ApertureAPI GTM Playbook — Overview

Audience: anyone. These pages assume zero prior context (the "16-year-old intern test"). Every step ends in a verifiable artifact — a logged row, a saved result ZIP, a signed PDF, a live URL. If a step does not produce its artifact, it is not done.

Doctrine change (2026-10-29): read page 39 — paid pilot, then purchase before this page. ApertureAPI is an instrument maker and optical computation company: one founder, built to order, no outside capital. The commercial path is a free bounded benchmark, $4,800 of milestone-paid pilot engineering (M1 $2,400 on signing, M2 $2,400 at the signed readout, credited in full to the instrument), then the one product in development built to order on a 50 % deposit: the bench surface profiler, H1 head on the D1 stand, $12,900 (second wavelength +$2,400; software and 90 days of onboarding included). The inline head, the I1 module and the optical computing card are portfolio, unpriced, after the first product qualifies. Parts cost is a design value; the first unit has not been built; prices are offers, not a quote book. Where this page still says "plan", "per run" or "credits", that is evaluation access, not the offer. Hardware costs, the demo-unit build and the ten-year view are on the private sales build (sales-flock sign-in and passcode).

Strategy: who we go after, why, and what we offer (updated 2026-10-29)

The headline: Turn Any Desktop Computer into a 3D Metrology Lab. The subheading: Precision 3D measurement, at cost.

The one sentence to remember: a camera and a laser record a hologram. Our software turns it into a calibrated 3D measurement with proof attached, and for labs without a laser and camera we sell the instrument too.

What we sell:

Who we go after, why, and which offer (full offer: page 22):

Both deals follow the bounded free validation (up to ~20 frames, one rig configuration, one reference measurement, 10 business days).

The supporting story for the Software deal. A lab with a rig today asks an LLM for a NumPy/SciPy reconstruction script and debugs it for a week. That is driving yourself; ApertureAPI is the Lyft: upload a frame, give wavelength, pixel pitch and distance, and get amplitude and phase back with a record of how it was made. Two things are true today and quotable:

• No code to write. Free no-account form and API, any sensor. On a public ground-truth set (inline-digital-holography-v3) our inline solver scores 17–30 dB complex SNR across the four non-speckle noise sets, against about 2 dB for the pinned open-source package PyHoloscope 1.1.3 on the same frames (proof table built from committed results). Benchmark similarity, not measurement accuracy. Speckle-dominated holograms stay low for everyone and are out of scope. Software-defined calibration recovers pixel pitch from a known target.
• Proof attached. Every result ZIP has an evidence record, a provenance ID, a quality report that marks weak runs pass, warn or fail, and a methods supplement. Accuracy against an independent physical reference is what the design partnership produces; it is not claimed in advance.

Still to build, never claim yet: a measured frame rate, multi-megapixel speed, accuracy against a physical reference, repeatability across instruments, any shipped instrument, any clinical or certified use.

Public lines (approved, all true today): - "Turn any desktop computer into a 3D metrology lab." - "Stop writing reconstruction code. Upload a frame, get amplitude and phase back with the record of how it was made. Free to try." - "Any sensor, any rig: tell us the wavelength, pixel pitch and distance; calibrate pixel pitch from a known target." - "Beat the open-source default: on public ground-truth holograms our inline solver scores 17–30 dB where the pinned open-source package (PyHoloscope 1.1.3) scores about 2 dB on the same frames — benchmark similarity, not measurement accuracy." - "No rig? We build the capture instrument to order inside a Hardware Design Partnership."

Not to say in public until measured: a frame rate, "clinical grade", "certified", an accuracy figure, an error-detection rate, or anything that implies an instrument has shipped. Page 09 and the claims boundary still govern every sentence.

What we are optimising: the pipeline, not signups

Our buyers are labs. A lab buys after its own frames have run through the service, beside a measurement it already trusts, with someone who controls a budget in the conversation. So the primary funnel is five stages, and every page in this playbook exists to move a lab from one stage to the next:

Signups are a secondary diagnostic. We still count them (the scoreboard on page 23), because a sudden drop says a listing or the upload form broke. They are not the goal, and no plan may trade pipeline work for signup volume.

Weekly pipeline KPIs (report every Friday)

Stage counts never go backwards in a report: a lab that stalls keeps its highest stage and gets a next_action_date. Do not round a stage up — a promise to send frames is still stage 1.

New employee? Start at page 23, then page 14

Page 23 — the 30-day pipeline plan is the front door for anyone starting this month: daily quotas on qualified conversations, workflow-fit checks, frames received, 48-hour returns, talks and clinics booked. Page 14 — Start here is the background, the path and the glossary.

The pipeline pages (primary)

Segment buyer pages (send the one that matches the lab, never all three): biomedical QPI / DHM · university holography labs · industrial QA. Evidence pages: validation docket · reproducibility bundle · methods supplement · procurement.

Sales-flock operations (standing pages)

Discovery plumbing (secondary)

Pages 01–08 are developer listings, registries, a demo host and community accounts. They are useful plumbing — they make the service findable by someone already searching — but they do not generate institutional demand on their own. Do them once, in this order, then leave them; no daily quota on the pipeline pages may be skipped to work on them.

Order matters: the MCP registry entry needs the PyPI package to exist first, and the public-apis PR needs the Postman collection URL.

Hard rules (read before touching anything)

1. Claims discipline. Copy-paste only from the prepared copy blocks in these pages. NEVER write a correlation/agreement figure, an error-detection percentage claim, or any frames-per-second number — those have no committed evidence behind them. The honest replacement when asked about agreement: "we publish our agreement numbers against a public dataset, and the report says where it falls short — I'll send it." Quotable without a figure: the SDC closed-loop calibration recovers pixel pitch very tightly on two committed targets (pitch only), and every result ships a re-computable provenance record. Full list of banned forms and their honest replacements: marketing/sales/claims-boundary.md.
2. Research tooling, not diagnosis; lab-side, not in-plant. To biomedical groups the service is research tooling for reproducibility and continuity — never a diagnostic. To industrial groups it is a lab-side comparison against their reference measurement — never an in-plant, in-line or line-speed promise.
3. The quickstart is two steps, not one. Calling the public endpoint requires first fetching a form token (GET /public/form_token), waiting at least 3 seconds, then POSTing to /public/reconstruct with the same User-Agent from the same IP. Every curl example in these pages includes the token step. Do not "simplify" it to one curl — it will 403.
4. Accounts: where a page says to create an account, use the company identity (operator supplies the email/credentials); never a personal one.
5. Log everything. Date, account used, URL or file, stage, artifact. Progress is countable, not described.
6. No paid acquisition. No paid ads, paid listings, bought or rented contact lists, sponsorships, booths, or paid demo-host upgrades (page 05 is free tier or skip). The one budgeted event is a visitor badge for SPIE Photonics West 2027, with an operator go. Anything else that costs money needs a written operator go and a reason it moves a lab along the five stages.
7. No outbound to production lines or do-not-approach targets (page 15, page 29). They may find us; we log them; we do not approach them.