How to Track Startups Before They Announce: The Six-Layer Observability Guide
The six observable layers of a pre-round startup, engineering, hiring, product, dependency, community, registry, ranked by lead time, plus the one-hour weekly routine that converts public traces into sourced deals.
Key Takeaway
Everything a startup does before announcing a round leaves public traces across six layers: engineering, hiring, product, dependency, community, and registry. This guide ranks the layers by lead time (engineering first at 3-6 weeks), defines the honest limits, and templates the one-hour weekly scan-cross-check-act routine that turns free observability into sourced deals with receipts.
The most valuable window in early-stage investing is the weeks before a round becomes public, and almost everything that happens in that window is observable for free. Startups cannot build in secret anymore: they commit code, hire engineers, ship changelogs, and adopt dependencies, and each action leaves a public trace ahead of the announcement. This guide assembles the full observability map and the weekly routine that turns it into sourced deals.
The six observable layers of a pre-round startup#
- Engineering layer: commit velocity, contributor growth, repository expansion on GitHub, the richest and earliest trace, covered in the GitHub patterns guide.
- Hiring layer: engineering job postings, usually two to eight weeks ahead of the announcement.
- Product layer: changelogs, release notes, docs expansion, feature flags flipping on.
- Dependency layer: new libraries appearing in package manifests as the stack grows.
- Community layer: founder posting cadence, conference talks, meetup appearances.
- Registry layer: the round itself entering Crunchbase and Form D, trailing everything else.
Each layer is free; the 47-source catalogue indexes them with limits and lag times.
Lead times, ranked#
Engineering signals lead announcements by three to six weeks in the tracked panel; hiring two to eight; product one to four; community zero to two; registry negative, it trails. The ordering is stable across sectors, though absolute levels differ; sector-normalized reading is required before comparing an AI company with an enterprise SaaS company. The leading vs lagging taxonomy formalizes which observables belong in a sourcing stack and which belong in verification.
The weekly routine#
One hour, three steps. Scan: pull the weekly momentum movers and the sector cut. Cross-check: for the top decile, layer hiring and product traces onto the engineering read, keep only multi-layer confirmations. Act: outreach with the evidence in the first line; log the receipt. The routine's power is cadence: single observations are noise, weekly sequences are evidence, a discipline the sourcing workflow guide templates hour by hour.
What this cannot do#
Observability is coverage-biased: open-source and developer-facing companies are over-represented, stealth hardware and enterprise-internal builds under-represented. It is evidence of preparation, not proof of a round, base rates matter and multi-layer confirmation is mandatory. And it decays: the company accelerating this week is being pitched everywhere next month, so the routine is worthless without the weekly cadence.
Key takeaways#
Six free layers, engineering, hiring, product, dependency, community, registry, ordered by lead time, with engineering earliest at three to six weeks. The routine is scan, cross-check, act in one weekly hour, keeping only multi-layer confirmations. Observability does not replace judgment; it schedules it, putting your attention on the right companies at the only moment the price of attention is still low.
Receipts and base rates: the honesty layer#
The routine only compounds if it is logged. For every outreach, record: the company, the layers that fired, the date, and later, the outcome, round announced or not, when. Two things come from that file. First, your own base rates: after a quarter you know what a real precursor looks like in your sectors, not in aggregate. Second, receipts: dated evidence you identified companies before announcement, which is the single most persuasive artifact in scout program applications, LP conversations, and founder outreach alike. The logging is the difference between a routine and an asset.
The discipline that keeps it honest: log the misses too. The company you flagged that never raised, the acceleration that turned out to be a hack week, these are not embarrassments, they are the denominator. Without them, every retrospective becomes a highlight reel and the base rates rot.
Tooling the routine#
Everything the routine needs is free. The weekly movers list and the signals API cover the engineering layer with sector cuts and deltas. Job boards and careers pages cover hiring. Changelogs and release feeds cover product. For the agent-inclined, the MCP server puts the whole panel inside Claude or Cursor, and the AI sourcing guide shows the Monday-loop pattern that compresses the scan to minutes. The spreadsheet is yours to build; the weekly workflow template provides the column structure that has survived real use.
The competitive context#
This observability window is not secret, but it is underused, because exploiting it requires cadence rather than budget. Funds with data budgets tend to buy the trailing layers, the databases that record what already happened, because those come standardized. The leading layers, engineering, hiring, product, remain the province of whoever runs the weekly routine. That asymmetry, budget buys the past, cadence buys the future, is the entire strategic argument for the observability stack, and it holds until the market standardizes on it, which the base-rate history of every previous signal says will eventually happen.