The commitment, stated plainly
Most engineering conversations die in the gap between "we have a problem" and "we have something a contracting officer, a department head, or a proposal manager can act on." The problem gets described in a call. Somebody promises to write it up. Two weeks later the write-up arrives as a nine-page capabilities deck with no numbers in it, and the close date has moved inside a week. We took that failure mode out of our process. Send us a problem description and you get one page back within 48 hours.
One page is the whole format. It carries an objective written as a single testable sentence, an approach broken into named steps with a named target stack, a deliverable list with file types attached, acceptance criteria expressed as numbers on a named instrument, and a range with the labor mix and period of performance behind it. It also carries the assumptions that would move the number, because a range without its assumptions is a guess wearing a suit.
The page is free and it is not a sales artifact. If the honest answer is that the work does not fit us, the page says so in the first line and names what kind of shop does fit. A capture manager forty-eight hours from a proposal deadline is better served by a fast no than by a slow maybe, and we would rather be the firm that gives clean answers than the firm that keeps a bad fit alive for three weeks.
What is on the page
Objective. One sentence that a reader outside the technical conversation can verify later. "Extract eleven named fields from scanned inspection reports at 98% field-level precision on a held-out set of 500 documents" is an objective. "Modernize the inspection workflow with AI" is a slogan, and slogans cannot be accepted or rejected at the end of a period of performance.
Approach. Three to five steps, each with the method named and the target environment named. Not "we will apply machine learning" but the actual shape: baseline on a rules extractor, fine-tune a document model, calibrate a confidence threshold, route low-confidence pages to a human queue, deploy to the customer's existing tenancy. The point is that a technical reader can argue with it. An approach nobody can argue with says nothing.
Deliverables. Artifacts with formats. A source repository with a tagged release, a container image, an evaluation notebook carrying the held-out results, a data dictionary, a test report, and the runbook an operations team needs to keep the thing alive after we leave. Every deliverable on our page is something the customer can open without us in the room.
Acceptance criteria. The numbers, the dataset they are measured on, and who declares pass. This is the line that decides whether an engagement ends cleanly.
Range and contract shape. A low and a high, the labor categories behind them, the period of performance, and the contract type the work fits. Firm-fixed-price under FAR 16.202 when the deliverable is definable at the start. Time-and-materials under FAR 16.601 when the discovery genuinely dominates the first weeks and neither side can price it honestly yet.
Assumptions and swing factors. Data access dates, whether an environment already has an authorization to operate, whether the records carry Controlled Unclassified Information, whether a government-furnished dataset shows up on time. Each one gets a line and a direction: this assumption breaking moves the estimate up, this one moves it down.
The acceptance criteria line is the one that matters
Everything else on the page is describable. Acceptance criteria are where scoping either becomes an engineering document or stays a brochure. Our page never accepts a criterion that cannot be measured by a third party. "Improves analyst efficiency" fails. "Reduces median triage time per case from 14 minutes to under 6 minutes, measured on the 200-case regression set, verified by the government technical lead" passes.
Three properties make a criterion usable. It names a metric with a direction and a threshold. It names the instrument the metric is measured on, which is almost always a frozen held-out dataset that neither side can tune against. And it names the person with authority to declare pass or fail. Criteria that skip the third property produce the worst kind of dispute, where both parties agree on the number and disagree about whether the number counts.
We also write the failure branch into the criteria. If the threshold is missed, what happens? Usually the answer is a defined remediation window with the labor already priced in, or a documented finding that the approach does not reach the target on the available data, which is a legitimate research outcome and should be treated as one. Contracts that pretend failure is impossible are the ones that end in a dispute over an unpaid invoice.
Why 48 hours is possible
The turnaround is not a promise about how fast we type. It is a consequence of what is already standing when your email arrives. We keep a bench of named engineers, licensed professional engineers, and domain specialists across defense, health, energy, transportation, and public-sector data, so the staffing question is a phone call rather than a search. We keep current labor categories and indirect rates, so a range does not require a fresh cost build-up. And we keep reference architectures for the problem shapes that recur across federal, state, and commercial work.
Where a 48-hour scope lands cleanly, by request type
Editorial weighting of how often a first-pass scope survives contact with the real environment. Illustrative, not a measured statistic.
