PageBound

1. One-liner

Checks every transcript against the rules of the jurisdiction it is actually going to, before you certify it.

2. Trend signal — why now?

Three things moved at once.

The people making the record are being replaced. AAERT’s 2025 Court Reporting Industry Trends Report (released 10 April 2025) puts the stenographer workforce down 21% over the decade to roughly 23,000, with stenography school enrollment down 74% and nearly half of programs closed. BLS shows the occupation at about 17,700 jobs with essentially flat projected employment 2024–2034 and ~1,700 openings a year — almost all backfill, not growth. Read those together and the story is not “fewer people take the record.” It is “the same seats are being refilled by someone else.” That someone else is the certified digital reporter, trained in a months-long online program, who never sat next to a veteran stenographer learning which margin rule applies in which county.

The rules those newcomers must satisfy got stricter and stayed fragmented. California’s Minimum Transcript Format Standards (16 CCR § 2473) were amended 11 September 2024, operative 1 January 2025. Subsection (b) is blunt: failure to comply “constitutes grounds for disciplinary action.” The federal standard is older and harder — the Judicial Conference format, in place since 1944, states that “minor changes result in significant monetary losses to parties” and that “no court, judge, supervisor, reporter, or transcriber may authorize a deviation,” because per-page rates “are based on strict adherence to the prescribed format.”

The specs genuinely conflict. Not stylistic differences — arithmetic ones:

Spec	Federal (Guide Vol. 6 Ch. 18)	California (16 CCR § 2473)	Nevada (NAC 656.360)
Lines of text/page	exactly 25	no fewer than 25	at least 25
Character density	10 per inch (~63/line)	≥56 chars/full line (52 if timestamped)	9–10 per inch
Left margin	1¾ inch	≤10 spaces to text; Q/A within first 8	≤1¾ inch
Right margin	⅜ inch	—	≤¾ inch
Colloquy	speaker ID at 10th space, text 3rd space after colon	≤10 spaces, text ≤2 spaces after colon	≤10 spaces
Parentheticals	open paren 5th space, remark 6th	≥35 chars/line if 2+ lines	≤15 spaces

Same profession, same week, three incompatible rulebooks. And it goes deeper than states: NY has its own Part 108 regime governing format and payment rates, Texas publishes a Uniform Format Manual, Washington (WAC 308-14-135) and Louisiana (RS 13:961.2) have their own, LA Superior Court publishes a Master Format Manual, and individual federal districts — Wyoming, Northern Indiana, Montana, Eastern North Carolina, the Court of Federal Claims — each post their own format PDF.

The kicker is the California Court Reporters Board’s own FAQ on the MTFS. Which rulebook applies is a conflict-of-laws question, not a preference: California’s standards govern a deposition unless the jurisdiction where the case is venued has its own; they apply to federal-court depositions because the FRCP does not specify deposition format; and — Question 1 — the CRB will discipline a California licensee for failing to follow another jurisdiction’s format. The reporter is accountable to rules from a state they may never have worked in.

Provenance:

• Signal 1 (demand/economic): AAERT 2025 Court Reporting Industry Trends Report — stenographer workforce −21% to ~23,000, school enrollment −74%, 76% of legal professionals cite scheduling difficulty, 55% report rising costs, 96% name accuracy the top KPI — https://aaert.org/industry-report/ — 10 Apr 2025
• Signal 2 (regulatory): California MTFS, 16 CCR § 2473, amended 11 Sep 2024 / operative 1 Jan 2025, with § 2473(b) making noncompliance grounds for discipline, plus the CRB FAQ establishing that venue determines the governing format and that CA will discipline for out-of-jurisdiction format failures — https://courtreportersboard.ca.gov/lawsregs/mtfs.pdf — retrieved 1 Aug 2026
• Signal 3 (feasibility/economic): Federal Judicial Conference transcript format, Guide Vol. 6 Ch. 18 — 25 lines, 10 cpi, 1¾”/⅜” margins, fixed speaker designations, 13 mandatory title-page fields, no deviation permitted, per-page rates premised on strict adherence — https://www.nced.uscourts.gov/pdfs/TranscriptFormatting.pdf — retrieved 1 Aug 2026; corroborated by divergent specs in NAC 656.360 — https://www.law.cornell.edu/regulations/nevada/NAC-656-360 Category: Underserved niche

3. The opportunity

Every incumbent in this workflow sells the wrong half of the job.

