Skip to content
Primores AI wiki

Discovery-Before-Scale — The Two-Phase Operational Framework for Organic Content

Discovery-Before-Scale

TL;DR: Discovery-Before-Scale is a two-phase architecture for organic content operations. Phase 1 (Discovery, 2–4 weeks): low volume, deliberate variation across pattern × niche × format. Identify which 2–3 patterns produce the target behavioral profile. Phase 2 (Scale, indefinite): full volume, pattern-locked, surface-varied. Volume compounds because every piece is in the validated zone. The math (Pirolli & Card’s Independence of Inclusion from Encounter Rate) makes the framework non-optional — pumping volume of un-validated patterns is mathematically guaranteed to fail.

The Problem It Solves

Most content operators conflate volume with strategy. The implicit theory: if we publish enough, something will work. The data says otherwise. Per glossary/information-foraging (Pirolli & Card 1999), the Independence of Inclusion from Encounter Rate principle states:

The decision to pursue a class of items is independent of its prevalence. The decision to include lower-ranked items in a diet is solely dependent on their profitability, and not upon the rate at which they are encountered.

Translation for content: a low-profitability content type cannot earn its way into a viewer’s attention diet by volume alone. Posting more of the same un-validated pattern doesn’t raise its inclusion probability; only profitability does. This is a theorem from optimal foraging theory, not opinion. Volume × low-profitability = noise.

Discovery-Before-Scale exists to solve this. The architecture forces profitability validation before committing to scale, so the volume math actually compounds when it runs.

The Two-Phase Architecture

Phase 1 — Content Discovery

Goal: identify which 2–3 patterns produce the target behavioral profile in this specific niche.

Duration: 2–4 weeks per niche. Discovery runs in calendar time, not piece-count time — feed algorithms need days to surface and stabilize signals.

Volume: low. Roughly 15–40 pieces of content total during the discovery window. Enough variation to read signals; not so much that the team commits to scale before validation completes.

Variables to deliberately vary:

  • Pattern — pick 4–6 of the 9 slideshow patterns to test. Don’t test all 9; spread is too thin.
  • Niche framing — within the chosen glossary/super-niche, test 2–3 framings of the same audience × problem × context.
  • Format surface — copy length, hook style, visual aesthetic. Vary, but secondary to pattern × niche.
  • Cadence — daily vs every-other-day vs sporadic. Algorithm-dependent; secondary signal.

Variables to hold constant:

  • Brand voice and glossary/distinctive-assets (consistency is the long-term lever, not the discovery-phase variable)
  • Production quality (don’t confound pattern signal with quality signal)
  • Posting time-window (rough, not precise)

Signals to read:

  • The four behavioral ratios: save/like, share/like, comment/like, follower/profile-view
  • View distribution across pieces (clustered around a few patterns, or roughly uniform?)
  • Algorithmic surfacing decisions (which pieces got pushed to non-followers?)

The validation criterion: a pattern is validated when its ratio fingerprint matches the target consistently across at least 3–4 pieces, and the algorithmic surfacing favors it over other patterns tested in the same window. Both signals together — single-piece success doesn’t validate; algorithmic preference for the pattern does.

What can go wrong:

  • Too few variants — the team commits to a pattern that won within the test set but wasn’t actually the best
  • Too many variants — signal-to-noise too low; no pattern gets enough volume to surface clearly
  • Pattern × niche confusion — different patterns work in different niches; test the chosen pattern × the right niche framing, not pattern in isolation
  • Aesthetic preference contaminating decisions — the team picks the pattern they like rather than the one with the best fingerprint

Phase 2 — Scale Distribution

Goal: maximize aggregate distribution within the validated pattern × niche zone.

Duration: indefinite. Phase 2 can run for months; the limit is when algorithmic patterns shift enough to invalidate the original validation, at which point a new Discovery loop fires.

Volume: full. Scale automation kicks in here. The Primores benchmark math: 9,000 pieces × 100 accounts = 6–9M views per active campaign. But only if patterns are validated.

Variables to vary deliberately:

  • Surface — copy, hook variations within the same pattern, visual variations. The pattern stays; the surface rotates.
  • Distribution accounts — multiple accounts running parallel campaigns. Each account develops its own distinctive identity within the same pattern envelope.
  • Niche sub-segments — once one pattern × niche works, related niche sub-segments often respond to the same pattern.

Variables to hold constant:

  • The validated pattern itself. Drift here destroys the volume math. Common failure mode: operator mid-scale boredom causes pattern drift; ratio fingerprints diverge; commercial outcome weakens.
  • Brand voice and distinctive assets. Mental availability is built across the entire scale phase; consistency is what makes the assets compound.
  • The validation criterion the pattern was selected against — review fingerprints continuously; if drift starts, return to Phase 1.

The Math

The optimistic case: pattern × niche fit validated in Phase 1, scale runs cleanly through Phase 2. 9,000 pieces × 100 accounts × 1,000 average view delivery = 9M aggregate views. Conversion at branded-search-lift rates (3–5x cold-traffic conversion) produces commercial outcomes proportional to that aggregate. Compounding kicks in across the months as algorithmic feeds keep surfacing older content to newly-discovered weak-tie bridges.

