vivechana

Deterministic discernment scoring for architectural decisions. Same input → same verdict. Score, don't argue.

Vivechana (विवेचना) — Sanskrit: discernment, the act of weighing and discriminating.

At any real scale, deciding "what do we build next?" by argument or gut is just interpretation — and interpretation is hallucination. Vivechana replaces the argument with a number you can reproduce. Give it your options, score five axes against a fixed anchor card, and it returns a ranked verdict that is identical on every machine, every run. No vibes, no recency bias, no loudest-voice-wins.

npx vivechana

No install, no account, no API key, no network. It runs offline.

60-second demo

npx vivechana                              # interactive wizard — walks you through the axes
npx vivechana init my-decision             # write a starter spec you can edit
npx vivechana my-decision.json             # score a spec file
npx vivechana my-decision.json --md out.md # also export a markdown record

Scoring a spec prints a pm2-style result:

  First out-of-box OSS product — universality vs domain depth   (2026-06-29)
  Vivechana V2 · K=I×L×R×Rep · V=K×F · anchors+grades enforced (no hand-waving)

  ┌────┬──────────────────────────────┬────┬────┬────┬─────┬────┬──────┬────────┬─────┐
  │  # │ Option                       │  I │  L │  R │ Rep │  F │    K │      V │ rev │
  ├────┼──────────────────────────────┼────┼────┼────┼─────┼────┼──────┼────────┼─────┤
  │ ▶1 │ Vivechana CLI                │ 6E │ 7E │ 8E │  9M │ 9M │ 3024 │  27216 │  R  │
  │  2 │ ankr-harness                 │ 6E │ 7E │ 7E │  8E │ 8M │ 2352 │  18816 │  R  │
  │  3 │ Bitmask encoding lib         │ 4E │ 8E │ 7E │  9M │ 7E │ 2016 │  14112 │  R  │
  │  4 │ ankr-ctl                     │ 5E │ 6E │ 6E │  7E │ 8M │ 1260 │  10080 │  R  │
  │  5 │ ComplyMitra (domain depth)   │ 4E │ 4E │ 3E │  3E │ 5E │  144 │    720 │  R  │
  │  6 │ Stay fully closed (null)     │ 1M │ 1M │ 1M │  1M │10M │    1 │     10 │  R  │
  └────┴──────────────────────────────┴────┴────┴────┴─────┴────┴──────┴────────┴─────┘

  ┌─ VERDICT ───────────────────────────────────────────────────────────────────┐
  │ Max-V option : Vivechana CLI                                                  │
  │ Score        : V=27216  (K=3024)   tier: flagship  [descriptive — not a gate] │
  │ Beats "Stay fully closed" : YES                                               │
  │ Reversibility: REVERSIBLE                                                     │
  │ Gate (F≥4)   : F=9 → BUILD NOW                                                │
  └───────────────────────────────────────────────────────────────────────────┘

That table isn't a toy — it is the actual scorecard by which this tool was chosen to be the first thing open-sourced out of a 290-service universe. Vivechana scored its own existence. The example ships in examples/oss-first-product.json; run it yourself:

npx vivechana "$(npx vivechana examples)"

The model

K = I × L × R × Rep      ← intrinsic worth (multiplicative: any weak axis drags the whole)
V = K × F                ← deliverable value today

| Axis | | Range | What it counts | |---|---|---|---| | I | Impact | 1–10 | Who/what is directly affected. Facts, not projections. | | L | Leverage | 1–10 | New capabilities unlocked downstream. | | R | Reach | 1–10 | How far the effect propagates. | | Rep | Replicability | 1–10 | Auto-propagation without human effort. 10 = works on case #1000 with no new code. 1 = manual every time. | | F | Feasibility | 1–10 | Buildable right now? The only axis recomputed each time — never stored. |

Because the formula is multiplicative, you can't paper over a fatal weakness with strength elsewhere — a brilliant idea that's manual-forever (Rep=2) or unbuildable (F=1) scores low, on purpose. This is the Jugaad Elimination Property: clever-but-unscalable hacks can't win.

The four guards — why the score is trustworthy

| Guard | Rule | |---|---| | 012 reversibility veto | An IRREVERSIBLE option cannot rank #1 on V alone — irreversible bets must earn it another way. | | 013 anchor card | Every axis value must cite a fixed anchor (9-10 / 5 / 1-2). No bare numbers. The tool refuses an un-anchored axis. | | 014 evidence grade | Every axis is graded Measured / Estimated / Guessed — so a confident-looking score wears its uncertainty on its sleeve. | | 015 null baseline | A scored "do nothing" option is required. If your best real option can't beat doing nothing, "not yet" is the honest verdict. |

Build-now gate is F≥4, not a V threshold

V is priority (which to do first), not a green light. The build-now gate is feasibility (F≥4). A high-V idea that isn't buildable yet is a futures contract, not a task. This keeps the frugal, high-feasibility, modest-V lane alive — which is usually where real shipping happens.

Use it as a library

import { scoreSpec } from 'vivechana';

const res = scoreSpec(spec);   // pure: no fs, no network, no env
// → { ok, ranked, bestReal, beatsNull, shipNow, vTier, ... }

scoreSpec is deterministic and side-effect-free — drop it into CI to gate a roadmap, or have an agent call it to justify a choice.

Provenance

Vivechana is a published method, deposited on Zenodo by its author:

• 3-axis original (prior-art anchor): 10.5281/zenodo.19248146
• 5-axis protocol: 10.5281/zenodo.19456053
• V2 Guards layer (012–015): 10.5281/zenodo.20589218

Author: Capt. Anil Kumar Sharma. This CLI is the reference implementation of the V2 protocol.

License

AGPL-3.0-only. The scoring engine, anchor card, and guards are open. Use it, fork it, build on it — derivatives stay open.