ASR vendors — Verbit, Sonix, Descript, the “AI for scopists” crowd — sell words. They are in a race to the bottom on accuracy-per-dollar and they compete with each other, not with me. None of them will tell you that your parenthetical ran 31 characters on a two-line notation and California says 35.

CAT software — Case CATalyst, Eclipse, StenoCAT — sells layouts. This is closer, but the burden sits on the human: the reporter must already know which layout is correct and must have built it. The state of the art is visible in an Orange County Superior Court document on transcript formats, which describes reporters hand-preparing matching templates across Case CATalyst, Eclipse and StenoCAT for both 26- and 28-character formats, and notes that when an out-of-state firm has particular requirements — 58 characters per line, 9 characters per inch — “an out of state reporting firm may send a sample transcript page so the reporter can set up their layout to match.”

Read that again. In 2026, the mechanism for satisfying another jurisdiction’s binding format rule is: someone mails you a picture of a page and you eyeball it.

That is the gap. Nobody sells the decision — which rulebook governs this specific transcript — or the proof that the finished document satisfies it. The specs are numeric, published, and mechanically checkable. The consequence of getting them wrong is a disciplinary complaint, a rejected transcript, or a page count that misprices the invoice. It is a compliance product wearing a document-tooling costume, and it is sitting unbuilt while the population that most needs it — newly-certified digital reporters — grows every year.

Why hasn’t it been built? Because it looks too small and too boring to the ASR companies chasing enterprise legal, and it requires patiently assembling a rules corpus that no single reporter needs in full. Any one reporter needs three or four jurisdictions. The corpus only pays off if you serve thousands of reporters — which is exactly the shape of a bootstrapped micro-SaaS and exactly the shape VCs walk past.

4. Target market

Primary customer: The freelance deposition reporter and the small reporting agency — 1 to 25 reporters — in the US, with a strong bias toward certified digital reporters (AAERT-credentialed) and the transcribers working behind them. Secondary: independent legal transcriptionists and scopists who deliver finished transcripts to agencies.

Why they buy: Their page rate is their paycheck, and format is what makes a page a page. Freelance per-page rates run roughly $3.50–$8.00 for originals; a 4-hour deposition yields ~150–250 pages and a full day 250–400, so a single transcript is $600–$1,500 of billable work. Lines-per-page is fixed by rule, which means a format error is a billing error — pages are the unit of sale. California compounds it: a reporter who redoes a transcript and blows the delivery window collects only 90% of the fee after 30 days and 80% after 60. And in California the downside is not merely commercial — § 2473(b) makes it grounds for discipline against the license they earn their living with.

Rough TAM reasoning: ~17,700 US court reporters per BLS, plus the digital reporters and transcribers that AAERT certifies who are not all captured in that occupational code, plus scopists. Call the realistic addressable population 25,000–35,000 individuals and a few thousand small agencies. At $39/mo for individuals, capturing 2,000 of them is ~$936K ARR. This is a small market. It is supposed to be — that is why it is empty.

Why now for them: The people entering these seats today did not inherit the oral tradition that used to carry format knowledge. A stenographer who trained for two years absorbed the local rules from a mentor. A digital reporter who completed an online certificate in four months did not, and is being handed multi-jurisdiction freelance work immediately because 76% of legal professionals report difficulty finding anyone at all.

5. Product sketch (MVP)

• Venue resolver — you enter the case caption, court and venue; it tells you which format regime governs this transcript and why, citing the specific rule. It handles the awkward cases: a California-licensed reporter on a federal deposition, a depo taken in one state for a case venued in another.
• Pre-certification check — upload the finished transcript (PDF or the export from your CAT software) and get a pass/fail against every numeric rule in the governing spec: lines per page, characters per line, margins, Q/A and colloquy indentation, parenthetical widths, page headings.
• Line-by-line violation report — not “format error” but “page 47, line 12: colloquy text begins 4 spaces after the colon; California allows no more than 2.”
• Title page and certificate auditor — checks the 13 federally mandated title-page elements are present, including the easily-missed one: the method by which proceedings were recorded and the method by which the transcript was produced.
• Speaker designation linter — flags speakers not conforming to required designations (THE COURT, MR./MRS./MS. + last name, THE WITNESS, THE INTERPRETER) and inconsistent naming across the document.
• Inaudible sweep — counts and locates every “inaudible”/“indiscernible” tag, because the federal guide says these are to be used only when transcription is impossible; a transcript with 40 of them is a rejection risk.
• Page-count certificate — an exportable statement of billable pages computed under the governing rule, so the invoice and the transcript agree.
• Jurisdiction diff — “you normally work federal; this one is Nevada. Here are the 6 things that change.”

