Final week — staff and sign-wave assignments

Rendered 2026-08-19 18:05 ET from FINAL_WEEK_plan.md. Do not edit this page by hand — edit the markdown and re-run render_plan.py.

Rebuilt Wed Aug 12 2026, evening, on the corrected ballot model

Supersedes the Aug 11 version entirely. Three things changed underneath it and all three move the plan:

1. The turnout projection is rebuilt on every pipeline run — currently 83,014, band 72,498–90,964. Election Day is 49% of the electorate (40,267 ballots), not the 43–52% guess. Read live values from deploy/site_context.json; do not quote this file for numbers.
2. The baseline was wrong. Pace was measured against 2024 ballot counts, which assumed registration held still. It did not — R registration is up 27% since 2018 and Leon is the only county that LOST Republicans. Counties that read "well behind" against 2024 are often at or above their own historical rate. Use the vs-rate column, not vs-2024.
3. Leon is surging and it is the biggest remaining pool in the district.

Amended Sun Aug 16 — Election Day numbers corrected

Every Election Day figure in this plan was hand-typed against the pre-Aug-12 projection and had drifted about 10% low. They are now tokens filled live from site_context.json and precinct_ed.json at render time, so they cannot go stale again. Bay's Election Day was written as 8,090; it is 8,985. Leon was 7,324; it is 8,135. Jackson was 3,232; it is 3,590. No conclusion in this plan changes — the PC-vs-Leon efficiency ratio is still 2.50x and the allocation still holds. See predict/leon_market_math.py and FREEZE.md Amendment 11.

Two things in §1 and §2 are now history, deliberately left in place. Early voting closed Aug 15 everywhere except Leon, Taylor and Gadsden, which closed Aug 16. So the "Leon has 2,821 early votes still to come" framing, the Saturday and Sunday candidate decisions, and the 1.49x tripwire reading are all spent — read them as the record of a call already made, not as live instruction. The tripwire has since crossed to 1.67x, and it did not change the allocation: Leon's net banked vote is 0.956x its 2024 figure, because its early-vote surge is mail voters walking in, not new turnout.

Also note §1, §2 and §3 still carry the July-wave market shares (22% Bay, 8.8% Leon) in the early-vote passages. The current model is 28.0% / 11.2%, which is what every Election Day figure below now uses.

1. The single most important number in this plan

Leon has 2,821 early votes still to come. Bay has 2,607. Leon is now the largest remaining early-vote pool in CD2, and its tripwire sits at 1.49x against a 1.50 threshold, direction +0.33 since D-10.

Leon's share of the electorate has moved from 20.7% in 2024 to a projected 23.3%. Rogers polls ~8.8% there. Evan Power is second district-wide at 15%, he is RPOF chairman, and Leon is his home county. The Leon surge is most plausibly Power consolidating Tallahassee.

This does not threaten the win — Rogers projects 28,059 against a p90 bar of 16,852, a cushion of +11,207 — but it is the only trend running against him and it is accelerating rather than flattening.

2. Should the candidate go to Leon Saturday and Sunday?

Sunday yes. Saturday no. The two days are not the same decision.

Saturday Aug 15 — NOT Leon

Saturday is the last early-voting day in 13 of 16 counties. Bay has 2,607 EV ballots left that day and never gets another. Leon votes again Sunday; Bay does not. Spending Saturday in Leon trades a market where Rogers gets 22% for one where he gets 8.8%, on the one day the good market is still open.

Sunday Aug 16 — YES, Leon

Only Leon, Taylor and Gadsden run early voting Sunday. Everything else is dark. Sunday-eligible EV totals 3,348 ballots and Leon is 2,821 of it — 84%.

The old plan said "Sunday — Taylor, and only Taylor." That is now wrong: Taylor has 309 EV ballots left and Gadsden 218. Together they do not justify a candidate day. Send one staffer to Perry and one to Quincy; the candidate goes to Tallahassee.

The constraint that makes Leon safe — read this twice

Do NOT run GOTV in Leon. Not sign waves reading "GO VOTE", not "polls are open", not turnout messaging of any kind. Leon breaks 8.8% Rogers. Every marginal Leon voter you turn out is net negative for him — you would be building Power's margin with your own volunteers.

The only safe activity in Leon is persuasion of people already in line. They are voting regardless; you are changing the ballot, not creating it. Contrast on Power, delivered to a captive audience, costs you nothing and can only gain.