The reference architectures matter more than anything else on that list. Across twenty years of production systems built for federal agencies at five consulting firms, three of them federal, the same handful of problem shapes keep reappearing under different names: structured extraction from unstructured records, entity resolution across systems nobody controls end to end, anomaly detection on operational time series, retrieval over a controlled document corpus, computer vision on inspection imagery, and migration of an analytic workload into a regulated cloud region. When a new problem arrives, the first question is which shape it is, and that question usually resolves in an hour.
Modeling feasibility is the second fast call. Our practice is led by a former professor in technology who ranks in the top 200 of more than 200,000 on Kaggle, and that record produces exactly the judgment a 48-hour estimate needs: whether the signal is plausibly in the data, what accuracy band is achievable before anyone builds anything, and where the estimate should widen because the data is thinner than the description suggests. Deployment target is the third, and with seven cloud certifications on the team the environment question is a lookup rather than an investigation.
The five things we need from you
The clock starts when the message contains enough to scope against. Five items get us there, and a paragraph each is plenty. Nobody needs to write a requirements document to get a one-page scope back.
- The problem in plain language, and who feels it today. What breaks, how often, and what it costs when it does.
- The data: what exists, where it lives, roughly how much, who owns it. "About 40,000 PDFs in a SharePoint library, going back to 2011" is enough to start.
- The decision the output supports and the person who makes it. Every useful system ends at a human doing something differently.
- The date that actually matters. A solicitation close date, a fiscal-year obligation deadline, a demonstration, a board review.
- The constraint set. Classification or CUI status, cloud region, on-premises requirements, whether an authorization to operate already exists, and any teaming or set-aside structure already in place.
What we do not need is a finished statement of work. If you already have one we will scope against it and flag the parts that will be hard to accept at the end. If you do not, the one-pager is often the thing your contracting shop turns into the statement of work, and that is a fine outcome. Several of the strongest engagements we have shaped started as three paragraphs pasted into an email at ten at night.
What "range" means, and the contract shape behind it
The range on the page is a low and a high, and the spread is information. A tight spread means the problem is well understood and the data is available; a wide spread means something specific is unknown and we name it. We do not compress a spread to look decisive. A number that looks confident and is wrong costs both sides more than a range that looks honest and holds.
| Situation | Contract shape we recommend | Why |
|---|---|---|
| Deliverable definable at the start | Firm-fixed-price, FAR 16.202 | Acceptance criteria carry the risk. Best value for the buyer when the target is measurable up front. |
| Discovery dominates the first weeks | Short T&M discovery under FAR 16.601, then FFP | Two to four weeks of paid discovery converts an unpriceable problem into a priceable one. |
| Subcontract workshare on a federal bid | FFP task with a defined technical volume contribution | The prime needs a number for the cost volume and a scope paragraph that survives evaluation. |
| Commercial buyer, product-shaped need | FFP milestones under a commercial-item structure, FAR Part 12 analogue | Milestone payment tied to accepted artifacts keeps both sides aligned without hourly accounting. |
| State or local pass-through of federal funds | FFP with 2 CFR 200 cost documentation | Uniform Guidance follows the money. Building the documentation in from day one avoids a painful audit later. |
The federal overlays we apply without being asked
A scope that ignores the compliance surface is not a scope. When a description mentions defense data, the page assumes DFARS 252.204-7012 applies and that the environment must meet NIST SP 800-171, and it prices the handling accordingly. When the records are Controlled Unclassified Information under 32 CFR Part 2002, the page names the CUI category and the marking and disposition obligations that come with it, because those obligations shape where the data can be processed and therefore what the work costs.
Data rights get a line every time. On defense work the relevant clauses are DFARS 252.227-7013 for technical data and 252.227-7014 for noncommercial computer software, and the difference between delivering with unlimited rights and delivering with government purpose rights is a real difference in both price and future position. We state which rights assertion the scope assumes so nobody discovers it at delivery. Precision Federal holds JCP / DD-2345 certification under CAGE 1AYQ0, so export-controlled technical data is a normal part of the intake rather than a reason to stall.
On SBIR and STTR workshare the arithmetic comes first. SBIR Phase I requires the small business to perform at least two-thirds of the work, and Phase II at least one half, so any subcontract we propose is checked against that ceiling before a dollar figure appears. STTR carries a different split, with the small business performing not less than 40% and the research institution not less than 30% under the SBIR/STTR Policy Directive, and eligibility rules live at 13 CFR 121.702. A prime bringing us in on a larger competitive procurement should also know that FAR 15.305(a)(2)(ii) lets a source selection consider the past performance of subcontractors that will perform major elements of the work, and that FAR 52.219-9 subcontracting plans, required above the threshold in FAR 19.702, create a documented reason to name a small business partner early rather than late.
