What changed
FACT (source: USAspending award ASST_NON_4671DRPRP00000001_070): DHS obligated $1,802,989,401.24 to the 'GOVERNOR'S AUTHORIZED REPRESENTATIVE' in Puerto Rico to 'PROVIDE REIMBURSEMENT TO STATE, LOCAL, TRIBAL, AND TERRITORIAL GOVERNMENT ENTITIES AND CERTAIN PRIVATE NON-PROFIT ORGANIZATIONS FOR EMERGENCY PROTECTIVE MEASURES TAKEN DURING THE PANDEMIC.' FACT (same source family): two additional DHS awards to the same Puerto Rico GAR and to the Government of the Virgin Islands total $35.3B (ASST_NON_4339DRPRP) and $22.0B (ASST_NON_4340DRVIP) for 'REPAIR OR REPLACEMENT OF DISASTER DAMAGED FACILITIES'. INFERENCE (not in source text): the money is a pass-through — the GAR/COR3 holds the award, and the actual filers are the 78 municipios, public corporations and private non-profits who must document costs to get paid. HYPOTHESIS: nothing in the award text states a deadline, a portal, a filer count, or that any subrecipient lacks tooling. Those are the load-bearing assumptions of this idea and none of them are sourced.
Why now
HYPOTHESIS, weakly supported. The award is real and large (FACT). But the trigger here is an award *record*, not a new mandate. DR-4671 is a pandemic-period declaration; emergency-protective-measures work under it was performed years ago, and obligated COVID-era Public Assistance awards are predominantly in closeout, appeal and audit posture rather than initial Project Worksheet formulation. If that is right — and the input contains no evidence either way — then the 'why now' is not 'a wave of new filings is about to start' but 'a backlog of closeouts and de-obligation disputes is being worked'. That is a different, smaller, harder product. The engine's own lesson that 'a stated award amount is HARD demand evidence' is being applied correctly to the existence of money, and incorrectly to the existence of *future filings*. Money already obligated is money whose paperwork has largely already been filed.
Converging signals
Three signals are cited but only one is independent. (1) The $1.8B DR-4671 PR award — the primary source. (2) 20 'FUNDED MANDATE' items retrieved at cosine 0.72-0.76, of which 15 are Medicaid entitlement grants (T19) to state health departments and one is an ACA §1332 waiver — these have no relationship to FEMA Public Assistance filing and are retrieval noise, matching on the shared surface pattern 'large dollar figure + federal agency + state recipient'. (3) The two other DHS disaster awards (PR 4339, VI 4340) are genuinely adjacent and do support the 'territorial disaster recovery is enormous' claim. Net: the convergence is one award plus two siblings, dressed up by seventeen irrelevant Medicaid rows. I am flagging this rather than scoring it as breadth.
Customer pain
HYPOTHESIS — zero direct evidence in the input. The demand_evidence array contains no PAIN item, no HIRING/SPEND item, and no complaint or job posting from a municipio, a PNP, or a recovery consultant. Per the founder profile I am instructed not to require chatter to validate a filing obligation, and I don't — but the obligation itself must be *shown*, and here the specific filing burden (Requests for Public Assistance, Project Worksheets, RFRs, quarterly progress reports) appears only in the convergence_description, which is the system's own prior narration, not the source. The FEMA PA documentation burden is real and well known outside this input; I am asserting it as background knowledge, not as something these sources establish.
Who pays
The plausible payers, ranked: (a) private non-profits with PA-eligible costs — small hospitals, shelters, food banks, churches — who have no grants department and cannot afford a 3-5% recovery consultant; (b) the recovery consultants themselves, as a white-label document-assembly tool; (c) the 78 municipios. (c) is the one the convergence names and it is the worst of the three: each municipio is a separate government purchase, in Spanish, requiring a local relationship, subject to municipal procurement rules and a mayoral election cycle. Selling to 78 governments one at a time is enterprise sales with the word 'enterprise' removed. The founder profile explicitly permits selling to subrecipients rather than procurement offices — but a municipio *is* a procurement office. Only (a) and (b) escape that.
Solved today
By consultants. Puerto Rico's recovery has been run for years by large program-management firms under COR3 contracts, and municipios engage their own PA consultants who bill hourly or as a percentage of recovery. HYPOTHESIS (not sourced in input): the incumbents include national disaster-recovery consultancies with existing COR3 relationships and Spanish-language staff. There is also FEMA's own Grants Portal, which is free to subrecipients and is where Project Worksheets actually live.
Why current solutions are bad
Consultants are expensive and slow, and the founder profile is right that undercutting a percentage fee with software is a legitimate wedge. But the honest counter: a percentage-of-recovery consultant is not selling data entry, they are selling *eligibility judgment* — deciding which costs FEMA will accept, defending them at appeal, and absorbing the blame when FEMA de-obligates. A document assembler does not carry that risk and therefore cannot capture that fee. The part of the consultant's bill that is pure assembly is the part they'd happily automate themselves.
Proposed product
Bilingual FEMA PA packet assembler: subrecipient uploads payroll registers, invoices, force-account equipment logs and contracts in Spanish; the system extracts, categorises against FEMA cost categories (A-G, plus Category B emergency protective measures), flags missing documentation against the FEMA PA Program and Policy Guide checklist, and emits an English-language, FEMA-format Project Worksheet narrative and cost-documentation binder ready for upload to Grants Portal. Note this is an *assembler*, not a submitter. The founder's ELDT edge was submitting into a federal portal; FEMA Grants Portal has no public submission API and its accounts are provisioned per-subrecipient with role-based access. He would be preparing documents a human uploads, which is a materially weaker product than the ELDT precedent and should not be scored as if it were the same shape.
