PhronesisThe chain
Design decides the spec: what counts as better, what gets measured, how much risk is acceptable, what must stay reversible. Engineering builds to the spec: instrument, test, maintain, plan for failure. The order matters. Engineering without design builds the wrong thing precisely; design without engineering stays a drawing. Designing engineering is step zero — choosing the requirements before anything is built. Applied to recovery, the stack reads engineering therapy.
The site already holds the inverse of this note: module 10 of the course, Return Paths: Therapy as Engineering, runs recovery as a control problem inside the displacement framework. This note runs the frame the other way — against the clinical record, to find where it holds and where it stops.
The kernel
The process side is established practice. Four bodies of work, all documented, none of them proposed here.
Measurement-based care. Lambert's group gave therapists session-by-session OQ-45 scores and flagged clients deviating from expected recovery curves (Lambert, Whipple, Smart, Vermeersch, Nielsen & Hawkins, 2001, Psychotherapy Research). The baseline that motivates it: 5–14% of clients worsen during treatment, and therapists are poor at spotting them — in one dataset, clinicians predicted 3 of 550 failures (documented in Lambert & Shimokawa, 2011). Pooled across the six trials, N = 6,151, feedback improved outcomes for at-risk cases (Shimokawa, Lambert & Smart, 2010, JCCP); the 2011 summary reports g ≈ 0.5–0.7 for not-on-track clients and claims the number who deteriorate “can be cut in half.” Two caveats belong in the record: nearly all of that early evidence comes from Lambert's own clinics, and the broad independent literature finds the effect smaller — d = 0.15 overall, d = 0.17 for at-risk cases, across 58 studies (de Jong et al., 2021, Clinical Psychology Review). Feedback works; it works less outside the developer's lab. A second system, PCOMS — two four-item scales, outcome and alliance — comes from Miller and Duncan (Miller, Duncan, Sorrell & Brown, 2005); its evidence is weaker and more developer-dependent (Østergård, Randa & Hougaard, 2020: g ≈ 0.27 overall, near zero in low risk-of-bias studies), and stands contested.
Protocolized treatment. Beck, Rush, Shaw & Emery (1979), Cognitive Therapy of Depression: the manual that made a therapy a specified, testable procedure — structured sessions, named techniques, homework — replicable enough to run in randomized trials.
Single-case design. Hersen & Barlow (1976; 2nd ed. Barlow & Hersen, 1984), Single Case Experimental Designs: baseline, one change, one measure, a predefined decision point, in a single person. Medicine formalized the same logic as the N-of-1 trial — randomized, blinded within-patient comparisons, continued until efficacy is established or disproved (Guyatt, Sackett, Taylor, Chong, Roberts & Pugsley, 1986, NEJM).
Failure-mode planning. Marlatt & Gordon (Eds., 1985), Relapse Prevention: high-risk situations mapped in advance, coping responses analyzed, the lapse distinguished from the relapse — catastrophizing a lapse is what converts one into the other — and return routes rehearsed before they are needed. Lapse pathways treated as analyzable and plannable.
So the kernel holds: therapy's process can be engineered — measured, tested, reversed, maintained. This much is documented.
The limit
The alliance — the working relationship — is among the strongest well-replicated predictors of outcome across modifiable treatment factors, and technique differences measure smaller. The number holds steady across three decades: r = .26 over 24 studies (Horvath & Symonds, 1991); r = .275 over 190 alliance–outcome relations (Horvath, Del Re, Flückiger & Symonds, 2011); r = .278 over 295 studies and more than thirty thousand patients (Flückiger, Del Re, Wampold & Horvath, 2018). Stated at size: r ≈ .28 is about 8% of outcome variance — modest in absolute terms, unusual in its consistency across raters, measures, treatments, and countries — and it is a correlation; causality is argued, with lagged within-patient analyses supporting alliance preceding change. “Among” is the fair word: Wampold (2015) tabulates alliance at d ≈ .57 against treatment differences d ≤ .20, protocol adherence d = .04, and rated competence d = .14 — while empathy and goal consensus measure comparable or larger. Wampold's wider case, that psychotherapy's benefits come mostly from factors shared across therapies (The Great Psychotherapy Debate, 2001; 2nd ed. with Imel, 2015), is his argued conclusion — influential and contested. The oldest form of the equivalence claim carries the “Dodo bird verdict” label: the image goes back to Rosenzweig (1936), who prefaced his common-factors paper with the Dodo's line from Alice in Wonderland — “Everybody has won, and all must have prizes” — and Luborsky, Singer & Luborsky (1975) made it a finding. It is contested: exposure-based treatments outperform for phobias, panic, OCD, and PTSD (Chambless, 2002; Tolin, 2010). No consensus, in either direction.
And the skill that reads a person in the room is tacit — “we can know more than we can tell” (Polanyi, 1966). The counterweight to this note already stands on the site: Phronesis in the Hands, on the 2024 finding that clinical practical wisdom is largely embodied and tacit. Cited here, not re-argued.
The division matches When Thought Becomes Artificial: the formalizable part of a practice can run without the practitioner — therapy's protocols already do, as workbooks and apps — and the tacit remainder is exactly the part that cannot. The spec covers what can be told.
The process takes a spec. The core takes a person.
The corrected claim
The proposal, sized to the evidence: therapy engineers well at the process level — measure each session, test one change at a time, predefine reversal, maintain against relapse — and resists engineering at the core, which is relational and tacit. The honest program is both at once: run the process like an engineer, hold the core like a practitioner. And design the engineering first — choose the measures, the risk ceiling, and the reversibility budget before changing anything. Requirements before build.
The use
For a person's own routine, process-engineering is small and literal:
- Baseline before change — measure the unchanged weeks first.
- One variable at a time.
- A measure chosen in advance, not after.
- A decide-by date.
- A reversal rule — the reading that sends the change back.
- Changes sized so they can be taken back.
This is the site's reversibility principle applied to self-experiments — the single-case logic of Hersen & Barlow and Guyatt, run informally, without blinding or randomization, and named as such. It is method: a way to change a routine and know what happened. Therapy is a licensed profession; this page is a lens on method, not therapy and not medical advice. When the problem is larger than a routine, that calls for a clinician — limits are listed at /dangers.
The instrument is live: /field/n-of-one runs a baseline, one change, one measure, and a decide-by date.
Kin to Phronesis in the Hands — the counterweight, clinical skill as embodied and tacit; When Thought Becomes Artificial — the formalizable part runs without the practitioner; Alter and Confabulation; module 10 of the course — the inverse, therapy as engineering inside the displacement framework (the store); and the live instrument, n-of-one.
Rests on: the feedback trials (Lambert et al., 2001; Shimokawa, Lambert & Smart, 2010; de Jong et al., 2021), the CBT manual (Beck, Rush, Shaw & Emery, 1979), single-case designs (Hersen & Barlow, 1976; Guyatt et al., 1986), relapse prevention (Marlatt & Gordon, Eds., 1985), and the alliance meta-analyses (Horvath & Symonds, 1991; Horvath et al., 2011; Flückiger et al., 2018). The proposal is only the boundary: the process engineers, the core resists.