The clock, hour by hour
Intake to signed scope
Step two is the step people underestimate. A single clarifying question sent within a few hours often changes the entire shape of the scope, and it is far cheaper to ask it on day one than to deliver a well-written page that solves the wrong problem. The question is usually about the data, sometimes about who signs off, and occasionally about whether the deadline in the message is the real deadline.
Five senders, five different pages
A prime capture manager gets a page written for a cost volume: workshare percentage, labor categories mapped to the prime's own structure, technical paragraphs that drop into a volume with light editing, and a statement of what we need from the prime by what date to hold the close. Speed matters more than polish here, and the page is built to be pasted.
A university researcher looking for a small-business partner on an STTR gets a page organized around the split: which tasks sit on the small-business side, which sit with the research institution, how the intellectual property lines are drawn, and what the budget shape is on both sides. Our engineers have shaped these partnerships with faculty across several disciplines, and the fastest ones start with the workshare table rather than the science.
A system integrator needing an AI bench gets a page about capacity and interfaces: who we assign, what they own, how the code lands in your repository, and what happens at the end so nothing is stranded. A commercial company gets outcomes and milestone payments. A state or local vendor gets the funding source, the documentation trail, and any in-state presence question that affects an evaluation.
When the answer is no
Roughly one request in five gets a no, and we send it fast. The usual reasons are a problem that is hardware-led rather than software-led, a timeline that cannot be met without doing worse work than the customer deserves, or a buyer who needs staff augmentation and would be better served by a staffing firm than by an engineering subcontractor. In every no we name what kind of firm fits, and when we know one, we say who.
That discipline is part of why the 48-hour number holds. A shop that says yes to everything eventually says "we will get back to you next week" to everyone.
Bottom line
The gap between a described problem and a scoped one is where most good engineering work goes to die. Closing that gap in two business days is an operational choice: keep the bench standing, keep the rates current, keep the reference architectures written down, and read every inbound message with an engineer rather than routing it through a sales funnel. One page that a technical lead and a contracting officer can both act on beats a deck nobody finishes.
Common questions before you send anything
Do I need an NDA in place first?
Not to get a scope. Send what you can share without one and we will scope against it. If the problem cannot be described without protected material, say so and we will execute a mutual NDA first; that adds a day or two to the clock and we will tell you the revised date up front. For export-controlled technical data our JCP / DD-2345 certification is already in place.
Is the one page binding?
The range is a good-faith estimate, not a quote, and the page says so. When the assumptions listed on the page hold, our final proposal lands inside the range. When an assumption breaks, we show which one and what it moved. Nothing becomes binding until there is a signed statement of work or a subcontract.
What if my deadline is tighter than 48 hours?
Say the deadline in the subject line and we will work to it. Same-day turnarounds happen regularly for teaming decisions on a closing solicitation. A same-day page carries a wider range and fewer assumptions resolved, and it will say which parts are provisional.
Will you scope work you would not bid on yourselves?
Yes. Primes, integrators, and universities regularly ask us to scope a workshare that sits inside their own bid. We write the page for their document, not ours, and we do not use the intake as a route to a competing position on the same opportunity.
Frequently asked questions
The problem in plain language, what data exists and where it lives, the decision the output supports, the date that matters, and any classification or cloud constraints. Five short paragraphs is more than enough. No deck, no requirements document, no formal request for information.
The one page is free. If the problem needs real discovery before anyone can price it honestly, we propose a short paid discovery task, usually two to four weeks under a time-and-materials structure per FAR 16.601, and its output is a firm-fixed-price scope for the build.
A named metric with a threshold and direction, a frozen held-out dataset neither party can tune against, and a named individual with authority to declare pass or fail. Criteria missing any of the three tend to produce disputes at closeout even when the technical work was sound.
Yes. We treat CUI under 32 CFR Part 2002 and the DFARS 252.204-7012 safeguarding obligations as ordinary scope inputs, and we price the handling environment into the estimate. Export-controlled technical data is covered by our JCP / DD-2345 certification under CAGE 1AYQ0.
Usually the page becomes the technical exhibit to a statement of work. We deliver a full statement of work within about a week of the page on request, with the deliverable schedule, acceptance procedure, data rights assertion, and payment milestones expanded from the summary lines.