MVP version
Take one real Category B pandemic-costs closeout package for one Puerto Rico PNP, obtained by working the file by hand for free. Build the Spanish-invoice/payroll extraction and the FEMA-format narrative generator around that single case. Ship nothing until one packet has actually been accepted.
30-day build
Do not build. Spend 30 days establishing whether the premise is true: (1) Confirm DR-4671's closeout posture — how many PWs remain unformulated vs in closeout, obtainable from FEMA's PA program data and COR3's Transparency Portal. (2) Interview five parties: two municipal finance directors, two PNP administrators, one PA consultant. Ask what they pay today and for what. (3) Determine whether Grants Portal permits any programmatic submission or bulk upload. If (1) shows the filings are done and (3) shows no API, this idea is dead as scoped and should be redirected to open declarations with active PW formulation.
60-day build
Only if 30-day findings hold: build the extraction + narrative generator against the single hand-worked packet. Recruit two more PNPs as design partners at zero cost. Target: three packets assembled, at least one accepted by COR3/FEMA review without material rework.
90-day revenue plan
Convert the design partners to paid at a flat per-packet fee. Approach two PA consultants with a white-label offer priced per packet — they are the only channel that can absorb 50 filings without 50 sales cycles. Realistic first revenue is a four-figure services-shaped invoice, not SaaS ARR. Treating this as a 90-day revenue play requires believing the discovery phase confirms rather than kills it, and I would put that below even odds.
Distribution path
This is the fatal weakness and the reason I would not lead with the Puerto Rico variant. There is no self-serve channel. The buyers are 78 municipal governments, an unknown number of PNPs with no shared directory, and a handful of consultancies. The founder sells through demonstrated value, not relationship sales — but relationship sales in Spanish, to Puerto Rican municipal government, from the mainland, is precisely what this requires. The convergence calls bilingual capability a moat; it is more accurately a barrier the founder must pay a contractor to cross, and one every incumbent consultant already crossed years ago.
Pricing hypothesis
$1,500-4,000 per assembled packet is defensible against consultant rates. Percentage-of-recovery pricing should be avoided: it implies eligibility risk the founder cannot underwrite and edges toward contingency-fee territory. Per-seat SaaS does not fit — a subrecipient files a handful of times in a disaster's life.
Technical difficulty
Moderate and honestly assessed: Spanish OCR/extraction over inconsistent municipal invoice and payroll formats is genuinely hard, and errors are expensive because a bad cost line triggers de-obligation and audit exposure for the customer. The FEMA narrative generation is the easy half. Nothing here is beyond a capitalised solo founder with contractors, but the failure mode is quiet wrong output, which is the worst kind in a compliance product.
Legal / regulatory risk
Low-to-moderate. Preparing federal grant documentation for a subrecipient is not a licensed activity, and the founder does not become a certified party — so this is not heavy_compliance in the founder's sense. But 2 CFR 200 subrecipient rules, PR government contracting rules, and the fact that the tool's output feeds federally audited claims mean errors have real downstream consequences and warrant an indemnity-limiting contract from day one.
Platform dependency
Correctly not a platform-policy risk — FEMA cannot deplatform a document assembler. But there is a real dependency of a different kind: if Grants Portal never exposes programmatic submission, the product is permanently a document generator sitting outside the system of record, which caps both price and stickiness. And COR3, as the pass-through entity, can standardise its own subrecipient tooling and eliminate the market by fiat.
Founder fit
Good but overstated by the convergence. Public records, operational documentation, and government paperwork are squarely in his strengths, and the ELDT app proves he can monetise a federal filing obligation. What does not transfer: ELDT had thousands of small discretionary buyers, a self-serve signup, an actual submission API, and English. This has ~78-200 buyers, no self-serve path, no submission API, and a Spanish-language sales motion. Same category, different business. I score founder_fit 6, not the 9 the 'public money' thesis would reflexively assign — the thesis is about shape, and the shape here is missing its two most valuable parts.
Breakout potential
Modest as scoped. The generalisable asset is the FEMA PA cost-documentation engine, not the Puerto Rico wrapper. If it works, the expansion is to every open major disaster declaration on the mainland — larger filer counts, English, no municipal-procurement gate. Puerto Rico is being framed as the beachhead because the dollar figure is biggest; but beachheads should be chosen for ease of landing, not size of prize, and by that test Puerto Rico is the worst entry point available, not the best.
Final recommendation
KILL AS SCOPED — but run the seven-day check before discarding the underlying engine. The Puerto Rico framing fails on three independent grounds, any one of which is sufficient: the filings are most likely already made, the buyer is municipal procurement, and the submission layer that made ELDT work does not exist here. The convergence's own claim that PR is 'the highest-value beachhead' inverts the correct test — it selected for prize size and against landability. What survives is narrower and worth a week, not a quarter: a FEMA Public Assistance cost-documentation engine, aimed at *currently open* declarations with active Project Worksheet formulation, sold to private non-profits and white-labelled to consultants, in English, on the mainland. Before writing any code, verify (1) whether any major declaration has meaningful unformulated PW volume, and (2) whether Grants Portal exposes any programmatic submission. If (2) is no, the whole family of ideas drops to a services business and should be judged as one. I want to be explicit that I am recommending against the system's primary thesis on this instance: the thesis says score public-money forced-filer shapes as maximal, and the shape matched — but a matched shape is not a matched business, and this one is missing the forced filer, the deadline, and the portal.
Next action
Spend two hours, not two weeks: pull FEMA's Public Assistance program data for DR-4671 and check how many Project Worksheets remain in formulation versus closeout. If the answer is near zero, mark this convergence dead and log the lesson that an 'award' trigger on an obligated multi-year disaster grant indicates completed paperwork, not pending paperwork — the engine is currently reading a receipt as an invoice.