Sizing it honestly: a 4-point shift across 2,821 Leon ballots is about 113 votes. That is not why you win. It is worth Sunday only because there is nothing else open on Sunday — the alternative use of that day is doors in Bay for Tuesday, which is also defensible and which I would take if the candidate is tired. Do not pretend Sunday in Leon is a big play. It is a free option on a dead day.

3. Where the plan actually gets won: Bay on Election Day

Bay's Election Day is three times either early-vote play. Election Day is 49% of the whole electorate and the PC DMA is 55% of it. Both early-vote rows are now history — early voting closed Aug 15 outside Leon, Taylor and Gadsden — so Bay's Election Day is the only live prize in this table.

Bay is also the largest ballot shortfall in the district at −4,233, and on the registration-adjusted read it is 0.82x, so that is genuine underperformance rather than a baseline artifact. Same for Holmes at 0.78x.

Everything below is built around getting Bay and Jackson to the polls Tuesday.

4. Day by day

Thu Aug 13 — doors in Jackson and Calhoun

The behind-and-big precincts are concentrated here. Jackson 3, 7, 11, 9, 1, 6, 12 and Calhoun 301, 101, 202 are all running 0.43–0.62x with 250+ estimated R ballots each. These are persuadable-turnout precincts where the vote has not happened yet.

Fri Aug 14 — Bay doors, shift to EV sites after 4pm

Last full weekday before EV closes. Also the hard deadline for anything that must ship if Trump endorses Monday — signage and door hangers especially, since they do not depend on a live send window.

Sat Aug 15 — Bay and Jackson EV sites, all day, all hands

The last early-voting day in 13 counties. 2,607 Bay ballots and 348 Jackson ballots are still available and will never be available again. This is the highest-value single day left. Do not dilute it.

Sun Aug 16 — Leon, contrast only

Candidate to Tallahassee. One staffer to Perry (Taylor), one to Quincy (Gadsden). No GOTV messaging in Leon — see §2. Contrast on Power to voters already queued. Everywhere else in the district is dark; use the day for Tuesday prep.

Mon Aug 17 — no early voting anywhere

The only day of the cycle with no early voting in any county, and 40,267 ballots — 49% of the whole electorate — are still uncast. Every one of them is reachable today. None is reachable Tuesday morning.

Field: doors and phones in Bay and Jackson.

The candidate — radio and volunteers, NOT doors. A candidate knocking doors on the last afternoon reaches perhaps fifty households. Drive-time radio in the Panama City market reaches thousands of Bay commuters the night before they vote, and Bay's Election Day alone is 8,985 ballots at 28.0% — about 2,516 votes in play in one county. WMBB / WJHG Monday evening news is the same trade.

• Afternoon: PC-market radio and TV. This is the last mass touch that exists.
• 5–6 PM: volunteer launch for Tuesday. His presence multiplies here in a way it cannot on a doorstep — a hundred volunteers leaving energised is worth more Tuesday than fifty doors Monday.
• From 6:45 PM: in Bay County, camera-ready, reachable. The Trump eve-of-election window is ~7:00 PM CT, confirmed twice this cycle. If it lands the playbook triggers and the clock is about twelve hours. Being in Tallahassee or in a car at that moment costs the news cycle it would win.
• Not Leon. He should not be in Tallahassee at all today.

Whoever owns the endorsement trigger must be named and reachable by 5 PM.

Tue Aug 18 — Election Day

The structural fact that shapes the day: the district votes in two zones, and his own zone votes an hour later.

So at 6 PM CT, when Leon has stopped voting, Bay still has an hour of polls open. Work it right through. Do not drive anywhere at closing time.

Election Day volume, projected: Bay 8,985 · Leon 8,135 · Jackson 3,590 · Wakulla 2,882 · Walton 2,681 · Washington 2,382 · Holmes 1,935.

Bay all day, or start in Leon and work back?

Stay west all day. Do not start in Leon. The math is not close, and Leon is worse than merely low-value.

Leon has nearly as many Election Day ballots as Bay, which is what makes it tempting. He gets 11.2% of them. Every 1,000 voters moved in the PC market is worth 2.50x what the same 1,000 are worth in Leon. Total at stake: Bay 2,516 Rogers votes, Jackson 1,005, Leon 911.

