Candle work · Witchcraft · AI · Sales playbook · Language

Four Levers, One Spine

Nothing here guarantees an outcome. What the research does support is a system where every input is guaranteed, counted, and pointed at the one number that pays you. Ritual gets you to the action. AI multiplies the actions. The playbook converts them. The log tells you the truth.

The guarantee question

What you can lock at 100%, and what nobody can

Yours to guarantee: inputs

  • The ritual happened before the hard action
  • The number of contacts made this week
  • Every lead answered inside one hour
  • Every open deal got its next follow-up
  • The log was filled in and reviewed on schedule

Not yours to guarantee: other people's yes

  • Any single buyer's decision
  • Timing, budgets, competitors, luck
  • That belief alone moves any of the above

Fantasy without an obstacle step predicts less effort, not more (Oettingen; Kappes & Oettingen 2011). The calculator below shows how close to certainty volume can take you, and why it never reaches it.

Architecture

The spine

One loop, run once per work block. Each lever has one job and hands off to the next.

01 · CANDLELightFixed 90-second ritual. Same candle, same words. It is the cue that the block has started.
02 · CRAFTContrastName the outcome, then the inner obstacle, then speak one if-then: "When I feel the stall, I dial."
03 · AILoadThe list, research and drafts are already queued. You open the block to a full pipeline.
04 · PLAYBOOKCloseLive conversations. Questions first, one clear ask, a next step booked before you hang up.
05 · LOGCountOne line: ritual done, contacts, replies, meetings, deals. Blow the candle out.
Weekly review reads the log and changes one thing → back to 01
The levers

What each one does, and how strong the evidence is

Evidence bars are my grading of the published research, not a measure of how much the practice matters to you.

Candle work

Job: get you into the hard action
EvidenceSuggestive, unreplicated
  • One ritual, never varied. Repetition is what turns it into a cue.
  • Fire it right before the highest-friction act: the dial, the price, the ask.
  • It scales only you. That is fine. Do not try to scale it.
Research

Brooks et al. (2016) found rituals improved performance by lowering anxiety, and only in high-anxiety tasks. I found no independent replication, and the paper's raw data is only partly available. A related "lucky charm" effect (Damisch 2010) failed replication (Calin-Jageman & Caldwell 2014). Treat ritual as a cheap, harmless cue, not a proven performance drug.

Witchcraft

Job: intention with the obstacle built in
EvidenceSmall but real (as MCII)
  • Write every spell as an if-then: "When X happens, I do Y."
  • Run the work on a 28-day cycle: set the target at the new moon, review at the full moon, close the books at the dark moon.
  • Your story is positioning. Use the same words in your offer, your pitch and your site.
Research

Mental contrasting with implementation intentions (MCII) shows a small-to-medium pooled effect, g ≈ 0.34 across 21 studies, and the authors say publication bias may shrink it further (Wang, Wang & Gai 2021). "I am the kind of person who…" identity phrasing has a mixed replication record, so the page leans on if-then plans, not identity labels.

AI

Job: volume, speed, persistence
EvidenceStrong mechanism, unmeasured for you
  • AI builds the list, researches each prospect, drafts the message and schedules every follow-up.
  • Target: every inbound lead answered inside the hour.
  • You send and you speak. AI-generated voices on calls count as "artificial" under the TCPA and need prior consent (FCC, Feb 2024).
Research

Oldroyd et al. (HBR 2011) reported firms that contacted leads within an hour were about seven times more likely to qualify them than those an hour slower; retellings disagree on the sample details. Monitoring progress improves goal attainment (d = 0.40, 138 studies), more so when it is physically recorded or made public (Harkin et al. 2016).

Sales playbook + language

Job: convert at the moment of contact
EvidenceMixed; NLP itself is weak
  • Keep from NLP only what is plain good practice: listen, mirror their words, ask before you tell.
  • Drop eye-cue reading, anchoring and "embedded commands". They do not hold up.
  • Filter for every blackbook tactic: would it still work if the buyer read it afterwards? If not, it costs you refunds and referrals.
Research

A systematic review found little evidence NLP improves outcomes, from just 10 usable studies out of 1,459 titles (Sturt et al. 2012). Even "replicated" compliance lines are shaky: "but you are free" pooled at g = 0.44, yet showed no effect in the seven low-bias studies. I have not seen your blackbook, so nothing here grades its specific tactics.

Simulation

How close to certain can volume take you?

Move the sliders to your real numbers. If you do not know them yet, that is the first thing the log fixes.

Chance of at least one closed deal
–

Expected deals:

Contacts needed for a 95% chance of one deal:

Treats each contact as independent with the same odds, which flatters the result. The curve approaches 100% and never touches it. Every lever on this page works by moving one of these five sliders.

Parallel think

Six angles on the same question

Run as six passes by one model, scored by my judgement on rigor and actionability. Not six independent agents.

