The Priority Engine

4 filters unified. 321 experiments ranked. Math dictates what to work on.

321
Experiments
15
Scored
306
Blocked
$2.44M
Top-10 Impact
98%
Avg Trust
94%
Avg Math
98%
Avg Method

The 4 Filters GATES

Every experiment must pass through four sequential gates. Fail any gate and it's blocked from the queue until remediated.

1

PROTOCOLS

Is this something we should work on at all? Blocks Russ legacy campaigns, experiments missing an owner, and experiments with no documented mechanism.
321 → 48passed
2

TRUST ENGINE

5-layer verification chain: observation verified, baseline proof, mechanism understood, metric measurable, target defensible. No trust = no score.
48 → 22passed
3

MATH ENGINE

10 checks: decomposition, Bayesian prior, z-score, Monte Carlo, SPC, sensitivity, constraint, backtest, dimensional analysis, logical consistency.
22 → 18passed
4

SCIENTIFIC METHOD

Observe → Hypothesis → Plan → Execute → Verify. Each experiment must have every phase documented before it earns a ranking.
18 → 15passed
321 → 15 passed all filters · 306 blocked

The 306 blocked experiments aren't dead — most are blocked on "no mechanism documented." Adding mechanism unlocks scoring.

Top 10 Priority Queue LIVE

Click any row to expand Trust / Math / Method breakdown. Score = (trust × math × method × impact × priority_weight × step_weight) / effort.

#IDNameScoreStep TrustMathMethodImpactEffort
1E315Weaponization Ladder Enforcement 600,0003 / 6 100%90%100% $500K3d

Trust Layers (5/5)

  • Observation verified
  • Baseline proof on file
  • Mechanism understood
  • Metric measurable
  • Target defensible

Math Checks (9/10)

  • Dimensional analysis: minor unit ambiguity

Method Phases (5/5)

  • Observe
  • Hypothesis
  • Plan
  • Execute
  • Verify

Ladder Position

  • Step 3: Timer Running
  • Next: Producing Output
  • Force-unblock in 14d if stuck
2E309Landing Page Launch Blitz 214,2862 / 6 100%100%100% $750K14d

Trust

  • All 5 layers green

Math

  • Full 10/10 checks passed

Method

  • Full cycle documented

Notes

  • Highest single-experiment impact on queue
  • Blocked on Audrey deliverables
3E300Smart Bidding Clean Data Recovery 200,0002 / 6 100%100%100% $350K7d

Trust

  • Phantom revenue documented
  • job_status filter enforced

Math

  • 10/10 checks passed

Method

  • Fix scope defined

Notes

  • Fixes "$128K phantom" pattern
  • Unblocks all ROAS reporting
4E306Antibody Actualization 180,0002 / 6 100%100%100% $200K3d

Trust

  • Prevention rules codified

Math

  • Full 10/10

Method

  • Brain chunk linkage confirmed

Notes

  • Converts past failures into harness rules
5E311Session Decay Timer 90,0003 / 6 100%100%100% $100K3d

Trust

  • Decay pattern documented

Math

  • All checks passed

Method

  • Re-read checklist every 5-7 tasks

Notes

  • Forces periodic self-audit
6E312Gemini Notes Auto-Pull 72,0002 / 6 100%100%100% $80K3d

Trust

  • Google Workspace API verified

Math

  • 10/10

Method

  • Full method cycle

Notes

  • Auto-captures meeting action items
7E301CRO Form Position Fix 51,4293 / 6 100%100%100% $180K14d

Trust

  • Snippet #55 redeployed

Math

  • Impact model validated

Method

  • A/B plan ready

Notes

  • Currently timer running, no output
8E305RAG Incident Ingestion Loop 45,0003 / 6 100%100%100% $50K3d

Trust

  • 2,808 chunks verified

Math

  • 10/10

Method

  • Auto-feedback loop live

Notes

  • Closes the institutional-memory gap
9E308Oxygen Builder Health Monitor 36,0002 / 6 80%80%80% $50K3d

Trust (4/5)

  • Target defensibility pending

Math (8/10)

  • Backtest and SPC pending

Method (4/5)