6. AI angle — what’s load-bearing

Be honest about the split, because it decides whether this is a real AI product or a regex with a marketing budget.

The numeric checks are deterministic and should be — margins, pitch, line counts and indentation are arithmetic, and a compliance tool that hallucinates is worse than no tool. Nobody wants a probabilistic answer to “how many lines are on page 12.”

AI is load-bearing in three places where deterministic parsing fails:

1. Reading the rules. The corpus is a few hundred PDFs of statutes, admin codes, format manuals and per-district guidance, written in inconsistent legal prose across decades. Turning “text is to begin no more than 10 spaces from the left margin” and “typing is to begin on each page at the 1-3/4 inch left margin” into a common machine-checkable schema — and re-doing it when a rule is amended, as California’s was in September 2024 — is exactly the extraction-and-normalization job that got tractable recently. Doing it by hand for 50+ jurisdictions is what makes this unattractive to a solo builder without AI.

2. Understanding document structure. Deciding that a given line is colloquy rather than Q&A rather than a parenthetical rather than a page heading — from a rendered PDF, across the layout quirks of three CAT packages — is classification, not arithmetic.

3. The venue decision. Mapping a case caption to a governing regime, where the correct answer depends on venue, on whether the reporter is official or freelance, and on whether the venue jurisdiction has its own standards at all.

Remove the AI and you can still build a federal-only checker for one CAT export. You cannot build the multi-jurisdiction corpus that makes it worth paying for, and the corpus is the product.

7. Localization angle

N/A — this is a US-only play, and deliberately so. The entire value is a rules corpus keyed to US federal districts and state boards. There is no version of this that ports; the UK, India and Australia have completely different record-capture regimes. Depth in one country beats breadth here. The natural expansion is not geographic but adjacent: captioners (CART/broadcast) and medical-legal transcription have their own format regimes and the same corpus machinery applies.

8. Business model — path to $1M–$5M ARR

• Pricing: $39/mo solo reporter (unlimited transcripts, all jurisdictions). $149/mo small agency up to 10 reporters with shared templates and an agency-level compliance log. $399/mo for 11–40 reporters. Per-seat above that.
• ACV: ~$470 solo, ~$1,800 agency. Blended target ~$700.
• Math to $1M ARR: 1,400 solo subscriptions + 250 agency seats ≈ $658K + $450K ≈ $1.1M. Against a population of 25,000–35,000 individuals, that is roughly 5% penetration of individuals plus a few hundred agencies. Achievable, not trivial.
• Math to $5M ARR: Requires going beyond deposition reporters — captioners, medical-legal transcription, and the transcript-production arms of the national agencies (Veritext, Esquire, U.S. Legal Support) buying at the firm level rather than reporters buying individually. Realistically $5M needs one enterprise motion, which is a different company. I would build this to be an excellent $1–2M business and treat $5M as optional upside, not the plan.
• Expansion path: Volume tiers, then a “certified clean” output stamp agencies can show clients, then the page-count certificate becoming the billing artifact — once invoices reference it, switching cost climbs sharply.

Gross margin is high: the only per-transcript cost is document parsing and some model inference, well under a dollar on a 300-page transcript.