StrategyConclusionScore
SimulationThe outcome is contacts × conversion. Certainty is bought with volume and it is asymptotic.86
ConstraintYour live hours are the only hard cap. AI must take everything that is not a live conversation.82
AdversarialThe likeliest failure is ritual replacing the dial. Ritual with no logged action after it counts as zero.80
FoundationalThree of the five things are inner tools and two are outer. Only outer tools touch other people.74
TemporalIn six months this dies from an unread log or from AI outreach that burned the list. Review weekly, cap sends.73
AnalogicalAthletes use pre-shot routines the same way: the routine starts the shot, the reps win the match.61

Where they disagree: if you already make plenty of contacts and avoid none of them, the ritual and craft levers add little and the playbook is the whole game. If you stall before the ask, the first two levers are worth the most. One week of honest logging tells you which.

Devil's advocate

The earlier answers, put on trial

Claim madeVerdictWhat the record shows
If-then plans give "roughly 2–3× attainment"OVERSTATEDThe 2006 meta-analysis found d = 0.65 across 94 tests. A later 642-test update drops to d ≈ 0.15–0.35 after bias correction. Useful, not a multiplier.
Visualization alone is "worse than nothing"PARTLYTrue for pure outcome fantasy in several Oettingen studies. Not a blanket finding about imagery, and the fix (MCII) is a small effect.
Rituals improve performance (Brooks 2016)SHAKYSingle paper, effect only under high anxiety, no independent replication found, partial data availability.
Identity statements drive behaviorWEAKMixed replication. Kept only as positioning copy, where it is a marketing choice and needs no psychology claim.
"AI and NLP are actual levers"HALF WRONGAI yes. NLP as a system has little supporting evidence.
AI scale is "effectively unbounded"FAILEDBounded by consent law (TCPA, CAN-SPAM), sender reputation and your hours for live calls.
"100% guarantee on execution"CONDITIONALHolds only when inputs are numeric, logged the same day, and seen by someone besides you.
The ritual is worth keepingDEFENDEDIt costs 90 seconds, risks nothing, and the worst case is a reliable start signal.
Scale plan

One 28-day cycle

Week 1 · Baseline

  • Fix the 90-second ritual and the one if-then line
  • Run the loop by hand, no new tools
  • Log all five numbers daily

Output: your real slider values.

Week 2 · Load

  • AI builds a 200-name list with a research note each
  • AI drafts first touch plus three follow-ups
  • Set a one-hour reply rule for inbound

Output: contacts per week doubles.

Week 3 · Sharpen

  • Pick the weakest slider from the log
  • Test one playbook change against it
  • Record calls; AI summarizes objections

Output: one conversion rate moves.

Week 4 · Judge

  • Compare week 4 to week 1, number by number
  • Keep what moved a slider, cut what did not
  • Set the next cycle's single target

Output: a system you can hand to a second person.

Outsourced craft

If a practitioner does the candle work for you

The research-backed part of ritual works through the person doing it. Work done on your behalf has no evidence of moving outcomes on its own, so every working of theirs is tied to an action of yours.

The brief you hand them

  • One target per cycle. A measurable aim for 28 days, such as "three new signed clients by the dark moon". Not "abundance".
  • The obstacle, named. Tell them the real inner blocker (stalling before the ask, not following up) and ask that the working address it.
  • A schedule you can sync to. Fixed dates and times: opening at the new moon, mid-cycle at the full moon, closing at the end.
  • Something physical back. A dressed candle, a petition or a phrase from each working. It becomes your 90-second cue.
  • No named targets. The work is for your own clarity, nerve and openings. No workings on specific prospects, clients or competitors, and no client details handed over.

Your side

  • Within 24 hours of each working: run your biggest outreach block. A working with no logged action after it counts as zero.
  • Daily: light their candle or read their line, speak your if-then, then make the contacts.
  • End of cycle: compare the five logged numbers with the last cycle. Keep the practitioner if contacts or conversion rose. Change something if not.

Guardrails on paying

  • Flat fee, agreed up front, per cycle. Cap it at an amount you would be fine losing.
  • Walk away from guaranteed results, escalating fees, "a curse or blockage that costs extra to remove", or pressure to keep it secret.
  • Treat it as a personal expense unless your accountant says otherwise.
Sources

Research used

  1. Gollwitzer & Sheeran (2006), implementation intentions meta-analysis, and the later 642-test update. Konstanz repository · summary
  2. Wang, Wang & Gai (2021), MCII meta-analysis, Frontiers in Psychology. PMC8149892
  3. Brooks et al. (2016), "Don't stop believing", OBHDP 137. record · BPS digest
  4. Sturt et al. (2012), NLP systematic review. Lincoln repository
  5. Oldroyd, McElheran & Elkington (2011), "The Short Life of Online Sales Leads", HBR. BYU record
  6. Carpenter (2013) and the pre-registered re-examination of "but you are free". Meta-Psychology
  7. Harkin et al. (2016), monitoring goal progress, Psychological Bulletin. White Rose
  8. FCC Declaratory Ruling, Feb 8 2024, AI voices under the TCPA. WSGR · Wiley
  9. Cited from memory, not re-checked today: Kappes & Oettingen (2011); Damisch et al. (2010) and Calin-Jageman & Caldwell (2014).