Why Leon is actively wrong, not just inefficient. A candidate at a precinct is turnout-stimulating. In Bay, where he runs 28% and leads, the marginal voter he pulls in helps him. In Leon, where he runs 11.2% and Power — the actual runner-up — is on home ground, the marginal voter he pulls in helps Power. Working Leon on Election Day means spending his own body to raise turnout in his opponent's best county. That is the same logic that governed the Sunday call, except Sunday had an excuse: nothing else was open. Tuesday everything is open.

The schedule cost is worse than the ratio suggests. Tallahassee to Panama City is roughly 100 miles, about 1h45m. A Leon-first day costs Bay's 7–9 AM rush (~1,498 ballots at 28.0%), two hours in a car returning nothing, and his own vote with a camera — he votes in Bay, and it is the one newsworthy candidate act of the day. It buys three Leon hours worth ~228 Rogers votes against the ~419 given up in Bay. Six hours spent to gain three, in a market worth 40% as much.

The strongest counter-argument: Leon is 8,135 ballots and 22.6% of the electorate — you cannot concede it. He is not conceding it; he contested it Sunday. There is nothing he can do on Tuesday that changes Leon's outcome except raise its turnout, which cuts against him. You beat a Leon surge by out-running it in Bay and Jackson, not by standing in it.

Jackson is the honest middle. Marianna is about an hour each way, the same 28% market, and Jackson held seven of the fourteen "big and behind" precincts in §5. A midday Jackson block also answers the district-wide optics without setting foot in an 11% market. Pure ballot math says Bay all day and skip Jackson too; keep Jackson live because a Panama-City-only Election Day has a real cost of its own.

What would change this: if Bay is visibly running hot by midday and Jackson visibly cold, shift the afternoon to Jackson. Nothing in the returns can justify going east — by the time Leon's numbers tell you anything, its polls have closed.

Morning, 6–9 AM CT - Vote first, with a camera. The one genuinely newsworthy thing a candidate does today, and it costs twenty minutes. - PC-market morning drive radio. Bay's Election Day voters are in cars 6:30–8:30. - The 7–9 AM rush at two or three of the highest-volume Bay precincts. No more than three — travel time is the enemy.

Midday–3 PM — Bay, then Jackson. Between them they carry the two biggest Election Day blocks in the district, and both are ~22% markets. Nothing else competes. He is not persuading anyone at this point; he is feeding volunteers and thanking poll workers, and the honest reason to do it is that a field team that sees the candidate at noon is still working at six.

150 feet from the entrance to the polling room per Fla. Stat. 102.031(4). Operate at 200 to leave margin.

3 PM–7 PM CT — the afternoon rush is the second peak and it runs to close in the Central counties. Highest-volume Bay precincts, until 7 PM CT. Not at headquarters, not in a car.

The evening — tell him this MONDAY, not Tuesday.

Shares are percent of ballots cast, which is what the board reports. Poll shares (with undecided in the denominator) run about a third lower; judging Tuesday against those would read a ten-point shortfall as on track.

He is going to appear to be losing for a full hour, and it is the forecast, not a collapse. Leon reports first and he polls 11.2% there; Leon alone is 44% of the 7 PM dump. If nobody has warned him, at 7:20 PM he will believe he has lost and will say things he cannot unsay.

The single number that matters all night is Bay's first report at 8:00 PM ET. Near 28% and it is over. At 22% or below the model is wrong and everything after is genuinely live. The worry line at 7:15 is below 14% — third place itself is expected.

Best advice for 6–8 PM: don't watch. Be with the volunteers who just worked a fourteen-hour day. The returns arrive whether he refreshes or not, and there is no decision left to make.

5. Precinct targets

Doors and turnout — biggest R volume that is ALSO behind

Jackson is seven of the top fourteen. That is where the recoverable vote is.

Washington 4, 7 and 9 — the Friday Aug 14 doors

Asked Washington or Calhoun on Aug 14, the answer is Washington, and it is not close. Washington has 2,382 Election Day ballots left against Calhoun's 1,205, and a Very Likely + Likely pool of 4,318 against 1,871 — 2.3x the gettable universe.

More important, the two are moving in opposite directions once registration is controlled for. Washington is at 0.94x its own historical rate; Calhoun is at 1.13x. On the raw 2024 column Calhoun looks worse (0.77x vs 0.84x) and that is the trap — Calhoun's registered Republicans are up 1.92x since 2018, so it is not underperforming, it is simply a much bigger county than 2024 implies. Knocking Calhoun means chasing voters who are already turning out.