The pessimistic case (un-validated scale): 9,000 × 100 × 200 average view delivery = 1.8M aggregate views, but with disorganized fingerprints. Engagement metrics look weaker; algorithmic distribution decays faster; branded-search lift is anemic; commercial outcomes underwhelm. Same hours invested, dramatically lower return.

The factor-of-5 difference between optimistic and pessimistic isn’t speculative — it tracks the difference between high-profitability content (favored under optimal-diet selection) and low-profitability content (disfavored regardless of volume). Per glossary/information-foraging, that’s a theorem.

When to Advance from Phase 1 to Phase 2

A decision tree, in roughly decreasing strictness:

  1. Are at least two patterns validated (each hitting target fingerprint across 3+ pieces)? If only one pattern validated, the niche may be category-narrow; consider testing one more sub-niche before advancing. If zero validated, run another 1–2 weeks of variation; if still zero, the niche may be wrong (return to glossary/super-niche selection).
  2. Did algorithmic surfacing distinguish between validated and non-validated patterns? If the algorithm pushed validated patterns to non-followers but ignored the others, that’s the strongest signal you can get for proceeding. If the algorithm treated all patterns similarly, validate further before scaling.
  3. Does the validated fingerprint match the strategic intent? If the team wanted save-bait but validated patterns are share-bait, the validation is real but the strategy is misaligned — either change the strategic intent or test save-bait patterns specifically before scaling.
  4. Is the team ready to hold the pattern constant for 3+ months? Scale phase requires discipline. If team aesthetics or boredom will likely cause pattern drift in week 4, postpone scale until that’s resolved (more variants in Phase 1; team alignment on the scale-phase rules).

If all four are yes, advance.

Common Failure Modes

  • Skipping Phase 1 entirely. “Let’s just start posting and see what works.” Almost always produces 2–3 months of disorganized content followed by realization that nothing’s compounding. The honest fix is admitting Phase 1 is necessary; pretending otherwise is expensive.
  • Phase 1 that runs forever. “We’re not sure yet, let’s test more.” Sometimes the patterns aren’t validating because the team can’t read the fingerprints; sometimes because the niche is wrong; sometimes because Phase 1 has lost discipline. After 4 weeks without validation, revisit niche selection and signal-reading methodology rather than running another 4 weeks of more-of-the-same.
  • Pattern drift during Phase 2. Most common after months of scale, when team boredom or aesthetic preference causes incremental drift. The pattern slowly shifts; ratio fingerprints diverge; the volume math stops compounding. Catch this early by reviewing fingerprints monthly; restore the validated pattern or return to Phase 1.
  • Scale without sufficient distribution accounts. 9,000 pieces from one account hits diminishing returns from the algorithm’s diversity penalty. Multiple accounts (each with its own distinctive identity within the pattern envelope) avoid this. Plan for multi-account scale before committing to Phase 2 volumes.

Why Discovery Is the Moat

The visible part of organic content scale operations is automation — bulk content production, multi-account distribution, scheduled posting, algorithmic-feed monitoring. Competitors can copy this in 2–3 months given clear goals.

The invisible part is the validated pattern × niche × fingerprint mapping that makes the automation actually work. That takes 2–4 weeks per niche to validate, and the validation work is hard to copy because it’s tacit — it lives in the team’s signal-reading skill, not in any automation script.

A competitor running the same volume on un-validated patterns produces an order of magnitude less value, even with identical infrastructure. Discovery is the moat. Cheap automation is the visible part.

Operational Notes

A few specific implementation choices Primores has converged on:

  • Discovery is calendar-bounded, not piece-bounded. A team that hasn’t validated after 4 weeks usually doesn’t validate by piece 80; the issue is reading or niche selection, not volume. Cap discovery at 4 weeks before reassessing.
  • Use multiple accounts during Phase 1, not just Phase 2. Algorithmic-feed signals at the account level differ from signals at the piece level; running 2–3 accounts during discovery gives cleaner pattern signal.
  • Don’t optimize Phase 1 for view count. Phase 1’s purpose is signal validation, not view production. Phase-2 view counts will dwarf Phase-1’s regardless.
  • Document the validated pattern’s specifics. When the pattern works, capture why — hook structures, visual rules, timing, voice constraints. This documentation is what survives team turnover and prevents pattern drift.

Key Takeaways

  • Phase 1 (Discovery, 2–4 weeks): low volume, deliberate variation across pattern × niche × format. Identify validated patterns matching target fingerprint.
  • Phase 2 (Scale, indefinite): full volume, pattern-locked, surface-varied. 9K pieces × 100 accounts = 6–9M views only if patterns are validated.
  • The math (Independence of Inclusion from Encounter Rate, Pirolli & Card 1999) makes the framework non-optional — pumping volume of un-validated patterns is mathematically guaranteed to fail.
  • Validation criterion: ratio fingerprint matches target across 3+ pieces and algorithm distinguishes validated patterns from non-validated.
  • Phase-2 pattern drift is the most common failure mode after months of scale; review fingerprints monthly.
  • Discovery is the moat; cheap automation is the visible part.

Sources

  • Pirolli, P. & Card, S. (1999). Information Foraging. Psychological Review, 106(4), 643–675. — The optimal-diet selection theorem and Independence of Inclusion from Encounter Rate.
  • Primores TikTok content operations — the working method this page documents.
  • marketing/organic-content-strategy — The pillar’s overall framing this page operationalizes.