9. Go-to-market wedge — first 100 customers

• Publish the format matrix as free tooling. Build a public, always-current comparison of transcript format rules by jurisdiction — the thing that currently does not exist and that reporters presently solve by mailing each other sample pages. Every reporter searching “Nevada transcript format” or “California 25 lines per page” lands on it. This is the top of funnel and it is also the credibility proof; it demonstrates the corpus before asking for money.
• Go where the newly certified are. AAERT credentials the digital reporters who need this most, and the online training programs (ed2go, and the university-branded versions at Texas State, Houston, Central Arizona, MATC) graduate cohorts continuously. Offer free accounts to students and a partner deal to the programs — a newly-certified reporter with no mentor is the single highest-intent user of a format checker, and they arrive on a predictable schedule.
• Cold outreach to small agencies with a real artifact. Agency directories and state board licensee lists are public. Do not send a pitch — run a sample transcript through the checker and send the actual violation report for a real page, plus what it would cost them at their page rate. The report is the demo. Target 500 agencies, expect a low double-digit reply rate on a personalized finding, close 5–8%.
• State association conferences and Facebook groups. Court reporters are a tight, association-heavy profession with active state groups (CCRA in California and its equivalents elsewhere). One talk titled “The six format rules that get transcripts kicked back” is a customer list.

10. Build complexity — justification

Low. No novel technology. PDF layout parsing to extract character positions, line counts and margins is a solved problem with off-the-shelf libraries; the rule checks are arithmetic; the AI work is document classification and one-time-per-jurisdiction rule extraction using standard model APIs. A solo builder ships a federal + California checker in 6–8 weeks and expands the corpus jurisdiction by jurisdiction after that. The genuine work is not engineering, it is the patient, correct assembly of the rules corpus — which is also the moat, so the effort is well spent rather than wasted.

The real risk to the build is accuracy discipline, not difficulty: a compliance tool that produces false passes is worthless. Budget for a validated test suite of real transcripts per jurisdiction.

11. Gating checklist

Gate	Pass?	Note
Legal in target market	✅	Checks documents against published public rules. Sells no legal advice and does not certify anything — the reporter still signs.
Ethical — no harm / dark patterns	✅	Helps a licensed professional comply with binding standards. Improves record accuracy, which is a public good.
Market exists (evidence above)	✅	Binding rules with disciplinary teeth, a documented manual workaround, ~17,700 reporters plus digital reporters and transcribers.
1–5 person team can build this	✅	Solo-buildable v1; corpus expansion is incremental.
Launchable with <$50K / ₹40L	✅	Realistically under $10K to first revenue.

All five pass.

12. Feasibility score

Axis	Weight	Score	Notes
Problem intensity	20	15/20	Real and recurring — every transcript, several a week. Consequences are concrete: disciplinary exposure under § 2473(b), rejected transcripts, mispriced invoices, CA late-delivery fee cuts of 10–20%. Held below 17 because most reporters working one familiar jurisdiction have an adequate workaround (a template that already works), so it is “painful when it bites” rather than daily hair-on-fire.
Demand evidence	15	10/15	Strong regulatory and structural evidence: binding conflicting rules from four primary sources, a documented manual workaround (mail a sample page), a named dated industry report showing workforce turnover. What is missing is the thing that matters most — I found no verbatim practitioner complaints and no evidence anyone is currently paying for this specific product. Cannot score above 10 without that.
Build feasibility	15	13/15	Solo build, off-the-shelf parsing and model APIs, 6–8 weeks to a two-jurisdiction v1.
Distribution clarity	15	11/15	Named, reachable channels: AAERT and the training programs, public licensee lists, state associations, and a free format matrix as SEO bait. Docked because the profession is conservative and slow-moving, and the free-tool funnel takes months to compound.
Revenue mechanics	15	11/15	Pricing fits the wallet — $39/mo against a $600–$1,500 transcript is trivially justified. $1M ARR is credible at ~5% penetration. Docked because the ceiling is genuinely low; $5M requires a different, enterprise motion.
Time to first revenue	10	8/10	Sellable the day the federal + California checker works. Individual buyers, no procurement, no enterprise cycle. Realistically 8–10 weeks.
Defensibility	10	7/10	The maintained multi-jurisdiction corpus compounds and is tedious to replicate — rules amend (CA did in Sept 2024) and staleness is fatal to a copycat. Not a 9: any competent builder could copy the federal checker in a month; the moat is breadth and freshness, earned over 12 months, not present at month 3.
Total	100	72/100

13. Qualitative modifiers

Founder-fit tags

technical-heavy · domain-expertise-required

Document parsing and corpus engineering are the core work, but this fails without a court reporter co-founder or a paid domain advisor. The edge cases — when does a page break not count against the 25, how do agencies actually compute billable pages — are exactly where a naive implementation loses trust permanently.