  • Verify phase TBD

Notes

  • Partial credit — still makes queue
10E307CRO Form Architecture Rebuild 30,8573 / 6 100%90%100% $180K14d

Trust

  • Full 5/5

Math (9/10)

  • Monte Carlo needs rerun

Method

  • Full cycle

Notes

  • Dependent on E301 outcome

The Master Formula ONE NUMBER

score = (trust × math × method × impact × priority_weight × step_weight) / effort
trust0.0 – 1.0 via 5-layer chain.
e.g. 1.00 = all verified
math0.0 – 1.0 via 10 checks.
e.g. 0.90 = 9/10 passed
method0.0 – 1.0 via 5 phases.
e.g. 1.00 = full cycle
impactProjected dollars ($).
e.g. 500,000
priority_weight1.0 base, 1.2 if Stephanie-requested.
e.g. 1.2
step_weightWeaponization ladder step multiplier.
1/1, 2/2, 3/3 ... 6/6
effortDays to complete.
e.g. 3 = three days

Worked example — E315: (1.00 × 0.90 × 1.00 × 500000 × 1.0 × 0.5) / 3 ≈ 600,000

Weaponization Ladder 6 STEPS

Where our 15 scored experiments currently sit. The jump from Step 3 → Step 4 is the gap — timers running but no verified output yet.

Step 1

Built

0
Good — nothing stuck pre-deploy
Step 2

Deployed

6
E309, E300, E306, E312, E308, E303
Step 3

Timer Running

9
E315, E311, E301, E305, E307, E310, E313, E314, E316
Step 4

Producing Output

0
NEEDS output verifier
Step 5

Output Read

0
Nobody reading reports yet
Step 6

Behavior Changed

0
The real finish line

306 Blocked Experiments BACKLOG

306 of 321 experiments are blocked from scoring. Adding a documented mechanism to each unlocks the queue. This is the single biggest cleanup opportunity.

241
No mechanism documented
34
Missing owner assignment
18
No measurable metric
9
Russ legacy (protocol block)
4
Failed dimensional check

Sample blocked experiments

  • E142 — Sewer camera email nurture sequence · Reason: no mechanism
  • E188 — Technician attribution bonus model · Reason: no measurable metric
  • E203 — Weather-triggered ad bid modifier · Reason: no mechanism
  • E221 — Review velocity engine v2 · Reason: missing owner
  • E244 — Facebook remarketing audience split · Reason: no mechanism

How To Use DAILY

Daily 5 AM CT

Priority engine recalculates. Queue refreshes overnight. Top-10 stored in stats.json.

Session Start

Hit GET /api/priority/next to know what to work on. No guessing.

Monday Standup

Pull the queue as talking points. Stephanie + Kalen see why each item is #1.

Execution Rule

Robert picks #1. Executes to Step 6 before starting #2. No parallel drift.

Meta Timer

If #1 stays #1 for 14 days: force-activate or archive. No zombies.

Session Logger

Every completed step logs back. Feeds the feedback loop. Compound learning.

API Endpoints PORT 8765

GET/api/priority/next
Returns the #1 priority experiment — full scoring breakdown.
curl http://localhost:8765/api/priority/next
GET/api/priority/queue?limit=N
Top N ranked experiments. Default 10, max 50.
curl 'http://localhost:8765/api/priority/queue?limit=10'
GET/api/priority/stats
Filter stats — scored vs blocked, average trust/math/method, total impact.
curl http://localhost:8765/api/priority/stats
GET/api/priority/all
Full ranked list (all 321 including blocked, with block reasons).
curl http://localhost:8765/api/priority/all | jq '.queue[0:5]'

Why This Changes Everything BEFORE / AFTER

BEFORE the engine

  • Robert asks "what should I work on?" and gets an opinion.
  • 321 experiments sit in a list with no rank.
  • Shiny new idea beats dollar-weighted evidence.
  • Zombies accumulate — timers running, nobody reading output.
  • Standups become vibes, not math.

AFTER the engine

  • Math answers. One score per experiment. One queue.
  • Trust + Math + Method + Weaponization Step = the number that wins.
  • Blocked experiments show the exact gate they failed — fix list is explicit.
  • 14-day meta timer kills zombies automatically.
  • Robert, Stephanie, Kalen all see the same #1. No debate.