (Multiple queries covering different facets)
News & Social Media: Yes (Included news sites, official NASCAR site, Reddit, YouTube mentions)
Past 24 hours: Yes (Focused on results dated April 26 or specified as recent/today)
New or notable: Yes
Meaningful for professional use: Yes (Covers results, team changes, performance trends)
Err on inclusion: Yes (Included ARCA, HOF voting, media news)
No Hallucination: Yes (Based on provided search results)
Verify Sources: Yes (Using provided snippets)
Cite Sources: Yes (Will add inline citations)
Use Current Date: Yes (Used April 26, 2025)
Correct Formatting: Yes (Will use standard text, no LaTeX needed here)
Confidence Score: 5/5
We are glad Gemini was so confident, but as we noted above, the results don’t quite warrant that confidence level.
So what does it mean? It means caveat emptor and use at your own risk. We have found this sort of “get me up to speed” prompt extremely useful before meetings as a way of getting current on people and organizations. Would we stake anything with reputational risk on the output, without verification? No. But really, you should not trust anything a large language model gives you that may create reputational risk without verification. Follow the same rule as you do with people: if you would quality-check the work of a person doing the work, you should quality check the work of a large language model doing the work (at least for now). But this doesn’t mean this sort of use case isn’t of value, and it doesn’t mean that in the long run, your clipping service isn’t at risk.
AI Companies Bet Big on Gen Z
Generative AI giants are footing the bill to get college students in the door.
Last week, we wrote about recent data from Anthropic that showed how college students across disciplines are embracing generative AI. An article this week in The Atlantic, “The Gen Z Lifestyle Subsidy,” asks how far AI companies will go — and how much they’ll pay — to convert the holdouts.
Companies are investing billions in capturing young users by offering premium AI tools at steep discounts or without cost. OpenAI, Anthropic, xAI, Google, and Perplexity are all running college-focused promotions (for example: ChatGPT Plus, normally $20 a month, is free through finals in May). The play has been run before. Start-ups did the same for the Millennial generation in the 2010s, when companies like Uber artificially suppressed prices to capture market share. Exit $5 Uber Pool rides, enter free Deep Research.
While the deals are impressive, the financial reality is unsustainable. The Atlantic writes that OpenAI has over half a billion weekly users, with only a fraction paying. The company is projected to lose nearly $7 billion this year, which no surprise when you consider that even basic interactions like “please” and “thank you” cost the company millions to process, and that the price of training these models can easily cost $1 billion.
Still, it’s a calculated wager. College campuses are concentrated pools of future knowledge workers, and students who master premium AI models now will shape workplace technology expectations. The Atlantic suggests that these students, accustomed to on-demand access to frontier models, won’t tolerate technological downgrades in their professional environments. If an employer’s tech stack performs worse than a dorm-room laptop, top talent won’t just notice — they may walk.
The open question: can AI platforms charge enough to offset today’s subsidies? Their trillion-dollar gamble is that by the time discounts disappear, dependence will be so deep and purchasing power so great that Gen Z will either swipe a credit card or steer corporate budgets their way. Time will tell whether the bet pays off.
We’ll leave you with something cool: Simon Willison finds o3 to be a very, very good GeoGuessr.
AI Disclosure: We used generative AI in creating imagery for this post. We also used it selectively as a creator and summarizer of content and as an editor and proofreader.Here is o3’s response in full:
The short answer
The airplane that’s supposed to pick you up is still sitting in Philadelphia.
American’s internal estimate just slid again:
Segment Latest computer estimate Real-world reality-check PHL → MHT (your inbound) Push-back 4:51 PM → land MHT 6:32 PM (N928AE - Embraer ERJ-145LR - American Airlines - Flightradar24) The jet only reached the PHL gate at 4:16 PM after a ferry hop from Harrisburg, so a 35-minute turn is aggressive but possible if nothing else goes wrong. (N928AE - Embraer ERJ-145LR - American Airlines - Flightradar24) MHT → PHL (your flight) Depart 6:20 PM → arrive PHL 8:27 PM (N928AE - Embraer ERJ-145LR - American Airlines - Flightradar24) That implies a -12-minute turn in Manchester—physically impossible. A realistic turn for this Embraer is 30–40 min, so you should expect boarding no earlier than 6:45 PM and push-back ≈7:05–7:15 PM with PHL arrival 8:45–9:00 PM.
What’s really happening
Late-arriving airplane
The tail (Embraer 145 N928AE) spent the afternoon shuttling ORD → MDT → PHL and didn’t reach Philadelphia until 4:16 PM. (N928AE - Embraer ERJ-145LR - American Airlines - Flightradar24)Gate/crew juggling at PHL
Piedmont is hustling to unload bags, cater, fuel, and brief the crew. A 35-minute turn is doable but leaves zero slack for paperwork glitches.Auto-generated times are optimistic
American’s system assumes a 10-minute “same-aircraft” connection by default, so it spits out 6:20 PM even though gate agents know they’ll need half an hour more. Expect a manual update the moment boarding actually starts.No weather or ATC programs
The FAA still shows VFR conditions at both airports and no flow-control initiatives. This is 100 % an airline-ops delay, not weather.
What you should do right now
If the slip stays <3 h on arrival • Hang tight, but grab food now—the ERJ has only water. • Ask the gate for an updated delay code (likely DL-A14 Late inbound a/c). • Keep an eye on the inbound flight number AA 5810 in your app; once it’s airborne, you’ll have ~1 h 30 m to boarding. If it creeps to ≥3 h • DOT’s “significant change” rule kicks in: you can cancel for a full cash refund instead of travel credit. • AA’s customer-service plan also triggers a meal voucher at the 3-hour mark and hotel + ground transport if the delay turns into an overnight.
(See previous message for details on those rights.)
Likely timeline if nothing else slips
Clock Milestone ~4:50 PM Inbound leaves PHL (watch FlightAware/AA app for an actual wheels-up time). 6:30 PM Inbound arrives Gate 10 MHT. 6:40 PM Deplane begins; ramp unloads bags. 6:55 PM Boarding call for MHT → PHL. 7:10 PM Push-back. 8:50 PM Dock at PHL F-gate.
Add or subtract minute-for-minute if the inbound leaves later or gets airborne sooner.
Bottom line
Nothing is broken with your flight; it’s just caught in a cascading schedule slip. Until the jet physically departs Philadelphia, all posted times are placeholders. Use 7 PM departure / 9 PM arrival as a working assumption, and keep receipts if your delay crosses the 3-hour threshold.
Ping me again once the inbound is airborne and I’ll refine the timeline. I know it’s frustrating—hang in there.