Precinct 4 is the whole county. Its Very Likely + Likely pool of 1,156 is larger than Calhoun's three biggest precincts combined (829). Its front-load index has risen to 1.12 as the early vote came in, so some of it has now been pulled forward — but that is still under the 1.15 "banking ahead" line, and it remains far and away the largest Election Day block in the county.

Two teams: one stays in 4 all day, the second takes 7 and 9. Twelve and five have volume but are banking ahead of schedule — those voters are largely handled. Skip 6, 11 and 1 — 111 Election Day ballots between them for a lot of driving.

If something forces a Calhoun trip anyway, 301 is the only precinct worth it (196 ED, index 0.87, most under-banked in that county). It is a third of Washington 4, and Calhoun is Central-time like Washington so there is no timing advantage to offset the volume gap.

Washington's precinct join is 100%, so unlike Bay these figures are solid at precinct level.

A better basis, added Aug 14

The table above is from precinct_pace.json, which allocates by the all-party county feed. precinct_ed.json now supersedes it for Election Day work: it allocates each county's Republican Election Day total by that precinct's share of its county's Republican 2024 primary turnout, taken from the voter file. That basis sums to 83,713 district-wide, exactly the model's BASE_2024.

It matters. The all-party proxy was sizing Walton 530 at 178 Election Day ballots when its real Republican volume is 554, and Bay 4 at 592 when it is 256. Where the two disagree, trust precinct_ed.json.

It also carries front_load_index per precinct — above 1.15 the precinct is banking ahead of its own 2024 share and has less left for Tuesday; below 0.85 it is holding back and has more. That is a diagnostic, deliberately not folded into the projection: the feed's banked figure is all-party, and subtracting it from a Republican total sent 49 precincts negative.

Signs — biggest R volume, regardless of pace

Bay 15 (1,489 est. R), Bay 20 (1,342), Bay 17 (1,190), Bay 18 (994), Bay 16 (890), Bay 9 (853), Washington 4 (786), Bay 10 (755).

These are already running 1.08–1.41x, so they do not need turnout help — they need name recognition at the moment of voting. Signs, not doors.

Washington 4 is the exception at 0.64x: big and behind. It gets both.

DO NOT target on these numbers

Bay 19, 6, 11, 4, 25 and Walton 420, 430, 520. These precincts have a join_quality of suspect — Bay, Leon and Walton redrew precinct lines between 2024 and 2026, and one Walton precinct shows 9.0x the eligible voters it had. Their pace figures are built from two different pieces of geography. Coverage is 0.85 in Bay and 0.93 in Walton. Use county-level numbers there, not precinct.

6. Statutory limit — get this right

150 feet from the entrance to the polling room is the no-solicitation boundary, per Fla. Stat. 102.031(4). Operate at 200 feet to leave margin. An earlier version of this plan cited 200 feet as the statute; that was wrong.

7. What the candidate should be doing

• Thu–Sat: Bay and Jackson, in that order of volume. Sites Saturday.
• Sun: Tallahassee, contrast only, no turnout messaging.
• Mon: PC-market radio and TV in the afternoon, volunteer launch 5–6 PM, in Bay and camera-ready from 6:45 PM for the endorsement window. Not doors.
• Tue: vote early with a camera, morning drive radio, then Bay and Jackson precincts through 7 PM CT. Do not watch the 7:15 PM ET returns.

The two things he must be told before Tuesday, not during it:

1. He will read as third until 8 PM ET. Leon closes an hour before Bay and he polls 11.2% there. This is in the model; it is not a surprise and not a collapse.
2. Bay's 8 PM ET report is the whole night. Near 28% and it is over. At 22% or below, the model is wrong.

The local endorsement drip is worth more than another out-of-state congressman — Graceville Mayor Terry Nichols (Jackson) and Bay Commissioner-Elect Scott Duvall are the two most useful names on the list, and Duvall's endorsement came with a real hurricane-rebuilding issue cue rather than generic résumé language. Use them in the counties they come from.

8. What would change this plan

• Leon crosses 1.50x — it will not change the allocation. It argues for more Bay, not for resources to Tallahassee. A recovered Bay voter is worth 2.5x a recovered Leon voter on margin.
• Trump endorses Power — the single largest downside risk in the race. Everything above is void; go to the playbook.
• The Thursday curve refit applies — current read is that it should not. The apparent trend is opening-day noise; the last three days are flat.