ach-pramana-analysis
Skill0 starsChanged 28 days agoMITView on GitHub
Analysis-of-Competing-Hypotheses discipline (Heuer/CIA method + Vedic Pramana source-tiering + anti-apofenia) for weighing decisions, claims, or priorities without anchoring on the first plausible story.
In this repository12 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- blind-checkSkill
- deep-debug-loopSkill
- evidence-stratifySkill
- fix-on-sightSkill
- kshetraSkill
- other-side-voiceSkill
- promote-patternSkill
- quarantineSkill
- simulate-forkSkill
- source-filterSkill
- verification-firstSkill
The file
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 15
Do NOT invoke for routine, low-stakes, single-obvious-answer tasks — the overhead isn't worth it there.
- Distinctive: line 22
Model missing/hidden data explicitly: If data could be hidden, withheld, or unrepresentative, say so as part of the analysis rather than …
- Distinctive: line 24
State the boundary of the provable: When the evidence runs out, say so plainly instead of extrapolating into a confident narrative.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
ACH + Pramana analysis
A discipline for reasoning under uncertainty, adapted from Richards Heuer's Analysis of Competing Hypotheses (developed for CIA analysts) and paired with a Vedic-derived (Pramana) source-reliability layer. Originally used together on a real investigative reconstruction (public-record analysis of a high-profile case); the same discipline is meant as standard practice in day-to-day work — business decisions, technical trade-offs, disputed claims, not just historical investigation.
When to invoke
- The user explicitly asks to weigh hypotheses, audit a claim, or invokes ACH / Прамана / «разбери по методу».
- A consequential decision (infra choice, business direction, technical judgment) has more than one plausible explanation or path, and picking wrong is costly.
- A disputed or surprising factual claim needs to be resolved from mixed-quality sources.
- Do NOT invoke for routine, low-stakes, single-obvious-answer tasks — the overhead isn't worth it there.
Steps
- Enumerate hypotheses up front, including the simplest/most boring explanation. Do not anchor on the first plausible story and start looking only for evidence that fits it.
- Score evidence by diagnosticity, not volume. For each piece of evidence, ask: does this discriminate between the hypotheses, or is it merely consistent with all of them? Non-diagnostic "supporting" evidence does not strengthen a case no matter how much of it piles up — this is the anti-apofenia check (apophenia = many non-diagnostic coincidences mistaken for a pattern).
- Tier every source explicitly (Pramana-style calibration) instead of treating all evidence as equally trustworthy. Единая шкала — ось цепочки из
<корень миниядра>/docs/pramana-panchaka.md§3: 🟢 первичный документ / прямые данные / воспроизводимый первоисточник · 🟡 достоверное свидетельство / вторичное сообщение / подтверждение специалиста · 🟠 один непроверенный источник или источник со ставкой · 🔴 спекуляция, вывод, нет цепочки. (Прежние Tier 1/2/3 = 🟢/🟡/🟠–🔴; отдельная ось «измеренности» — уevidence-stratify, не смешивать. Карта осей:<корень миниядра>/docs/methods-map.md.) State the tier when it affects the conclusion. - Model missing/hidden data explicitly. If data could be hidden, withheld, or unrepresentative, say so as part of the analysis rather than silently assuming what's visible is complete.
- Select by elimination, not accumulation. The surviving hypothesis is the one with the fewest disconfirming diagnostic contradictions — not the one with the most confirmations. A hypothesis with zero diagnostic evidence against it beats one with many non-diagnostic points for it.
- State the boundary of the provable. When the evidence runs out, say so plainly instead of extrapolating into a confident narrative. "Unresolved" is a valid, honest output.
- Seek independent external validation where possible — if another credible, independent source reached the same boundary, note it; it's a real signal, not proof.
Verification before presenting a conclusion
- Did I list more than one hypothesis, including the boring one?
- Did I actually check each piece of evidence for diagnosticity, or just count things that "fit"?
- Are source tiers stated where they matter to the verdict?
- Did I model what could be missing, rather than assuming completeness?
- Am I stating a real boundary of uncertainty, or dressing up a guess as a conclusion?
When NOT to use
Routine tasks with one clearly correct answer, small reversible decisions, or anything where the overhead of formal hypothesis-weighing would slow the user down for no real benefit. This is for the cases where being wrong is costly and more than one story fits the surface facts.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