Key assumptions to validate (3–5)

1. Assumption: Reporters experience format rejections or corrections often enough to feel it — not a once-a-year event. How to test: 25 structured interviews with freelance reporters and small agency owners, asking for specific counts over the last 12 months. This is the assumption the whole idea rests on and the one I have the least evidence for.
2. Assumption: A meaningful share of reporters work across more than one jurisdiction. If most work one venue with one correct template, the venue resolver is worthless and this collapses to a thin linter. How to test: In the same interviews, ask how many distinct jurisdictions they billed into last year.
3. Assumption: Transcripts can be parsed accurately enough from real CAT exports across Case CATalyst, Eclipse and StenoCAT to check margins and character positions without false passes. How to test: Collect 20 real transcripts across the three packages and hand-verify the checker’s output against manual measurement.
4. Assumption: Individual reporters will pay personally rather than expecting the agency to buy it. How to test: Price test — put up a paid tier and see who converts before building the agency features.

Risk flags

1. Incumbent absorption: Stenograph or a CAT vendor could add a jurisdiction-aware format check as a feature. They have the file-format access and the installed base. Mitigation is corpus breadth and speed, and the fact that CAT vendors have shown no urgency here for years — but this is the genuine strategic risk.
2. Market ceiling: ~17,700 reporters is a hard demographic constraint, and BLS projects it flat. This is a $1–2M business unless it expands into captioning or medical-legal. Fine for a bootstrapper, wrong for anyone wanting more.
3. Accuracy liability: If the tool passes a transcript that later draws a disciplinary complaint, trust evaporates and there may be exposure. Requires clear terms — the reporter certifies, not the tool — and conservative flagging that prefers false positives.
4. Rule staleness: The product’s value is entirely its currency. A jurisdiction amends its standards and a wrong “pass” is worse than no product. Corpus maintenance is an ongoing operational cost, not a one-time build.
5. Unverified customer voice: I could not source a single verbatim practitioner quote. The pain is inferred — soundly, from binding rules, financial consequences and a documented manual workaround — but inferred. That is why confidence is Medium and why assumption 1 is the gate.

14. Structured verdict

Score:                  72/100
Verdict:                GO
Confidence:             Medium
Best-fit builder:       Technical solo founder who can parse documents cleanly,
                        paired with a working court reporter as advisor or co-founder
Time to revenue:        8–10 weeks
Capital to launch:      $5–10K (₹4–8L) — mostly domain advisor time and model API spend
Top 3 assumptions to validate first:
  1. Format rejections/corrections are frequent, not rare — 25 structured interviews
     asking for specific 12-month counts
  2. Reporters routinely work multiple jurisdictions — same interviews, count
     distinct jurisdictions billed last year
  3. Real CAT exports parse accurately enough to avoid false passes — hand-verify
     20 transcripts across CATalyst, Eclipse, StenoCAT
Kill criteria:
  - Abandon if fewer than 8 of 25 interviewed reporters report a format-driven
    correction, rejection or billing dispute in the last 12 months
  - Abandon if more than 70% of interviewed reporters bill into only one jurisdiction
  - Abandon if the checker cannot exceed 98% agreement with manual measurement on
    the 20-transcript validation set
  - Abandon if a CAT vendor ships jurisdiction-aware format validation before v1

15. Next step — 1-week validation sprint

• Day 1–2: Build the format matrix for six jurisdictions — federal (Judicial Conference), California, Nevada, Texas, New York, Washington — from primary sources only. This is real work product, not a mockup, and it doubles as the free tool. Publish it.
• Day 3–4: Recruit 25 reporters from state association groups and AAERT-adjacent communities. Ask three questions and shut up: how many jurisdictions did you bill into last year; how many times in the last 12 months did a transcript come back or get questioned on format or page count; what did it cost you in time and money. Do not pitch the product until the questions are answered.
• Day 5: Hand-verify the parser against 20 real transcripts collected during outreach, measuring margins and line counts manually against the tool’s output.

Falsifiable outcome: Go only if ≥8 of 25 reporters report a format-driven correction, rejection or billing dispute in the last 12 months, AND ≥30% bill into multiple jurisdictions, AND the parser hits ≥98% agreement with manual measurement. Miss any one and this is a linter, not a business — and I would rather learn that in a week than after a quarter of corpus-building.